Psoc Res IV Year

Renewable Integrated
Power System Stability

and Control
IEEE Press
445 Hoes Lane
Piscataway, NJ 08854
IEEE Press Editorial Board

Ekram Hossain, Editor in Chief
Jón Atli Benediktsson David Alan Grier Elya B. Joffe

Xiaoou Li Peter Lian Andreas Molisch
Saeid Nahavandi Jeffrey Reed Diomidis Spinellis
Sarah Spurgeon Ahmet Murat Tekalp
Renewable Integrated Power System Stability
and Control
Hêmin Golpîra
University of Kurdistan
Kurdistan, Iran
Arturo Román-Messina
The Center for Research and Advanced Studies of IPN
Guadalajara, Mexico
Hassan Bevrani
University of Kurdistan
Kurdistan, Iran
This edition first published 2021
© 2021 John Wiley & Sons, Inc.
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or
otherwise, except as permitted by law. Advice on how to obtain permission to reuse material from
this title is available at http://www.wiley.com/go/permissions.
The right of Hêmin Golpîra, Arturo Román-Messina, and Hassan Bevrani to be identified as the
authors of this work has been asserted in accordance with law.
Registered Office
John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA
Editorial Office
111 River Street, Hoboken, NJ 07030, USA
For details of our global editorial offices, customer services, and more information about Wiley
products visit us at www.wiley.com.
Wiley also publishes its books in a variety of electronic formats and by print-on-demand. Some
content that appears in standard print versions of this book may not be available in other formats.
Limit of Liability/Disclaimer of Warranty
While the publisher and authors have used their best efforts in preparing this work, they make
no representations or warranties with respect to the accuracy or completeness of the contents
of this work and specifically disclaim all warranties, including without limitation any implied
warranties of merchantability or fitness for a particular purpose. No warranty may be created or
extended by sales representatives, written sales materials or promotional statements for this work.
The fact that an organization, website, or product is referred to in this work as a citation and/or
potential source of further information does not mean that the publisher and authors endorse
the information or services the organization, website, or product may provide or recommendations
it may make. This work is sold with the understanding that the publisher is not engaged in
rendering professional services. The advice and strategies contained herein may not be suitable
for your situation. You should consult with a specialist where appropriate. Further, readers
should be aware that websites listed in this work may have changed or disappeared between when
this work was written and when it is read. Neither the publisher nor authors shall be liable for any
loss of profit or any other commercial damages, including but not limited to special,
incidental, consequential, or other damages.
Library of Congress Cataloging-in-Publication Data

Names: Golpîra, Hêmin, author. | Román-Messina, Arturo, author. | Bevrani,
Hassan, author.
Title: Renewable integrated power system stability and control / Hêmin
Golpîra, Arturo Román-Messina, Hassan Bevrani.
Description: Hoboken, NJ : Wiley-IEEE Press, 2021. | Includes index.
Identifiers: LCCN 2021003857 (print) | LCCN 2021003858 (ebook) | ISBN
9781119689799 (cloth) | ISBN 9781119689829 (adobe pdf) | ISBN
9781119689775 (epub)
Subjects: LCSH: Electric power systems. | Renewable energy sources.
Classification: LCC TK1001 .G65 2021 (print) | LCC TK1001 (ebook) | DDC
621.31/7–dc23
LC record available at https://lccn.loc.gov/2021003857
LC ebook record available at https://lccn.loc.gov/2021003858
Cover Design: Wiley

Cover Image: © kertlis/Getty Images
Set in 9.5/12.5pt STIXTwoText by SPi Global, Pondicherry, India
10 9 8 7 6 5 4 3 2 1
To our mothers, Himêra, Diana and Ayesha
vii
Contents
Preface xiii
Acknowledgments xvii
Nomenclature xviii
List of Abbreviations and Acronyms xxi
1 Introduction 1
1.1 Power System Stability and Control 1
1.2 Current State of Power System Stability and Control 4
1.2.1 Frequency Control 5
1.2.2 Voltage Control 6
1.2.3 Oscillation Damping 7
1.3 Data-Driven Wide-Area Power System Monitoring and Control 8
1.4 Dynamics Modeling and Parameters Estimation 10
1.4.1 Modeling of Frequency, Voltage, and Angle Controls 11
1.4.2 Parameters Estimation 12
1.5 Summary 14
References 14
2 MG Penetrated Power Grid Modeling 25

2.1 Introduction 25
2.2 Basic Concepts 26
2.2.1 Dynamic Equivalencing 26
2.2.2 Background on Study Zone and External System 27
2.3 Power Grid Modeling 28
2.3.1 The Notion of Center-of Gravity (COG) 28
2.3.1.1 Key Concept 28
2.3.1.2 Basic Assumptions 32
2.3.1.3 Modeling Formulation 32
2.3.1.4 Local Frequency Estimation 33
viii Contents
2.3.1.5 Simulation Results 35

2.3.2 An Enhanced COG-Based Model 46
2.3.2.1 Key Concept 46
2.3.3 Generalized Equivalent Model 50
2.3.3.1 Basic Logic 50
2.3.3.2 Simulation and Results 51
2.4 MG Equivalent Model 53
2.4.1 Islanded Mode 54
2.4.1.1 Synchronous-Based DG 54
2.4.1.2 Genset Model Validation 57
2.4.1.3 Inverter-Based DG 58
2.4.1.4 Inverter-Based DG Model Validation 61
2.4.2 Grid-Connected Mode 61
2.4.2.1 Basic Logic 61
2.4.2.2 Model Validation 63
2.5 Summary 67
References 67
3 Stability Assessment of Power Grids with High Microgrid

Penetration 71
3.1 Introduction 71
3.1.1 Motivation 71
3.1.2 Relations with Previous Literature 72
3.2 Frequency Stability Assessment 73
3.2.1 Background on Frequency Indices 73
3.2.1.1 Rate of Change of Frequency 73
3.2.1.2 Frequency Nadir 74
3.2.1.3 Delta Frequency Detection 74
3.2.2 Frequency Stability Assessment Under High MG Penetration Levels 74
3.2.3 Sensitivity Factors 74
3.2.3.1 Frequency Response 74
3.2.3.2 Delta Frequency Detection 77
3.2.4 Simulation and Results 78
3.3 Maximum Penetration Level: Frequency Stability 80
3.3.1 Basic Principle 80
3.3.2 Background on MG Modeling 81
3.3.3 Minimum Inertia Related to Frequency Nadir 82
Contents ix
3.3.4 Minimum Inertia Related to Delta Frequency Detection 84

3.3.5 Minimum Inertia Related to RoCoF 85
3.3.6 Maximum Penetration Level 86
3.3.7.1 Analysis Tools 86
3.3.7.2 Dynamical Simulation Results 86
3.4 Small-Signal Stability Assessment 90
3.4.1 Basic Definition 90
3.4.2 Key Concept 90
3.5 Maximum Penetration Level: Small-Signal Stability 93
3.5.1 Basic Idea 93
3.6 Voltage-Based Realization of the MG-Integrated Power Grid 94
3.6.1 Key Concepts 95
3.6.2 Jacobian Sensitivities 95
3.6.2.1 V-P Sensitivity 95
3.6.2.2 V-Q Sensitivity 96
3.7 Summary 99
References 100
4 Advanced Virtual Inertia Control and Optimal Placement 103

4.1 Introduction 103
4.2 Virtual Synchronous Generator 104
4.2.1 Concept and Structure 105
4.2.2 Basic Control Scheme and Applications 106
4.2.3 Application in Power System Dynamic Enhancement 108
4.2.3.1 Scenario 1: 10-MW Load Increase at Bus 9 109
4.2.3.2 Scenario 2: 20-MW Power Command Decrease of G3 109
4.2.4 Application to Power Grids with HVDC Systems 110
4.3 Dispatchable Inertia Placement 113
4.3.1 Frequency Dynamics Enhancement 113
4.3.1.1 Background: Literature Review 113
4.3.1.2 Virtual Inertia Modeling 114
4.3.1.3 Experimental Verification 116
4.3.1.4 Economic Modeling 119
4.3.1.5 Simulation and Results 124
x Contents
4.3.1.6 Sensitivity Analysis 134

4.3.2 Small-Signal Stability 136
4.3.2.1 Objective Function 136
4.4 Summary 139
References 139
5 Wide-Area Voltage Monitoring in High-Renewable Integrated Power

Systems 145
5.2 Voltage Control Areas: A Background 147
5.2.1 Voltage Sensitivities 148
5.2.2 Electrical Distances 149
5.2.3 Reactive Control Zones and Pilot Nodes 150
5.2.3.1 Selection of Optimal Pilot Buses 151
5.2.3.2 Selection of Control Plants 151
5.2.4 Other Approaches 152
5.3 Data-driven Approaches 153
5.3.1 Wide-Area Voltage and Reactive Power Regulation 154
5.3.2 PMU-Based Voltage Monitoring 155
5.4 Theoretical Framework 155
5.4.1 Dynamic Trajectories 156
5.4.2 Spectral Graph Theory 157
5.4.3 Kernel Methods 157
5.4.3.1 Markov Matrices 159
5.4.3.2 The Markov Clustering Algorithm 162
5.4.4 Spatiotemporal Clustering 164
5.5 Case Study 165
5.5.1 Sensitivity Studies 165
5.5.2 Data-Driven Analysis 169
5.5.3 Measurement-Based Reactive Control Areas 171
5.5.3.1 Diffusion Maps 171
5.5.4 Direct Clustering 175
5.5.5 Correlation Analysis 176
5.5.5.1 Direct Analysis of Concatenated Data 178
5.5.5.2 Two-Way Correlation Analysis 179
5.5.5.3 Partial Least Squares Correlation 179
5.6 Summary 181
References 181
Contents xi
6 Advanced Control Synthesis 185

6.2 Frequency Dynamics Enhancement 186
6.2.1 Background: The Concept of Flexible Inertia 186
6.2.2 Frequency Dynamics Propagation 189
6.2.3 Inertia-Based Control Scheme 191
6.2.4 Flexible Inertia: Practical Considerations 192
6.2.5 Results and Discussions 194
6.3 Small Signal Stability Enhancement 200
6.3.1 Key Concept 200
6.3.2 Control Scheme Design 201
6.4 Summary 207
References 207
7 Small-Signal and Transient Stability Assessment Using Data-Driven

Approaches 211
7.1 Background and Motivation 212
7.2 Modal Characterization Using Data-Driven Approaches 213
7.2.1 Modal Decomposition 213
7.2.2 Multisignal Prony Analysis 215
7.2.2.1 Standard Prony Analysis 215
7.2.2.2 Modified Least-Squares Algorithm 218
7.2.2.3 Multichannel Prony Analysis 219
7.2.2.4 Hankel-SVD Methods 221
7.2.3 Koopman and Dynamic Mode Decomposition Representations 222
7.2.3.1 The Koopman Operator 223
7.2.4 Dynamic Mode Decomposition 223
7.2.4.1 SVD-Based Methods 225
7.2.4.2 The Companion Matrix Approach 228
7.2.4.3 Energy Criteria 230
7.3 Studies of a Small-Scale Power System Model 231
7.3.1 System Data and Operating Scenarios 231
7.3.2 Exploratory Small-Signal Analysis 234
7.3.3 Large System Performance 236
7.3.3.1 Cases B–C 236
7.3.3.2 Case D 238
7.3.4 Mode Shape Identification 241
7.3.5 Temporal Clustering 242
xii Contents
7.4 Large-Scale System Study 244

7.4.1 Case Study Description 244
7.4.2 Renewable Generator Modeling 245
7.4.3 Effect of Inverter-Based DGs on Oscillatory Stability 245
7.4.4 Large System Performance 246
7.4.5 Model Validation 246
7.4.5.1 Reconstructed Flow Fields 250
7.4.6 Identification of Mode Shapes Using DMD 253
7.5 Analysis Results and Discussion 253
References 255
8 Solar and Wind Integration Case Studies 259

8.1 General Context and Motivation 259
8.2 Study System 261
8.3 Wind Power Integration in the South Systems 263
8.3.1 Study Region 263
8.3.2 Existing System Limitations 266
8.4 Impact of Increased Wind Penetration on the System Performance 266
8.4.1 Study Considerations and Scenario Development 266
8.4.2 Base Case Assessment 267
8.4.2.1 System Oscillatory Response 269
8.4.3 High Wind Penetration Case 271
8.5 Frequency Response 274
8.5.1 Frequency Variations 274
8.5.2 Wind and Hydropower Coordination 277
8.5.3 Response to Loss-of-Generation Events 280
8.6 Effect of Voltage Control on System Dynamic Performance 283
8.6.1 Voltage Support and Reactive Power Dispatch 283
8.6.2 Effect of Voltage Control Characteristics 283
8.7 Summary 288
References 288
Index 293
xiii
Preface
Increased needs for electrical energy as well as environmental concerns besides

growing attempts to reduce dependency on fossil fuel resources have caused power
system industries all around the world to set an ambitious target of renewable
generation. Therefore, the capacity of installed inverter-based distributed genera-
tors (DGs) and renewable energy sources (RESs), individually or through the
microgrids (MGs), in power systems is rapidly growing; and this increases the
significance of renewable integrated power system stability and control as a
challenging issue.
It is well known that low penetration of MGs/DGs has little influence on host
grid stability and dynamics and thus the associated dynamics could be studied
through simple power flow analysis. However, modern power grids face new
technical challenges arising from the increasing penetration of power-electronic-
interfaced MGs/DGs. Increasing renewable power penetration level may adversely
affect frequency response, voltage and system control and lead to degraded perfor-
mance of traditional control schemes. This, in turn, may result in large deviations
and, potentially, system instability.
Moreover, the increasing penetration of inverter-interfaced DGs motivates the
need to develop additional ancillary services to control undesired system dynam-
ics. Advanced control of grid-connected MGs, however, has the potential to offset
the intermittent nature of distributed energy resources and provide control sup-
port to the host utility during emergency conditions. To this end, new trends in
power system modeling and dynamic equivalencing should be discussed.
In response to the above mentioned challenges, this book provides a comprehen-
sive coverage of penetrated power grid dynamic analysis, data-driven estimation,
modeling, and control synthesis. This book summarizes a long-term academic/
industry research outcome and contributions and is influenced by the authors’
practical experiences on power system dynamics and stability, MGs, electric net-
work monitoring, and control and power electronic systems in several countries,
xiv Preface
universities, and power electric companies. The book also provides a thorough
understanding of the basic principles of the penetrated power system modeling,
model order reduction, and grid-connected MG equivalent model derivation.
One of the main parts of this book covers the modeling of power systems using
PMU data for the purpose of renewable integrated dynamics identification and
parameters estimation issues, as well as oscillation damping, voltage control,
and frequency control design problems. For this purpose, in addition to real net-
work data, several standard IEEE power system models are used as benchmarks
for generating data that are used in system identification. Furthermore, in addition
to dynamic stability analysis and controller synthesis, inertia challenge require-
ments and control levels are discussed, and recent advances in visualization of vir-
tual synchronous generators (VSGs) and the associated effects on system
performance are addressed.
This book could be useful for engineers and operators working on power systems
dynamic, control, and operation, as well as postgraduate students and academic
researchers. The book describes renewable integrated power system dynamics
modeling and control issues from introductory to the advanced steps. This book
is organized into eight chapters.
Chapter 1 discusses the term of power system stability and control with an
updated brief review on the areas of frequency, voltage, and angle controls, con-
cerning the penetration of RESs/DGs. In response to the existing challenges in
penetration of more RESs/DGs to the grid, the necessity of using data-driven mod-
eling, parameters estimation, and control synthesis in wide-area power systems is
emphasized; a general scheme for wide-area measurement system and wide-area
control is described.
Chapter 2 deals with dynamic equivalencing of penetrated power grid. Several
methods are introduced to model the host grid as well as the distribution network.
A center of gravity (COG)-based equivalent model is addressed to represent the
power system dynamic behavior in terms of slow power and frequency dynamics.
The relationship between the frequency of the COG and the motion of local cen-
ters of angle is analytically determined to compute local frequency deviations fol-
lowing major disturbances.
Chapter 3 addresses the power grid stability analysis from frequency, small sig-
nal and voltage points of view. Some analytical approaches have been discussed to
determine maximum penetration level of MGs concerning the upward system sta-
bility. The given methods explicitly rely on the basic power system equations
which, in turn, make the proposed indices completely independent of the test case;
and this helps to solve the associated difficulties with the system dimensions.
Chapter 4 explains the VSG concept and its applications in renewable integrated
power grids. The positive dynamic impacts of VSGs in a power system are dis-
cussed, and recent relevant achievements in the application of the advanced
Preface xv
control methods in emulating virtual inertia are clarified. Afterwards, according to

relevant dynamical metrics, dispatchable inertia is optimally placed in the system
to enhance system stability and dynamics performance.
Chapter 5 examines the application of a measurement-based analysis technique
to identify voltage control areas in renewable integrated power systems. The pro-
posed technique combines the inherent abilities of graph theoretical techniques
with spectral clustering and visualization methods to identify voltage control areas
and reconstruct system behavior. The evaluation of voltage stability problems is
done using both static and dynamic techniques. Numerical relevant issues are also
discussed.
Chapter 6 proposes advanced control schemes rely on inertia manipulation in
the system to improve frequency, voltage, and small signal stabilities. The pro-
posed control approaches, which use stochastic equivalent model of power system
(to enable high penetration levels of MGs), combine an adaptive dispatch strategy
for energy storage systems with an MG-controlled islanding scheme to provide sta-
bility support for the host power grid.
Chapter 7 addresses some important issues in understanding the oscillatory per-
formance of wind and solar PV penetrated power systems. A fundamental study of
the characterization of power system dynamic behavior with increased RESs is
presented. The study is motivated by the need to further clarify the participation
of wind and PV farms in inter-area oscillations. The performance of data-driven
model extraction techniques using two simple and complex power system exam-
ples is evaluated. Some analytical criteria to describe the energy relationships in
the observed oscillations are derived, and a physical interpretation for the system
modes is suggested.
Chapter 8 describes an experience in the analysis of wind and solar integration in
a large-scale practical power system to examine the impact of high variable renew-
able generation on power system security. Both active and reactive power control
strategies are considered. The study assesses the impact of large amount of wind
and PV on system dynamics and identifies ways of improving system dynamic
performance and stability through control and operating practices.
Hêmin Golpîra, University of Kurdistan

Arturo Román-Messina, The Center for Research
and Advanced Studies of IPN
Hassan Bevrani, University of Kurdistan
January 2021
xvii
Acknowledgments
Most of the contributions, outcomes, and insight presented in this book were
achieved through a long-term teaching and research cooperation on the renewable
integrated power systems over the last 15 years. The materials given in the present
book are mainly the research outcomes and original results of authors in Smart/
Micro Grids Research Center-SMGRC, University of Kurdistan (Sanandaj, Iran)
and in The Center for Research and Advanced Studies-CINVESTAV of the
National Polytechnic Institute of Mexico (Guadalajara, Mexico). It is a pleasure
to acknowledge the received supports from these sources, and the awards from
Iran Grid Management Company (IGMC), Iran National Science Foundation
(INSF), and Alexander von Humboldt (AvH) Foundation.
The authors would like to thank their colleagues Prof. J. Raisch and Prof. Bruno
Francois for their kind support. Finally, the authors offer their deepest personal
gratitude to their families for their patience during the preparation of this book.
xviii
Nomenclature
X,Y data matrices

ε Kernel bandwidth
λ eigenvalue
σ singular value
ξ damping
υ vector of natural modes
ψ,φ eigenvectors
ΨΦ matrices of eigenvectors
Γ inflation operator
θ phase angle
Λ diagonal matrix of eigenvalues
Σ diagonal matrix of singular values
δs mechanical rotor angle (rad)
δ rotor angle position (rad)
δCOI i rotor angle position of the COI (rad)
ω angular speed (rad/s)
ωs mechanical rotor angular speed (rad/s)
ω0 rated angular speed (rad/s)
Tm(t) mechanical input torque (p.u.)
Te(t) electrical output torque (p.u.)
M inertia constant of the system (s)
M ESS
i
inertia of ESS in area i (s)
ESS minimum required ESS inertia, in compliance with RoCof, in area i (s)
M i, min
ESS minimum required ESS inertia, in compliance with frequency nadir,
M i, min
in area i (s)
D damping coefficient
I(t) impulse response of the system
P(n) data sequence of interest
Nomenclature xix
Pin injected power of ESS to the host grid

K number of sinusoidal components in noise
L length of P(n)
Lx,Ly latent variables
J moment of inertia
ak magnitude
Φk initial phase angle
ωk harmonic frequency in radius
Ak complex magnitude of the kth-harmonic
si eigenvectors associated with the noise subspace
e signal eigenvector
eU complex-conjugate transpose of e
Ccap capital costs ($/kW)
CPCS power conversion system costs ($/kW)
Cstor storage section costs ($/kWh)
CBOP power balance costs ($/kW)
tch charging/discharging time (h)
CO & M operation and maintenance costs ($/kW-year)
CR,a annualized replacement costs ($/kW-year)
Ccap,a annualized total capital costs ($/kW-year)
CLCC,a annualized life cycle costs ($/kW-year)
CRF capital recovery factor
CR replacement costs ($/kWh)
CFOM,a fixed operation and maintenance costs ($/kW-year)
CVOM,a variable operation and maintenance costs ($/kWh)
ncycle number of discharge cycles per year
ζc charging efficiency of the battery (%)
ζd discharging efficiency of the battery (%)
η power angle-based stability index
i (j) area (bus) index
f frequency (Hz)
Ptie
•
virtual transferred power (pu)
X tie
•
fictitious reactance (pu)
TtieCOIi,j applied torques from bus j to COI
TtieCOIi,COG applied torques from COG to COI
Ai area i
ΔPi size of disturbance in area i
ξ− deviations from the target value in negative direction
xx Nomenclature
ξ+ deviations from the target value in positive direction

ξ target value
ps probability of each scenario
s scenario counter
IC internal combustion
SM synchronous machine
MP slope of P-ω droop
KPI integral control gain
KPP proportional control gain
FCMD command fuel signal
ECMD exciter control signal
Pmeas measured value of real power
Qmeas measured value of reactive power
I• line current
Ktf torque to fuel conversion ratio
ηthr thermal constant
Kcv calorific value
Kfr fuel rate at rated speed
Km mechanic losses constant
τe exciter machine time constant
PMG injected power of MG to the host grid
ωMGs angular speed at the point of common coupling
ζ DGs re-dispatching time (s)
υ DGs islanding time (s)
n number of areas
β frequency bias
PGenset generation of Genst
L level arm length (m)
M Conv
COI i
conventional synchronous inertia (s)
TD delivery time of primary frequency response (s)
Kf(s) transfer function of the phase-locked loop
MMMG muti-micro-grid inerta constant (s)
V0 initial values of terminal voltage
YH Hankel matrix
. norm
xxi
List of Abbreviations and Acronyms
A amplitude
AGC automatic generation control
AQR automatic reactive power regulator
AVR automatic voltage regulator
B residue
BFV best fitness value
COG center of gravity
COI center of inertia
DER distributed energy resources
DFIG doubly-fed induction generator
DG distributed generation
DM diffusion map
DMD dynamic mode decomposition
E energy
EMT electromagnetic transient
ESS energy storage system
EV electrical vehicle
GA genetic algorithm
HVDC high-voltage direct current
KT Kumaresan–Tuft
LCC line commutated converter
MCL Markov clustering
MG microgrid
MMGs multi-MGs
NERC North American Electric Reliability Corporation
NYNE New York New England
PC principal component
PCTVAR percentage of variation
PF participation factor
xxii List of Abbreviations and Acronyms
PI proportional-integral
PLL phase-locked loop
PLS partial least squares regression
PLSC partial least squares correlation
PMSG permanent magnet synchronous generators
PMU phasor measurement unit
POIS point of interconnection with the system
PS pseudo spectrum
PSS power system stabilizers
PV photovoltaic
RES renewable energy source
RoCoF rate of change of frequency
SC synchronous condenser
SCADA supervisory control and data acquisition
SG synchronous generator
SLB static load bank
SOC state of charge
SS static switch
SVC static VAR compensator
SVD singular value decomposition
T-D time domain
TSO transmission system operator
UCTE Union for the Coordination of the Transmission of Electricity
UFLS underfrequency load shedding
ULTC under load tap changer
VAR volt–ampere reactive
VSC voltage source converter
VSG virtual synchronous generator
V2G vehicle-to-grid
Vvand Vandermonde matrix
WAMS wide-area measurement system
WF wind farm
WT wind turbine
z complex amplitude
1
Introduction
The term power system stability and control is used to define the application of con-
trol theorems and relevant technologies to analyze and enhance the power system
functions during normal and abnormal operations. Power system stability and
control refers to keep desired performance and stabilizing power system following
various disturbances, such as short circuits, loss of generation, and load.
The capacity of installed inverter-based distributed generators (DGs) and renewa-
ble energy sources (RESs) individually or through the microgrids (MGs) in power sys-
tems is rapidly growing, and a high penetration level is targeted for the next few
decades. In most countries including developing countries, significant targets are con-
sidered for using the distributed microsources and MGs in their power systems for
near future. The increase of DGs/RESs in power systems has a significant impact
on CO2 reduction; however, recent studies have shown that relatively high DGs/RESs
integration will have some negative impacts on power system dynamics, frequency
and voltage regulation, as well as other control and operational issues. Decreasing
system inertia and highly variable dynamic nature of DGs/RESs/MGs are known
as the main reasons. These impacts may increase for the dynamically weak power
systems at the penetration rates that are expected over the next several years.
In this chapter, a brief discussion on the power system stability and control in
modern renewable integrated power systems and the current state of this topic are
given. Data-driven wide-area power system monitoring and control is emphasized,
and the significance of measurement-based dynamic modeling and parameter esti-
mation is shown.
1.1 Power System Stability and Control
Power system stability and control was first recognized as an important problem in
1920s [1]. Over the years, numerous modeling/simulation programs, synthesis/
Renewable Integrated Power System Stability and Control, First Edition.
Hêmin Golpîra, Arturo Román-Messina, and Hassan Bevrani.
© 2021 John Wiley & Sons, Inc. Published 2021 by John Wiley & Sons, Inc.
2 1 Introduction
analysis methodologies, and protection schemes have been developed. Power grid
control must provide the ability of an electric power to regain a state of operating
equilibrium after being subjected to a physical disturbance, with most system vari-
ables, i.e., frequency, voltage, and angle, bounded so that practically the entire sys-
tem remains intact. Thus, the main control loops are known as frequency control,
voltage control, and rotor angle (power oscillation damping) control [2].
In many power systems, advanced measurement devices such as phasor meas-
urement units (PMUs) and modern communication devices are already being
installed. Using these facilities, the parameters of existing power system control-
lers can be adjusted by an online data-driven control mechanism [3]. The PMU
data after filtering are used to estimate some important parameters in the system
(scheduling parameters). These parameters are then used in the control tuning
algorithm that will adapt the controller parameters in frequency control, voltage
control, and power oscillation control. Therefore, the controller’s parameters are
adapted according to the current status of the system.
One of the important steps of reliable and performant control system design
is defining the performance specifications. It depends on the features of the con-
troller design method, the constraints on the controller structure, the achievable
performance that is limited by the physical constraints, the industrial standards on
the limit of the variables, the limits of the actuators, etc. Finding the control spe-
cifications and making them compatible with the controller design approach
require a deeper understanding of the physical system to be controlled.
The characteristics of three main control loops, i.e., frequency control, voltage
control, and angle control, should be studied to enable the definition of achievable
performance specifications and designing an effective control system.
• Frequency control: Since the frequency generated in an electric network is

proportional to the rotation speed of the generator, the problem of frequency
control may be directly translated into a speed control problem of the turbine
generator unit. This is initially overcome by adding a governing mechanism that
senses the machine speed and adjusts the input valve to change the mechanical
power output to track the load change and to restore frequency to nominal
value. Depending on the frequency deviation range, different frequency control
loops, i.e., primary, secondary, and tertiary, may be required to maintain power
system frequency stability [4].
The secondary frequency control which is also known as load frequency con-
trol (LFC) initializes a centralized and automatic control task using the assigned
spinning reserve. The LFC is the main component of an automatic generation
control (AGC) system [5]. In large power systems, this control loop is activated
in the time frame of few seconds to minutes after a disturbance. In a modern
AGC system, based on the received area control error (ACE) signal, an online
1.1 Power System Stability and Control 3
tuning algorithm must adjust the LFC parameters to restore the frequency and
tie-line powers to the specified values.
• Voltage control: The generators are usually operated at a constant voltage by

using an automatic voltage regulator (AVR) which controls the excitation of
the machine via the electric field exciter system. The exciter system supplies
the field winding of the synchronous machine with direct current to generate
required flux in the rotor. A system enters a state of voltage instability when
a disturbance changes the system condition to make a progressive fall or rise
of voltages of some buses. Loss of load in an area, tripping transmission lines,
and other protected equipment are possible results of voltage instability. Like
frequency control, the voltage control is also characterized via several control
loops in different system levels. The AVR loop which regulated the voltage of
generator terminals is located on lower system levels and responds typically
in a time scale of a second or less.
• Angle control: Rotor angle stability is the ability of the power system to maintain
synchronization after being subjected to a disturbance. Angle stability refers to
damping of power oscillations inside subsystems and between subsystems on an
interconnected grid during variation beyond specified threshold levels. The risk
of losing angle stability can be significantly reduced by using proper control
devices inserted into the power grid to find a smooth shape for the system
dynamic response.
The power oscillation damping has been mainly guaranteed by power system
stabilizers (PSSs). A PSS is a controller, which, beside the turbine-governing sys-
tem, performs an additional supplementary control loop to the AVR system of a
generating unit. Depending on the type of PSS, the input signal could be the
rotor speed/frequency deviation, the generator active power deviation, or a com-
bination feedback of rotor speed/frequency and active power changes. This sig-
nal to be passed through a combination of a lead-lag compensators. The PSS
output signal is amplified to provide an effective output signal.
In order to damp the inter-area oscillations, which have smaller oscillation
frequency than the local oscillatory modes, a wide-area control (WAC) system
is required. The WAC system is a centralized controller that uses the PMU sig-
nals and produces auxiliary control signals for the PSSs.
• Virtual synchronous generator: Additional flexibility may be required from var-

ious control levels so that the system operator can continue to balance supply
and demand on the modern power grids in the presence of DGs/RESs/MGs.
The contribution of DGs/RESs in regulation task refers to the ability of these
grids to regulate their power output, by an appropriate control action. This
can be regarded as adding virtual inertia to the grid and considered as a solution.
Virtual inertia emulation requires the inverter to be able to store or release an
amount of energy depending on the grid frequency’s deviation from its nominal
4 1 Introduction
value, analogous to the inertia of a conventional generator. This setup, which is

known as virtual synchronous generator (VSG), will then operate to emulate
desirable dynamics, such as inertia and damping properties, by flexible shaping
of its output active and reactive powers as conceptually shown in Figure 1.1.
This VSG provides a promising solution to improve power grid stability and per-
formance in the presence of a high penetration of DGs/RESs/MGs. The VSG is not
only applicable for improving of frequency regulation and oscillations damping,
particularly during the transient state following a disturbance, but also it is useful
to support the voltage stability. The VSG system can use the available DGs/RESs,
as primary sources to participate in power oscillation damping by adjusting their
active and reactive power generations. The VSG is more discussed in Chapter 4.
1.2 Current State of Power System Stability and Control
Power system stability and control can take different forms, which are influenced
by the type of instability phenomena. A survey on the basics of power system con-
trols, literature, and achievements is given in [6, 7].
PMUs are sophisticated digital recording devices that communicate global
positioning system (GPS) synchronized high sampling rate dynamic power sys-
tem’s data to the central control and monitoring stations. The recorded data by
PMUs provide valuable information about the dynamic of the power system that
can be used for data-driven modeling. An overview of system identification tech-
niques for modeling of power systems using PMU data is given in [8]. In [9], a sub-
space identification method is used to identify a reduced order model for power
oscillation control. The PMU data are used for the calibration of the parameters
of the reduced-order model of a power generator in [10]. The feasibility of
multi-input multi-output (MIMO) identification of power systems using low-level
Power Grid
Primary DC/DC
Source Pin DC/AC Pout
Inverter
AC/DC
VSG Control
Figure 1.1 Conceptual structure of a virtual synchronous generator.

probing signal is shown in [11]. An online algorithm is used in [12] to identify the
frequency response of power system dynamics, while it is combined with a
selective modal analysis. The transfer function and state-space model identifica-
tions using PMU data are compared in [13] for electromechanical oscillation
damping estimation. Several identification methods are compared for analysis
of inter-area oscillatory modes of power systems [14].
The data from PMUs have already been used for estimation of some important
power system parameters. The electromechanical modes of a power system and
their confidence intervals are estimated using PMUs operational data in [15,
16]. Amplitude, frequency, and damping of power system oscillations are esti-
mated using PMU measurements in [17, 18]. The PMU data are used in [19–21]
to identify the topology (or change in topology) of a power system. Recently, some
system identification methods have been employed to estimate the power system
inertia using the operational PMU data (with no external excitation signal)
[22, 23].
1.2.1 Frequency Control

Preliminary efforts in the field of power grid frequency regulation are reported in
[24]. Subsequently, an IEEE working group prepared some standard definitions
of significant terms and concepts on power system frequency control [25]. Consid-
ering the physical constraints and to cope with the advances in technologies
and the changed system environment, dynamic modeling developments, security
constraints, and communication delays, as well as modifications on the frequency
control definitions, have been discussed over the years [26–30]. A comprehensive
survey and exhaustive bibliography on frequency control up to 2014 are given
in [31, 32].
Frequency control analysis, frequency response modeling, nonlinearity and
uncertainty presentation, specific applications, frequency bias calculation, control
performance standards, load characteristics impacts, and parameters identifica-
tion are presented in several documents [29, 32–42]. A Considerable research
on the time-delayed system is contained in [4, 30]. In addition, regarding paramet-
ric uncertainty, several self-tuning, adaptive, and robust control strategies are
widely applied for power grid LFC system synthesis over the years [4, 43–51].
Dynamic impacts of intermittent DGs and high penetration of RESs on power
grids frequency response are discussed in [32, 52–56]. A low inertia can negatively
affect the grid frequency dynamic performance and stability. A number of recent
works have suggested the application of inverter-based virtual inertia emulators to
improve frequency stability and frequency response performance [57–61]. Fur-
thermore, numerous research works have been recently focused on the use of
DGs, RESs, MGs, electric vehicles, and storage devices to provide frequency
6 1 Introduction
control supports in the power grids [62–69]. Providing frequency control support
via controllable loads and smart load technologies using the concepts of demand
response (DR) is discussed in [41, 42, 70–76]. Two recent works in this area are [77,
78], that discuss the impact of a high integration of MGs on the frequency control
of power systems, and propose a decentralized stochastic frequency control
of MGs.
PMU-based/data-driven online tuning frequency control approach is not
addressed in the abovementioned worldwide published works. In most cases,
the secondary frequency control is designed using conventional frequency
response model, which is very difficult to realize in a modern power grid with a
highly variable structure and penetration of DGs/RESs.
1.2.2 Voltage Control

Since 1990s, supplementary control of generator excitation systems, static var com-
pensator (SVC), and high voltage direct current (HVDC) converters is increasingly
being used to solve power system oscillation problems [7]. There has also been a
general interest in the application of power electronics-based controllers known as
flexible alternating current transmission system (FACTS) controllers for the
damping of system oscillations [79]. Following several power system collapses
worldwide [80–82], in 1990s, voltage stability has attracted more research
interests.
Recently, following the development of PMUs, communication channels, and
digital processing, wide-area power system stabilization and control have become
areas of interest [83, 84]. A typical generic of different voltage control levels is dis-
cussed in [85]. Optimal voltage control has long been successfully implemented in
power systems, including the three-level hierarchical automatic voltage control in
Europe [86–88], and the adaptive zone division method in China [89].
A supervisory voltage control strategy for large-scale solar photovoltaic (PV)
integration in power network is proposed in [90, 91] to enhance the voltage sta-
bility. A survey of methods, mostly based on PMU data, for long-term voltage
instability detection is given in [92]. In [93], a two-stage distributed voltage control
scheme is proposed. The first stage is the local control of each DG based on sen-
sitivity analysis, and the second stage acquires reactive power support from other
DG units. In [94], a consensus-based cooperative control is proposed to regulate
voltage by coordinating electric cars and active power curtailment of PVs. In
[95], a distributed voltage stability assessment considering DG units is developed
based on distributed continuation power flow. Coordinated voltage control is a
technique which provides voltage control by means of adjusting, sequencing,
and timing various kinds of controllers within a system. Some relevant works
are reported in [96–98].
As mentioned above, several PMU-based voltage control methodologies have

been reported worldwide; however, mostly presented a voltage recovery approach
in an off-normal or emergency condition. Among existing three hierarchical levels
of voltage control (primary, secondary, and tertiary controls), only few works are
mainly focused on optimal supervisory on secondary voltage control, which is
required to coordinate adjustment of the set-points of the existing voltage control-
lers. In this regard, the online adaptive tuning of available voltage control systems
in a power grid with high integration of DGs/RESs is not well addressed. Further-
more, the overlap between voltage dynamics and frequency/active power as well
as rotor angle dynamics in a modern power grid has not been highlighted in the
published reports.
1.2.3 Oscillation Damping

Traditionally, the power system oscillations are damped through the generator
local controllers, such as the exciter and governor, which are designed to ensure
only the local stability of the generator (1–2 Hz). In order to increase the stability of
the system, PSSs and power electronic converter-based FACTS are added into the
grid [99–101]. In a broader context, the power system oscillation problem has also
been related to voltage stability. The control interaction is discussed in [102, 103].
The exploitation of the wide-area measurements, provided by PMUs, for monitor-
ing and controlling the power system led to the introduction of the wide-area mon-
itoring and control (WAMC) systems [104]. The advent and application of
synchronized measurement technology has enabled the detection and observation
of poorly damped oscillations (such as the inter-area modes) and became the back-
bone for more development of the WAMC systems [105]. Inter-area oscillations
are characterized by low frequency (0.2–1 Hz) and occur when generators of
one group swing against generators of another group [106]. Integration of RESs
into the WAC scheme for damping power oscillations is discussed in [107, 108].
The utilization of a networked control system model for the WAC design, accord-
ing to linear matrix inequality techniques, is proposed in [111]. Furthermore, Ref.
[110] presents a WAC design, based on particle swarm optimization, for improving
the performance of the power system through the control of wind farms.
More specifically, WAC aims to utilize the synchronized phasor measurements
in order to provide coordination signals to the local controllers, making them capa-
ble of damping effectively all the inter-area oscillations [100]. In the literature, var-
ious works deal with the development of a WAC system. The proposed WAC
schemes are segregated mainly according to the components of the power system
that the WAC is intended to coordinate [3, 110]. Multiple control methodologies
have been developed for damping the inter-area oscillations deploying a WAMC.
In [83], a decentralized/hierarchical architecture for wide-area damping control
8 1 Introduction
using PMU remote feedback signals was discussed. References [100, 111] proposed
the design of wide-area damping controllers that provide supplementary damping
control to synchronous generators (SGs). A networked control system model for
wide-area closed-loop power systems is applied in [109]. A power oscillation
damping controller is introduced in [112] based on a modal linear quadratic Gaus-
sian methodology. A combination of controlling SGs and renewable sources in
order to increase the overall damping capability of the system is shown in [101,
107, 108, 113]. Few LPV control solutions to power oscillation damping are pro-
posed that use either a low-order first principle model of the system [114] or a
reduced-order parametric LPV identified model [115].
In comparison of frequency and voltage control, a higher number of reports have
been published in PMU-based oscillation damping (rotor angle control) field.
However, most of the reported approaches require the detailed and accurate
knowledge of the complete network model (both topology and parameter values),
that is unavailable or corrupted in practice as a result of communication failures,
bad data in state estimation etc. In addition, the impact of disturbances on the
inter-area oscillations cannot be well captured by these methods.
1.3 Data-Driven Wide-Area Power System Monitoring

and Control
Power grids modeling and control has become a more challenging issue due to
the increasing penetration of RESs, changing system structure and the integration
of new storage systems, controllable loads and power electronics technologies,
and reduction of system inertia. Conventional modeling and control designs
may not be any more effective to satisfy all specified objectives in various operation
modes of modern power grids. These challenging issues set new demand for
the development of more flexible, rapid, effective, precise, and adaptive
approaches for power system dynamic monitoring, stability/security analysis,
and control problems. Thanks to recent advances in control, communication,
and computing technologies, it is possible to tackle mentioned challenges by
implementing a data-driven-based modeling and control framework as shown
in Figure 1.2.
The system data are collected from the distributed PMUs in the grid through
a secure communication network. The development of information and commu-
nication technology (ICT) enables more flexibility in wide-area monitoring of
power system with fast and large data transmission. Especially, the wide-area
measurement system (WAMS) with PMUs is a promising technique as one of
the smart grid technologies in the bulk power grid.
1.3 Data-Driven Wide-Area Power System Monitoring and Control 9
Power System Data

(PMU, Real-time Measurement)
Data
Storage
Pre-filtering and De-noising
Parameters Estimation
and Dynamic Modeling
Emergency Control and Automatic

Protection Schemes Continuous Control
Figure 1.2 An overall data-driven control framework for renewable integrated power
systems.
The measured data are locally saved and then collected by phasor data concen-
trators (PDCs) for the post analysis or sent to a remote location via a standard
data format. These data with the time stamp of the synchronized GPS in real time
may applied for parameter and state estimations and finally used for the system
protection and/or real-time control. Figure 1.3 shows how a PMU-based WAMS
can provide data for the power system control center to generate continuous (in
normal states) and discontinuous control (in off-normal states) commands.
Before any application, the collected PMU data need to be cleaned and de-noised
and employed by the data processors for estimation, modeling, and control pur-
poses. The proposed de-noising method may use a rolling-averaging window with
pre-specified length to remove noise from the recorded data. The block of para-
meters estimation algorithms contains high fast and precise algorithms for estima-
tion of some important parameters and transient characteristics that are required
to use in control tuning algorithm or to detect a contingency and triggering the
emergency control and protection schemes. In case of crossing the assigned thresh-
olds showing an off-normal and emergency condition, the recorded data and some
estimated parameters are used to detect the amount of mismatch (size of distur-
bance) for the emergency control and protection schemes such as load shedding
10 1 Introduction
GPS
Continuous control
Data Storage
Data Analysis,
Parameter Estimation
and Control Center
Synchrophasor
Communication PDC
System (SPCS)
Emergency control
Power Grid
Discontinuous control and protection
Figure 1.3 PMU-based wide-area measurement system and control.
algorithms. Otherwise, the estimated parameters such as scheduling parameters

are employed by the continuous control systems.
As mentioned, using significant number of distributed micro-sources into power
systems adds new technical challenges. As the electric industry seeks to reliably
integrate large amounts of DGs/RESs into the power system in regulated environ-
ment, considerable effort is needed to accommodate and effectively manage the
installed micro-sources. A key aspect is how to handle changes in topology and
dynamics caused by penetration of numerous DGs/RESs in the network and
how to make the power grid robust and able to take advantage of the potential
flexibility of distributed micro-sources. In a modern control framework, a part
of power produced by available DGs/RESs in the grid are used as a primary energy
source of inertia emulator to provide virtual inertia as a supporting control for
abovementioned controllers (like a fine tuner) to improve power grid stability.
1.4 Dynamics Modeling and Parameters Estimation
From a system dynamic point of view, the bulk generating units, due to their high
inertia, provide a long time constant; such that the rotor speed and thus the grid
frequency cannot alter suddenly, while the load changes. Hence, the total rotating
mass enhances the dynamic stability. In future, a significant share of DGs/RESs/
MGs in the electric power grids is expected. This increases the total system gener-
ation power, while does not contribute to the system rotational inertia. System
dynamics are faster in power systems with low rotational inertia, making control
and power system operation more challenging [32].
A complete understanding of reliability considerations via effective modeling/

aggregation techniques is vital to identify a variety of ways that power grids can
accommodate the large-scale integration of the distributed micro-sources in
future. An accurate dynamic model is needed for the stability analysis and control
synthesis in a grid with a high degree of DGs/RESs penetration. A proper dynamic
modeling and aggregation of the DGs/RESs and MGs, for performance and stabil-
ity studies, is a key issue to understand the dynamic impact of distributed micro-
sources and simulate their functions in new environment.
The power system is a nonlinear multivariable time-varying system. It is repre-
sented by a nonlinear set of equations for the generators (swing equations), for the
transmission lines and for the loads, which for a typical power system has a few
hundreds of states. For the control design purpose, usually a reduced-order line-
arized model around an operating point is used and it is assumed that all system
parameters are known and time-invariant. These assumptions, however, are not
valid in a real power system with dominated DGs/RESs/MGs. The main dynamic
modes of the system are varying stochastically during a day because of the
variation of load and aggregated inertia. The dynamic modes will change more sig-
nificantly by integration of new RESs into the power system (e.g. because of
long-term variation of the mean value of the aggregated inertia). Therefore, a fixed
linearized time-invariant model will not represent correctly the behavior of the
power system.
The frequency response of the system can be identified offline/online using the
data for different load and generation configurations (when the share of DGs/RESs
is increased) and saved in a database for the models. The small variation of the
system (originated from measurement noise, load variation, and system nonline-
arity) will be modeled by frequency domain uncertainty. The long-term effect of
change in system inertia can be considered by identifying several frequency-
domain models for different levels of RES penetration. One can represent this
model’s database by an LPV model [116]. It should be mentioned that the model
of the power system for the frequency, voltage, and rotor angle is different because
they have different inputs and outputs and scheduling parameters.
1.4.1 Modeling of Frequency, Voltage, and Angle Controls

The participant bulk SGs with different participation factors are the main actua-
tors for the frequency control system. Following a disturbance, the variation of fre-
quency and tie-line power is applied to the LFC system via the ACE signal. Then,
depending on the accessible amount of regulation power, the LFC system will be
activated to compensate the power grid frequency and return it to the nominal
value. The LFC system can attenuate the frequency and active power changes from
tenth of seconds to few minutes. Therefore, the ACE signal may provide the output
12 1 Introduction
of system model for frequency control. Considering the frequency response

dynamics [32], the candidate scheduling parameters are system inertia, aggregated
generating time constant, droop and damping coefficient. The measurement-based
dynamics identification and system modeling will be for adaptive control and
online parameters tuning of the LFC system. The increasing size and diversifica-
tion of demand/power sources magnify the importance of this issue in the modern
power grids.
Unlike grid frequency, since the voltage is known as a local variable, a higher
number of measured points are required. For proposing the data-driven models,
several concepts like Thevenin equivalent system and oscillation model will be
applied and the results will be used for the grid voltage analysis, and then optimal
tuning of AVRs. The measurements of voltage, current, and phase deviations of
existing nodes are considered as the most important inputs of model. The system
output can be the terminal voltage change of the SGs. In order to construct an
appropriate LPV model for voltage control, relevant scheduling parameters must
be selected. There are several choices for the scheduling parameters (low-
frequency resonance mode, reactive power of the system, Thevenin equivalent
impedance/admittance, etc.) that must be compared and discussed using effective
analytical and simulation-based studies. In choosing efficient scheduling para-
meters, a tradeoff between accuracy and simplicity of the resulting LPV model
is needed. The measurements are also used to perform some important graphical
tools and curves to evaluate the stress conditions and to analyze the voltage sta-
bility criteria. For instance, the data are fitted to the active power–voltage curve
(PV curve) of the equivalent system by suitable fitting approaches such as least
squares method.
The SGs equipped with a PSS and RESs participating in power oscillation damp-
ing are considered as the main actuators of the control system. It is assumed that
some of the RESs are not operating at their maximum generated power, so that they
can help for power oscillation damping by reducing/increasing a small percentage
of their power generation. The system to be controlled is a multivariable system,
where the inputs are the reference voltages for the AVR of the SGs as well as the
auxiliary signals that will be added to the reference active and reactive powers of
RESs. The system outputs can be the active power or the speed (or both, depending
on the type of PSSs) of the SGs and the measured active and reactive powers of the
RESs. In order to construct an appropriate model for power oscillation damping, we
need to choose the scheduling parameters.
1.4.2 Parameters Estimation

For estimation of all required parameters, the recorded data from the installed
PMUs can be used. As shown in Figure 1.2, firstly, a de-noising methodology,
mostly based on rolling averaging windows, can be employed to prepare the

received PMU data for further processing. Afterwards, some data-driven-based
algorithms are used to estimate the most important parameters (e.g., system iner-
tia, droop characteristic, and damping factor) required for building low order
power system models such as frequency response model [32] and oscillation model
[3]. In real-time operation, an accurate and fast estimation of parameters is
required. The estimated parameters can be used in the auto-tuning algorithm of
the controllers.
In a modern power system, in addition to scheduling parameters, several algo-
rithms need to be developed for online estimating of other important system para-
meters such as synchronizing coefficient between various areas, rate of change of
frequency/voltage (ROCOF/ROCOV), frequency/voltage nadir, and time occur-
rence of frequency/voltage nadir. The estimations must be fast enough and should
cover the issues related to the existing time delay. These data-driven-based estima-
tion algorithms can be analytically developed based on the concept of swing equa-
tion, base-case systems, regression, and curve fitting. These measurement-based
dynamics identification and system modeling can be used for adaptive online para-
meters tuning of the targeted controllers. The increasing size and diversification of
demand/power sources magnify the importance of this issue in the modern power
grids. The estimation approaches may be applicable for both on-line and off-line
methods. However, for on-line applications, shorter data windows must be used.
PMU data-based power-load imbalance estimation is a key estimation to suc-
cessfully handle the emergency control strategies, e.g. load shedding, and protec-
tion plans. In case of detecting a contingency or an emergency condition, following
comparing of frequency, voltage, and their rate of changes with the specified
threshold values, an estimation algorithm must estimate the size of disturbance
to use in the available emergency control systems and special protection schemes.
This online estimation is an important issue to realize a successful load-shedding
scheme with minimum amount of shed load.
Conventionally, for the estimation of the size of load-power mismatch, the swing
equation is used. The estimated imbalance based on this method may far from real
power mismatch as it relies on three worst assumptions: (i) there is no additional
active power variation except for that of disturbance, (ii) there is a negligible reac-
tive-power imbalance in response to sudden active power imbalance, and (iii) the
inertia constant assumed to be known. This under/over-estimation causes inaccu-
rate calculation of total amount of load to be shed.
In [117], for estimating the size of disturbance, some appropriate base-case fea-
tures are selected from a set of pre-defined base cases. Moreover, an efficient yet
simple logic is defined to select appropriate base-case for the received data. This
approach benefits from the use of PMU data to precise calculation of the required
amount of load to be shed, in one-step and in a short time in comparison with the
14 1 Introduction
actual time. The proposed scheme relies on fast, yet iterative, estimation of fre-
quency nadir, and time of minimum frequency occurrence. Accordingly, the iner-
tia constant as well as the size of power mismatch are estimated which, in turn,
compares with the maximum size of imbalance, satisfying the pre-specified thresh-
olds, to determine the amount of shed load.
1.5 Summary
Modern power grids face new technical challenges arising from the increasing
penetration of power-electronic-connected RESs/DGs. Increasing MGs/DGs pen-
etration level may adversely affect frequency response and voltage and system con-
trol and lead to degraded performance of traditional control schemes. This, in turn,
may result in large deviations and, potentially, system instability.
This chapter provides the pre-requirement terminology and general background
for the next chapters of this book. The term power system stability and control with
an updated brief review on the areas of frequency, voltage, and angle controls, con-
cerning the penetration of RESs/DGs, is discussed. In response to the existing chal-
lenges in penetration of more RESs/DGs to the grid, the necessity of using data-
driven modeling, parameters estimation, and control synthesis in wide-area power
systems is emphasized.
References
1. Steinmetz, C.P. (1920). Power control and stability of electric generating stations.
Transactions of the American Institute of Electrical Engineers XXXIX (2):
1215–1287.
2. Smith, J.R., Andersson, G., and Taylor, C.W. (1996). Annotated bibliography on
power system stability controls: 1986–1994. IEEE Transactions on Power Systems
11 (2): 794–800.
3. Bevrani, H., Watanabe, M., and Mitani, Y. (2014). Power System Monitoring and
Control. Wiley.
4. Bevrani, H. and Hiyama, T. (2009). On load-frequency regulation with time
delays: design and real time implementation, IEEE Transactions on Energy
Conversion 24 (1): 292–300.
5. Bevrani, H. and Hiyama, T. (2011). Intelligent Automatic Generation Control.
CRC Press.
6. Kundur, P. (1994). Power System Stability and Control, vol. 7. New York:
McGraw-Hill.
References 15
7. Kundur, P., Paserba, J., Ajjarapu, V. et al. (2004). Definition and classification of
power system stability ieee/cigre joint task force on stability terms and definitions.
IEEE Transactions on Power Systems 19 (3): 1387–1401.
8. Pierre, J.W., Trudnowski, D., Donnelly, M. et al. (2012). Overview of system
identification for power systems from measured responses. IFAC Proceedings
Volumes 45 (16): 989–1000.
9. Eriksson, R. and Soder, L. (2011). Wide-area measurement system-based subspace
identification for obtaining linear models to centrally coordinate controllable
devices. IEEE Transactions on Power Delivery 26 (2): 988–997.
10. Zhou, N., Lu, S., Singh, R., and Elizondo, M.A. (2011). Calibration of reduced
dynamic models of power systems using phasor measurement unit (PMU) data.
2011 North American Power Symposium, Boston, MA (2011), pp. 1–7. doi: https://
doi.org/10.1109/NAPS.2011.6024873.
11. Zhang, J., Lu, C., and Han, Y. (2013). MIMO identification of power system with
low level probing tests: applicability comparison of subspace methods. IEEE
Transactions on Power Systems 28 (3): 2907–2917.
12. Wiseman, B.P., Chen, Y., Xie, L., and Kumar, P. (2016). PMU-based reduced-order
modeling of power system dynamics via selective modal analysis. In: 2016
IEEE/PES Transmission and Distribution Conference and Exposition (T&D),
1–5. IEEE.
13. Liu, H., Zhu, L., Pan, Z. et al. (2016). Comparison of MIMO system identification
methods for electromechanical oscillation damping estimation. 2016 IEEE Power
and Energy Society General Meeting (PESGM), Boston, MA (2016), pp. 1–5.
doi: https://doi.org/10.1109/PESGM.2016.7741834.
14. Tuttelberg, K., Kilter, J., and Uhlen, K. (2017). Comparison of system
identification methods applied to analysis of inter-area modes. Proceedings of
International Power Systems Transients Conference 2017, Seoul, South Korea
(26–29 June 2017).
15. Ghasemi, H. and Canizares, C.A. (2008). Confidence intervals estimation in the
identification of electromechanical modes from ambient noise. IEEE Transactions
on Power Systems 23 (2): 641–648.
16. Dosiek, L., Pierre, J.W., and Follum, J. (2013). A recursive maximum likelihood
estimator for the online estimation of electromechanical modes with error
bounds. IEEE Transactions on Power Systems 28 (1): 441–451.
17. Uhlen, K., Warland, L., Gjerde, J.O. et al. (2008). Monitoring amplitude,
frequency and damping of power system oscillations with PMU measurements.
2008 IEEE Power and Energy Society General Meeting – Conversion and Delivery of
Electrical Energy in the 21st Century, Pittsburgh, PA (2008), pp. 1–7, doi: https://
doi.org/10.1109/PES.2008.4596661.
18. Tripathy, P., Srivastava, S.C., and Singh, S.N. (2011). A modified TLS-ESPRIT-
based method for low frequency mode identification in power systems utilizing
16 1 Introduction
synchrophasor measurements. IEEE Transactions on Power Systems 26 (2):

719–727.
19. Rogers, K.M., Spadoni, R.D., and Overbye, T.J. (2011). Identification of power
system topology from synchrophasor data. 2011 IEEE/PES Power Systems
Conference and Exposition, Phoenix, AZ (2011), pp. 1–8, doi: https://doi.org/
10.1109/PSCE.2011.5772462.
20. Nabavi, S. and Chakrabortty, A. (2013). Topology identification for dynamic
equivalent models of large power system networks. 2013 American Control
Conference, Washington, DC, (2013), pp. 1138–1143. doi: https://doi.org/10.1109/
ACC.2013.6579989.
21. Wang, X., Bialek, J.W., and Turitsyn, K. (2018). PMU-based estimation of
dynamic state jacobian matrix and dynamic system state matrix in ambient
conditions. IEEE Transactions on Power Systems 33 (1): 681–690.
22. Tuttelberg, K., Kilter, J., Wilson, D., and Uhlen, K. (2018). Estimation of power
system inertia from ambient wide area measurements. IEEE Transactions on
Power Systems 33 (6): 7249–7257.
23. Zeng, F., Zhang, J., Zhou, Y., and Qu, S. (2020). Online identification of inertia
distribution in normal operating power system. IEEE Transactions on Power
Systems 35 (4): 3301–3304. https://doi.org/10.1109/TPWRS.2020.2986721.
24. Concordia, C. and Kirchmayer, L. (1953). Tie-line power and frequency control of
electric power systems [includes discussion]. Transactions of the American
Institute of Electrical Engineers. Part III: Power Apparatus and Systems 72 (3):
562–572.
25. System Controls Subcommittee of the Power System Engineering Committee of
the IEEE Power Group (1970). IEEE standard definitions of terms for automatic
generation control on electric power systems. IEEE Transactions on Power
Apparatus and Systems PAS-89 (6): 1356–1364.
26. I. C. Report (1973). Dynamic models for steam and hydro turbines in power
system studies. IEEE Transactions on Power Apparatus and Systems PAS-92 (6):
1904–1915.
27. Jaleeli, N., VanSlyck, L.S., Ewart, D.N. et al. (1992). Understanding automatic
generation control. IEEE Transactions on Power Systems 7 (3): 1106–1122.
28. Pathak, N., Bhatti, T.S., and Verma, A. (2017). Accurate modelling of discrete
AGC controllers for interconnected power systems. IET Generation,
Transmission & Distribution 11 (8): 2102–2114.
29. Moawwad, A., El-Saadany, E.F., and El Moursi, M.S. (2018). Dynamic security-
constrained automatic generation control (AGC) of integrated ac/dc power
networks. IEEE Transactions on Power Systems 33 (4): 3875–3885.
30. Ledva, G.S., Vrettos, E., Mastellone, S. et al. (2018). Managing communication
delays and model error in demand response for frequency regulation. IEEE
References 17
31. Ibraheem, P., Kumar, and Kothari, D.P. (2005). Recent philosophies of automatic
generation control strategies in power systems. IEEE Transactions on Power
Systems 20 (1): 346–357.
32. Bevrani, H. (2014). Robust Power System Frequency Control, 2e. Gewerbestrasse,
Switzerland: Springer.
33. Ulbig, A., Borsche, T.S., and Andersson, G. (2014). Impact of low rotational inertia
on power system stability and operation. IFAC Proceedings Volumes 47 (3):
7290–7297.
34. Jaleeli, N. and VanSlyck, L.S. (1999). NERC’s new control performance standards.
35. Hain, Y., Kulessky, R., and Nudelman, G. (2000). Identification-based power unit
model for load-frequency control purposes. IEEE Transactions on Power Systems
15 (4): 1313–1321.
36. Chang-Chien, L.R., Hoonchareon, N.-B., Ong, C.-M., and Kramer, R.A. (2003).
Estimation of /spl beta/ for adaptive frequency bias setting in load frequency
control. IEEE Transactions on Power Systems 18 (2): 904–911.
37. Wilches-Bernal, F., Concepcion, R., Neely, J.C. et al. (2018). Communication
enabled fast acting imbalance reserve (CE-FAIR). IEEE Transactions on Power
Systems 33 (1): 1101–1103.
38. Zhang, G. and McCalley, J.D. (2018). Estimation of regulation reserve
requirement based on control performance standard. IEEE Transactions on Power
Systems 33 (2): 1173–1183.
39. Polajzer, B., Brezovnik, R., and Ritonja, J. (2017). Evaluation of load frequency
control performance based on standard deviational ellipses. IEEE Transactions on
Power Systems 32 (3): 2296–2304.
40. Avila, T., Gutierrez, E., and Chavez, H. (2017). Performance standard-compliant
secondary control: the case of Chile. IEEE Latin America Transactions 15 (7):
1257–1262.
41. Douglas, L.D., Green, T.A., and Kramer, R.A. (1994). New approaches to the AGC
nonconforming load problem. IEEE Transactions on Power Systems 9 (2): 619–628.
https://doi.org/10.1109/59.317682.
42. Trovato, V., Sanz, I.M., Chaudhuri, B., and Strbac, G. (2017). Advanced control of
thermostatic loads for rapid frequency response in Great Britain. IEEE
43. Delavari, A. and Kamwa, I. (2018). Improved optimal decentralized load
modulation for power system primary frequency regulation. IEEE Transactions
44. Pan, C. and Liaw, C. (1989). An adaptive controller for power system load-
frequency control. IEEE Transactions on Power Systems 4 (1): 122–128.
45. Vajk, I., Vajta, M., Keviczky, L. et al. (1985). Adaptive load-frequency control of
the Hungarian power system. Automatica 21 (2): 129–137.
18 1 Introduction
46. Wang, W., Li, Y., Cao, Y. et al. (2018). Adaptive droop control of VSC-MTDC
system for frequency support and power sharing. IEEE Transactions on Power
Systems 33 (2): 1264–1274.
47. Prostejovsky, A.M., Marinelli, M., Rezkalla, M. et al. (2018). Tuningless load
frequency control through active engagement of distributed resources. IEEE
48. Stankovic, A.M., Tadmor, G., and Sakharuk, T.A. (1998). On robust control
analysis and design for load frequency regulation. IEEE Transactions on Power
Systems 13 (2): 449–455.
49. Rerkpreedapong, D., Hasanovic, A., and Feliachi, A. (2003). Robust load
frequency control using genetic algorithms and linear matrix inequalities. IEEE
50. Ojaghi, P. and Rahmani, M. (2017). LMI-based robust predictive load frequency
control for power systems with communication delays. IEEE Transactions on
Power Systems 32 (5): 4091–4100.
51. Zhang, C., Jiang, L., Wu, Q.H. et al. (2013). Delay-dependent robust load
frequency control for time delay power systems. IEEE Transactions on Power
Systems 28 (3): 2192–2201.
52. Zhao, J., Mili, L., and Milano, F. (2018). Robust frequency divider for power
system online monitoring and control. IEEE Transactions on Power Systems 33 (4):
4414–4423.
53. Aliabadi, S.F., Taher, S.A., and Shahidehpour, M. (2018). Smart deregulated grid
frequency control in presence of renewable energy resources by EVs charging
control. IEEE Transactions on Smart Grid 9 (2): 1073–1085.
54. Wang, D., Liang, L., Hu, J. et al. (2018). Analysis of low-frequency stability in grid
tied DFIGs by non-minimum phase zero identification. IEEE Transactions on
Energy Conversion 33 (2): 716–729.
55. Liu, Y., Jiang, L., Wu, Q.H., and Zhou, X. (2017). Frequency control of DFIG-
based wind power penetrated power systems using switching angle controller and
AGC. IEEE Transactions on Power Systems 32 (2): 1553–1567.
56. Pradhan, C. and Bhende, C.N. (2017). Frequency sensitivity analysis of load
damping coefficient in wind farm-integrated power system. IEEE Transactions on
Power Systems 32 (2): 1016–1029.
57. Golpira, H., Seifi, H., Messina, A.R., and Haghifam, M. (2016). Maximum
penetration level of microgrids in large-scale power systems: frequency stability
viewpoint. IEEE Transactions on Power Systems 31 (6): 5163–5171.
58. Leon, A.E. (2018). Short-term frequency regulation and inertia emulation using
an MMC-based MTDC system. IEEE Transactions on Power Systems 33 (3):
2854–2863.
References 19
59. Rakhshani, E., Remon, D., Cantarellas, A.M. et al. (2017). Virtual synchronous
power strategy for multiple HVDC interconnections of multi-area AGC power
systems. IEEE Transactions on Power Systems 32 (3): 1665–1677.
60. Li, D., Zhu, Q., Lin, S., and Bian, X.Y. (2017). A self-adaptive inertia and damping
combination control of VSG to support frequency stability. IEEE Transactions on
Energy Conversion 32 (1): 397–398.
61. Wu, Y., Yang, W., Hu, Y., and Dzung, P.Q. (2019). Frequency regulation at a wind
farm using time varying inertia and droop controls. IEEE Transactions on Industry
Applications 55 (1): 213–224.
62. Fang, J., Li, H., Tang, Y., and Blaabjerg, F. (2018). Distributed power system
virtual inertia implemented by grid-connected power converters. IEEE
Transactions on Power Electronics 33 (10): 8488–8499.
63. Li, Y., Xu, Z., Ostergaard, J., and Hill, D.J. (2017). Coordinated control strategies
for offshore wind farm integration via VSC-HVDC for system frequency support.
IEEE Transactions on Energy Conversion 32 (3): 843–856.
64. Ahmadyar, A.S. and Verbic, G. (2017). Coordinated operation strategy of wind
farms for frequency control by exploring wake interaction. IEEE Transactions on
Sustainable Energy 8 (1): 230–238.
65. Izadkhast, S., Garcia-Gonzalez, P., Frias, P., and Bauer, P. (2017). Design of plug-
in electric vehicle’s frequency-droop controller for primary frequency control and
performance assessment. IEEE Transactions on Power Systems 32 (6): 4241–4254.
66. Hwang, M., Muljadi, E., Jang, G., and Kang, Y.C. (2017). Disturbance-adaptive
short-term frequency support of a DFIG associated with the variable gain based
on the ROCOF and rotor speed. IEEE Transactions on Power Systems 32 (3):
1873–1881.
67. Attya, A.B.T. and Dominguez-Garcia, J.L. (2018). Insights on the provision of
frequency support by wind power and the impact on energy systems. IEEE
Transactions on Sustainable Energy 9 (2): 719–728.
68. Tielens, P. and Van Hertem, D. (2017). Receding horizon control of wind power to
provide frequency regulation. IEEE Transactions on Power Systems 32 (4): 2663–2672.
69. Garmroodi, M., Verbic, G., and Hill, D.J. (2018). Frequency support from wind
turbine generators with a time-variable droop characteristic. IEEE Transactions
on Sustainable Energy 9 (2): 676–684.
70. Khooban, M., Dragicevic, T., Blaabjerg, F., and Delimar, M. (2018). Shipboard
microgrids: a novel approach to load frequency control. IEEE Transactions on
71. Benysek, G., Bojarski, J., Smolenski, R. et al. (2018). Application of stochastic
decentralized active demand response (DADR) system for load frequency control.
IEEE Transactions on Smart Grid 9 (2): 1055–1062.
20 1 Introduction
72. Vrettos, E., Ziras, C., and Andersson, G. (2017). Fast and reliable primary
frequency reserves from refrigerators with decentralized stochastic control. IEEE
73. Short, J.A., Infield, D.G., and Freris, L.L. (2007). Stabilization of grid frequency
through dynamic demand control. IEEE Transactions on Power Systems 22 (3):
1284–1293.
74. Molina-Garcia, A., Bouffard, F., and Kirschen, D.S. (2011). Decentralized
demand-side contribution to primary frequency control. IEEE Transactions on
Power Systems 26 (1): 411–419.
75. Zhao, H., Wu, Q., Huang, S. et al. (2018). Hierarchical control of thermostatically
controlled loads for primary frequency support. IEEE Transactions on Smart Grid
9 (4): 2986–2998.
76. Yao, E., Wong, V.W.S., and Schober, R. (2017). Robust frequency regulation
capacity scheduling algorithm for electric vehicles. IEEE Transactions on Smart
Grid 8 (2): 984–997.
77. Ferraro, P., Crisostomi, E., Raugi, M., and Milano, F. (2017). Analysis of the
impact of microgrid penetration on power system dynamics. IEEE Transactions
78. Ferraro, P., Crisostomi, E., Shorten, R., and Milano, F. (2018). Stochastic
frequency control of grid connected microgrids. IEEE Transactions on Power
Systems 33 (5): 5704–5713.
79. Larsen, E. and Sener, F. (1996). Facts Applications. Catalogue No. 96TP116-0.
80. IEEE (1990). Voltage Stability of Power Systems: Concepts, Analytical Tools and
Industry Experience. IEEE Technical Report 90YH0358-2-PWR. IEEE/PES.
81. Balu, C. and Maratukulam, D. (1994). Power System Voltage Stability.
McGraw-Hill.
82. Van Cutsem, T. and Vournas, C. (2007). Voltage Stability of Electric Power Systems.
Springer Science & Business Media.
83. Kamwa, I., Grondin, R., and Hebert, Y. (2001). Wide-area measurement based
stabilizing control of large power systems – a decentralized/hierarchical
approach. IEEE Transactions on Power Systems 16 (1): 136–153.
84. Taylor, C.W., Erickson, D.C., Martin, K.E. et al. (2005). WACS wide-area stability
and voltage control system: R & D and online demonstration. Proceedings of the
IEEE 93 (5): 892–906.
85. Andersson, G., Bel, C.A., and Canizares, C. (2009). Frequency and voltage control.
In: Electric Energy Systems: Analysis and Operation. CRC Press.
86. Ilic, M.D., Liu, X., Leung, G. et al. (1995). Improved secondary and new tertiary
voltage control. IEEE Transactions on Power Systems 10 (4): 1851–1862.
87. Corsi, S., Pozzi, M., Sabelli, C., and Serrani, A. (2004). The coordinated automatic
voltage control of the Italian transmission grid-Part I: reasons of the choice and
References 21
overview of the consolidated hierarchical system. IEEE Transactions on Power

Systems 19 (4): 1723–1732.
88. Corsi, S., Pozzi, M., Sforna, M., and Dell’Olio, G. (2004). The coordinated
automatic voltage control of the italian transmission grid-Part II: control
apparatuses and field performance of the consolidated hierarchical system. IEEE
89. Guo, Q., Sun, H., Zhang, M. et al. (2013). Optimal voltage control of PJM smart
transmission grid: study, implementation, and evaluation. IEEE Transactions on
Smart Grid 4 (3): 1665–1674.
90. Xiao, W., Torchyan, K., El Moursi, M.S., and Kirtley, J.L. (2014). Online
supervisory voltage control for grid interface of utility-level PV plants. IEEE
91. A. Awadhi, N. and Moursi, M.S.E. (2017). A novel centralized PV power plant
controller for reducing the voltage unbalance factor at transmission level
interconnection. IEEE Transactions on Energy Conversion 32 (1): 233–243. https://
doi.org/10.1109/TEC.2016.2620477.
92. Glavic, M. and Van Cutsem, T. (2011). A short survey of methods for voltage
instability detection. 2011 IEEE Power and Energy Society General Meeting,
Detroit, MI (2011), pp. 1–8, doi: https://doi.org/10.1109/PES.2011.6039311.
93. Robbins, B.A., Hadjicostis, C.N., and Dominguez-Garcia, A.D. (2013). A two-stage
distributed architecture for voltage control in power distribution systems. IEEE
94. Zeraati, M., Hamedani Golshan, M.E., and Guerrero, J.M. (2019). A consensus-
based cooperative control of PEV battery and PV active power curtailment for
voltage regulation in distribution networks. IEEE Transactions on Smart Grid
10 (1): 670–680.
95. Li, Z., Guo, Q., Sun, H. et al. (2018). A distributed transmission-distribution
coupled static voltage stability assessment method considering distributed
generation. IEEE Transactions on Power Systems 33 (3): 2621–2632.
96. Popovic, D.H., Hill, D.J., and Wu, Q. (2002). Optimal voltage security control of
power systems. International Journal of Electrical Power & Energy Systems 24 (4):
305–320.
97. Larsson, M. and Karlsson, D. (2003). Coordinated system protection scheme
against voltage collapse using heuristic search and predictive control. IEEE
98. Ma, H. and Hill, D.J. (2018). A fast local search scheme for adaptive coordinated
voltage control. IEEE Transactions on Power Systems 33 (3): 2321–2330.
99. Ghahremani, E. and Kamwa, I. (2016). Local and wide-area PMU-based
decentralized dynamic state estimation in multi-machine power systems. IEEE
22 1 Introduction
100. Raoufat, M.E., Tomsovic, K., and Djouadi, S.M. (2016). Virtual actuators for wide-
area damping control of power systems. IEEE Transactions on Power Systems
31 (6): 4703–4711.
101. Mohagheghi, S., Venayagamoorthy, G.K., and Harley, R.G. (2007). Optimal wide
area controller and state predictor for a power system. IEEE Transactions on
102. Mithulananthan, N., Canizares, C.A., Reeve, J., and Rogers, G.J. (2003).
Comparison of PSS, SVC, and STATCOM controllers for damping power system
oscillations. IEEE Transactions on Power Systems 18 (2): 786–792.
103. Bian, X.Y., Geng, Y., Lo, K.L. et al. (2016). Coordination of PSSs and SVC damping
controller to improve probabilistic small-signal stability \\of power system with
wind farm integration. IEEE Transactions on Power Systems 31 (3): 2371–2382.
104. Padhy, B.P., Srivastava, S.C., and Verma, N.K. (2017). A wide-area damping
controller considering network input and output delays and packet drop. IEEE
105. Giri, J. (2015). Proactive management of the future grid. IEEE Power and Energy
Technology Systems Journal 2 (2): 43–52.
106. Wu, X., Dorer, F., and Jovanovic, M.R. (2016). Input-output analysis and
decentralized optimal control of inter-area oscillations in power systems. IEEE
107. Zacharia, L., Hadjidemetriou, L., and Kyriakides, E. (2018). Integration of
renewables into the wide area control scheme for damping power oscillations.
108. Surinkaew, T. and Ngamroo, I. (2016). Hierarchical coordinated wide area and
local controls of DFIG wind turbine and PSS for robust power oscillation
damping. IEEE Transactions on Sustainable Energy 7 (3): 943–955.
109. Wang, S., Meng, X., and Chen, T. (2012). Wide-area control of power systems
through delayed network communication. IEEE Transactions on Control Systems
Technology 20 (2): 495–503.
110. Mokhtari, M. and Aminifar, F. (2014). Toward wide-area oscillation control
through doubly-fed induction generator wind farms. IEEE Transactions on Power
Systems 29 (6): 2985–2992.
111. Zhang, Y. and Bose, A. (2008). Design of wide-area damping controllers for inter-
area oscillations. IEEE Transactions on Power Systems 23 (3): 1136–1143.
112. Zenelis, I. and Wang, X. (2018). Wide-area damping control for interarea
oscillations in power grids based on PMU measurements. IEEE Control Systems
Letters 2 (4): 719–724.
113. Youseian, R., Bhattarai, R., and Kamalasadan, S. (2017). Transient stability
enhancement of power grid with integrated wide area control of wind farms and
synchronous generators. IEEE Transactions on Power Systems 32 (6): 4818–4831.
References 23
114. El-Guindy, A., Schaab, K., Schurmann, B. et al. (2017). Formal lpv control for
transient stability of power systems. In: 2017 IEEE Power & Energy Society General
Meeting, 1–5. IEEE.
115. Nogueira, F.G., Junior, W.B., da Costa Junior, C.T., and Lana, J.J. (2018).
LPV-based power system stabilizer: identification, control and field tests. Control
Engineering Practice 72: 53–67.
116. Toth, R. (2010). Modeling and Identification of Linear Parameter-Varying Systems,
vol. 403. Springer.
117. Golpira, H. and Bevrani, H. (2020). Frequency Analysis Based Centralized Load
Shedding and Island Detection Using PMU Data. Technical Report, Iran Grid
Management Company, IGMC1927, (In Persian).
25
MG Penetrated Power Grid Modeling
Power grids worldwide have experienced a significant transformation, which has

been characterized by increased penetrations of distributed renewable generation.
With advances in communication, measurement, and control technologies, the
technical and economical merits of utility-scale microgrids (MGs), as a group of
distributed energy resources (DERs), interconnected loads, and energy storage
systems (ESSs), make these technologies attractive to enhance the dynamic perfor-
mance of future power grids.
Because of uncertainty in future power system projections, together with dif-
ferent load scenarios, and the uncertain behavior of inverter-based distributed
generations (DGs), MG-integrated power system modeling is a critical aspect in
the development of control strategies and coordinated operation with other gen-
eration resources. Two specific aspects of interest for successful integration of MGs
are (i) the modeling of the host power grid and (ii) the development of MG
dynamic equivalents as seen from their point of interconnection with the systems.
This chapter addresses the problem of power system modeling, with emphasis
on the development of aggregate MG models for electromechanical stability stud-
ies. Both deterministic and stochastic models for distribution and transmission
applications are considered, and the concept of power system dynamic equivalen-
cing is introduced.
2.1 Introduction
The increased demand for electrical power, environmental concerns, pressing

need to reduce dependence on fossil fuel, and technological developments have
caused many countries to set an ambitious target for deployment of DERs
[1, 2]. From the network perspective, aggregate models are needed that circumvent
the need for central dispatch of a massive number of DERs and the inherent
26 2 MG Penetrated Power Grid Modeling
difficulties associated with the control and coordination of DERs [3, 4]. MGs, act-
ing as single controlled entities on the grid, can overcome some of the aforemen-
tioned limitations through local control of DERs/DGs, separation of generation
and corresponding loads from the distribution system in the presence of distur-
bances, and providing high local reliability for loads. In addition to the DERs/
DGs’ potential benefits, the conversion of only one-third of fuel energy into elec-
trical power and the poor efficiency of bulk power generation and high voltage
transmission systems make DERs/DGs in distribution network more pervasive
[3]. Moreover, the falling investment cost of small-scale power plants, develop-
ment of data communications and control technologies, the emerging potential
of DERs/DGs, and short installation time are the main incentives toward integrat-
ing significant amounts of DERs’ generation [5, 6]. It is envisaged that a large num-
ber of DERs/DGs will be connected to the host grid shortly, which will increase the
power system dimension and complexity.
In this chapter, a systematic methodology for modeling distributed MGs for inte-
gration studies is presented. This framework can be used to investigate the inte-
gration of high penetrations of DERs/MGs energy into the system.
2.2 Basic Concepts
2.2.1 Dynamic Equivalencing

High penetration of MGs may seriously affect distribution system dynamics
which, in turn, may affect overall power system stability and dynamics. To analyze
the associated impacts, a simple representation of the distribution network as a
constant power load (PQ) is not appropriate [4]. Moreover, conventional
approaches to model the host power grid fail to handle the complexity and uncer-
tainties of the MGs-penetrated grid.
To facilitate the dynamical studies of MGs-penetrated power grids, appropriate
dynamical models should be derived. Power system dynamic equivalencing is a
powerful means to study system dynamics when a significant integration of
MGs operating in grid-connected mode is considered. In this type of modeling,
the structure (and dynamics) of part of the system is simplified, while attempting
to preserve the important characteristics of the underlying dynamical processes
being investigated.
Generally, power system equivalent models can be classified into three main
categories: high frequency, low frequency, and wideband models depending
on the transient phenomena of interest [6–8]. The focus of this chapter is on
the derivation of low-frequency models to be used in stability studies. Commonly,
2.2 Basic Concepts 27
low-frequency models are formally utilized in simulating rotor-angle stability of

synchronous machines as well as for frequency stability studies.
Most of the research efforts in this area started in the 1970s and 1980s when com-
puting power was dramatically less than today. The technical subcommittee of
IEEE Power and Energy Society (PES) successfully reviewed the advances and
challenges in the state of the art [7]. While the current industry practice is to sim-
ulate the full model due to the availability of fast computers, this is not the case for
MGs-penetrated power grids. In other words, despite the availability of fast com-
putation power, the need for low-frequency equivalent models is still relevant,
especially in cases where the influence of power-electronics devices on low-
frequency oscillations is of interest.
2.2.2 Background on Study Zone and External System

In dynamic equivalencing studies, the power network of interest is commonly
divided into the study zone and external system. When the transient phenomena
of interest occur in a study zone, the external system is commonly replaced with
an appropriate equivalent model that preserves the main dynamic of concern [7].
Accordingly two commonly used power system simulation tools are (i) electro-
magnetic transient (EMT) models and (ii) transient stability models or “phasor
models”.
In EMT models, the conventional synchronous generator is replaced with a volt-
age source. This causes the elimination of electromechanical low-frequency dynam-
ics which is not the case for transient studies. The EMT representations are well
suited for the simulation of lightning and switching over voltages that is beyond
the scope of this book. On the other hand, in modern power grids where power-elec-
tronic-interfaced sources play an important role, the high-frequency equivalent net-
work representation for the external system is inadequate. By contrast, quasi-
steady-state-based approaches rely on the assumption that the required frequency
that defines the phasors and system parameters is equal to the nominal value [8].
Such an assumption is acceptable as long as only the rotor speed variations of syn-
chronous machines are considered to regulate the frequency. However, an increas-
ing number of devices characterized with zero rotational inertia, including flexible
loads providing load demand programs, DERs, ESSs, and high voltage direct current
(HVDC) transmission systems, are expected to contribute to frequency regulation.
These devices do not generally impose the frequency at their connection point with
the grid [8, 9].
Thus, there is, from a modeling point of view, the need to define with accuracy the
local frequency at crucial buses of the network. Therefore, the power system equiv-
alencing process considering the high penetration of renewable sources should
accurately incorporate the computation of the local frequency to the system model.
2.3 Power Grid Modeling
This section firstly discusses conventional power system modeling considering

local frequency estimation using a center-of-gravity (COG) formulation in physics.
To account for the uncertain system behavior, a stochastic approach is utilized to
model MGs/DGs in integration studies. Based on these representations, a meas-
urement-based approach is combined with the model to derive a simple, yet ana-
lytical equivalencing approach.
2.3.1 The Notion of Center-of Gravity (COG)

The same reasoning used in mechanics to introduce the concept of the COG can be
extended to define the center of inertia (COI). In physics, the COG is a unique
point in a body or group of particles, where the resultant torque due to gravity
forces vanishes [10, 11]. This section introduces COG-based approaches for power
system dynamic equivalencing as well as local frequency estimation.
2.3.1.1 Key Concept

To introduce the proposed formulation, assume that an interconnected power sys-
tem is divided into n areas {Ai, i = 1, …, n}, where a local COI is associated with
each area. The areas may be determined using coherency identification techniques
or be associated with the use of time-varying single (multi)-machine equivalent
(SIME) methods [12].
The motion of the COI for area Ai is given by [13]
d2 δCOI i t
M COI i = T mCOI i t − T eCOI i t , i = 1, …, n 21
dt 2
where δCOI i = j Ai M j δ j j Ai M j is the position of the COI, expressed in terms
of the individual rotor angles δj, M COI i = j Ai M j ωo is the equivalent inertia,
and T mCOI i t , T eCOI i t are the time-varying mechanical input power and electrical
output power, respectively, which include the effects of turbine dynamics and
other controllers [12].
With the COI description (2.1), used to represent the area response to external
forces, valuable information about local (bus) frequency behavior and inter-unit
synchronizing oscillations between generators are eliminated. Ideally, fast syn-
chronizing oscillations are suppressed when areas are chosen based on the
coherency identification techniques (uniform frequency for an area), such that
f j f Ai = f COI i , where f COI i = j Ai M j f j j Ai M j .
It is convenient to define fictitious transmission lines between key generators
(buses) and the COI to study the key generators’ motion, as illustrated in
Area n
Area 1
G1 Gj
δ1, M1 δj, Mj
δCOI , MCOI
1 1
XCOI1,I
COI XCOI1,j
XCOI1,m
Gm
δm, Mm
Figure 2.1 Area i illustrating the notion of the local COI. Dashed lines indicate virtual
connections between the motion of the area i COI and key system buses.
Figure 2.1. Conceptually, this is equivalent to use weighted strings or plumb line
techniques to find the COG in physical objects or systems.
Once the location of the COI is determined in Figure 2.1, expressions relating the
local frequencies fj, to the COI behavior can be obtained as
∂f j
fj= Δ f COI i , j Ai 22
∂ f COI i
where the sensitivities are determined from the parameters of the fictitious tie-
lines of Figure 2.1. By knowing these sensitivities, the propagation of slow fre-
quency oscillations and tie-line power exchanges may be evaluated.
Drawing on the above framework, Figure 2.2 schematically represents the posi-
tion of the COG relative to the local centers of inertia. Here, in analogy with the
single-area representation, Ptie tie
COI i ,COG and P COI i ,j represent virtual power flows across
the fictitious tie-lines interconnecting the ith local COI to the COG and the jth bus
Area 2
Local COI
dynamics
Area 1 δCOI2, MCOI2

tie
PCOI 1,COG
tie
PCOI 2,COG
δCOI1, MCOI1
COG
tie
Area k PCOI j,COG δCOG , MCOG
tie
PCOI n,COG
δCOIk , MCOIk Area n

δCOIn , MCOIn
Figure 2.2 Interconnected power system divided into areas illustrating the notion of the
COG. Dotted lines indicate virtual connections between the motions of centers of inertia and
the system’s COG.
to the associated local COI that are of interest for calculation. Two issues are of key
interest here:
1) The determination of the COG and its associated parameters, and

2) The computation of physical relationships between the COG and the local cen-
ters of inertia variables. These issues are discussed in detail below.
The motion of the COG expresses in terms of the individual COIs as

n n n
d2 δCOG d2 δCOI i
M COG = T mCOG − T eCOG = M COI i = T mCOI i − T COI i
dt 2 i=1 dt 2 i=1 i=1
2 3a
subject to the boundary conditions
n
Ptie
n
COI i ,COG
T tie
COI i ,COG = =0 2 3b
i=1 i=1
2π f COG
in which, power transfers from the ith local COI to the COG are assumed to be of
the general form [14]:
V COI i V COG
Ptie
COI i ,COG = sin δCOI i − δCOG 24
X tie
COI i ,COG
where X tie
COI i ,COG is the virtual reactance between area i and the COG. On the other
hand, the condition for the equilibrium of the COI relative to the COG and local
buses requires that
n n Ptie
COI i ,j Ptie
COI i ,COG
T tie tie
COI i ,j + T COI i ,COG = + =0 25
j=1
2π f COI i 2π f COI i
j=1
i n
where T tie tie

COI i ,j T COI i ,COG are the applied torques from bus j and COG, respectively.
Substituting (2.4) in (2.3b) results in
n n
Ptie
COI i ,COG
n
1 V COI i V COG
T tie
COI i ,COG = = tie
sin δCOI i −δCOG = 0
i=1 i=1
2π f COG i=1
2π f COG X COI ,COGi
26
and similarly, for area Ai,
n
1 V j V COI i Ptie
COI i ,COG
T tie tie
COI i ,j + T COI i ,COG = sin δ j −δ COI i + =0
j
2π f COI i X tie
COI i ,j 2π f COI i
j=1
i n
27
In interpreting this model, note that the first term of (2.7) on the right-hand side
explains the virtual transferred power between bus j and the local COI, while the
second one describes the interactions between the local COI and the COG. The set
of equations (2.1)–(2.7) describes the multi-area dynamic energy balance and is
well suited for the efficient analysis of power and frequency transients of large
interconnected power systems. In the new equivalent system of Figure 2.2, the
equations are expressed in terms of variables that have the potential to approxi-
mate frequency behavior using a simplified model from which both local and
global properties can be analyzed. In what follows, the relationship between the
global frequency and the local frequencies is investigated. First, some basic
assumptions are discussed to derive the equivalent model of Figure 2.2.
2.3.1.2 Basic Assumptions

Initially, several assumptions are introduced in the above framework to derive
simple algebraic relationships between the frequency of the COG, the frequencies
of the individual COIs, and the frequencies of key system buses, namely:
• The virtual impedances, i.e. X tie tie

COI i ,COG and X COI i ,j in (2.6) and (2.7), are calculated
based on the steady-state frequencies. This assumption can be easily under-
standable by considering that virtual impedances are calculated in relation to
each other. In other words, since at the equilibrium condition, the resultant tor-
que in the COG framework and the summation of local COI torques are zero, it
seems reasonable to assume that the ratio of virtual impedances remains
constant.
• Two different time horizons are assumed for voltage consideration in the
procedure of calculation of virtual impedances: (i) the period from when the
disturbance occurs to the time when the first under load tap changer (ULTC)
tap movement takes place and (ii) the time in which the ULTCs operate and bus
voltages increase to the reference values [15]. During the early period, bus voltage
magnitudes gradually come down to values lower than the pre-disturbance values.
The sensitivity of changes in the virtual impedances to the changes in bus voltage
magnitude, in these two periods, is assumed to be negligible. This means that the
ratio of impedances remains approximately constant over the time. Therefore, bus
voltage magnitudes are set to 1 pu in the process of calculation of impedances.
2.3.1.3 Modeling Formulation

To determine the equivalent parameters of the COG, assume that COI angles and
voltages are determined using the above procedures. It follows, therefore, that
the problem of calculation of the equivalent reactances X tie tie
COI i ,COG and X COI i ,j can
be posed as the solution of an optimization problem of the form of
n
1 V COI i V COG
min sin δCOI i − δCOG 28
tie
X COI ,COG i=1
2π f COG X tie
COI ,COG
i i
and
1 V j V COI i
min sin δ j − δCOI i + T tie
COI i ,COG 29
tie
X COI
i ,j j
2π f COI i X tie
COI i ,j
Since the minimum value of an absolute function is zero, the solutions of (2.8)
and (2.9) lead to the same results as solving (2.6) and (2.7). As the values of the
virtual impedances in (2.8) and (2.9) are used to weight the participation of each
area or bus in the multi-area dynamic energy balance, the optimization problem is
enforced to satisfy the conditions 0 < X tie

• < 1. The aim is to calculate the fictitious
impedances which produce the local and global torques that add up to zero, and
thus there is no need to guarantee the optimal solution. Genetic algorithms (GA)
could be an efficient method to calculate the fictitious reactances.
The computational procedure involved in determining the fictitious reactances
by the GA is given in the following steps:
• Step 1. Given rotor angle positions δj, j = 1, …, ng and bus voltage magnitudes
Vj, minimize the absolute resultant torque in (2.8) and (2.9) using a GA. An
initial population of 200 chromosomes characterizes the GA. For a system
including n areas, each chromosome consists of 10 × n genes, representing
the X tie
COI i ,COG; each unknown variable, i.e. the fictitious reactance may be repre-
sented by 10 binary bits. Minimization of (2.9) requires defining 10 × (m + 1)
genes, where m is the number of generator buses in the associated area. This
means that m + 1 fictitious reactances, including m X tie tie
COI i ,j and one X COI i ,COG ,
interact with each other to make that the resultant torque vanishes for the local
COIs. As a result, only m X tie COI i ,j are considered as unknown variables.
•• tie
Step 2. Set X COI i ,COG to the value obtained from (2.8).
Step 3. Compute frequency (and power) sensitivities in (2.2).
• Step 4. Calculate local and global frequency responses and tie-line net flows.
2.3.1.4 Local Frequency Estimation

From (2.3a), any instantaneous change in the COG frequency, caused by initial
compensation of a disturbance by the kinetic energy of the equivalent rotating
plant, would be represented by the swing equation as [13, 16]:
d Δ f COG 1
= ΔPmech − ΔPelec 2 10
dt 2πM COG
It then follows that the motion of the centers of inertia can be written as [17]:
1
f COI i = ΔPi − 2πDi Δ f COI i − Ptie
COI i ,COG , i = 1, 2, …, n 2 11
2πM COI i
where f COI i and Di are the COI frequency deviation for the area i following distur-
bance ΔPi and load damping parameter, respectively. The equivalent damping
coefficient D can be calculated using Prony analysis (refer to Section 7.2 for details
of Prony analysis). The last term in (2.11) has a physical interpretation of interest.
The net tie-line power exchange PtieCOI i ,COG can be rewritten in the alternative form
of (2.4). By knowing the virtual impedances, the term Ptie
COI i ,COG and (2.11) can be
evaluated. Dividing (2.11) by (2.10) gives
M COG ΔPi − 2πDΔ f COI i − PCOI i ,COG

tie
df COI i
= 2 12
df COG M COI i ΔPmech − ΔPelec
then
M COG ΔPi − 2πDΔ f COI i − PCOI i ,COG

tie
Δ f COI i = Δ f COG 2 13
M COI i ΔPmech − ΔPelec
∂ f COI i
∂ f COG
df COI i
ΔPi in (2.13) is nonzero only for the disturbed area. Moreover, the term
df COG
describes the sensitivity of the COI frequency to the COG frequency. Multiplying
df COI i
by the COG frequency variations for a given energy mismatch ΔfCOG gives
df COG
area’s i COI frequency changes.
Rearranging terms in (2.13) results in:
M COG ΔPi − PCOI i ,COG

tie
M COG 2πDΔ f COI i
Δ f COI i + Δ f COG = Δ f COG
M COI i ΔPmech − ΔPelec M COI i ΔPmech − ΔPelec
2 14
or, equivalently,
M COG ΔPi − PCOI i ,COG

tie

M COG 2πD
1+ Δ f COG
∂ f COI i
∂ f COG
Equation (2.15), which represents the area i-th COI frequency based on COG
frequency (2.10), can be employed to analyze the propagation of disturbances to
neighboring systems through tie-lines. It should be noted that as the overall rotat-
ing masses and loads are considered to be aggregated in the COG, the term
ΔPmech − ΔPelec in (2.15) represents the magnitude of load disturbance.
Following the same procedure as in (2.15), each bus frequency can be estimated
based on the associated local COI frequency. For this purpose, the dynamic behav-
ior of each local bus frequency fj is expressed as
1
fj= ΔP − 2πD j Δ f j − Ptie
COI i ,j 2 16
2πM j
Dividing (2.16) by (2.11) and following the same procedure as in (2.15) gives:
M COI i ΔPi − PCOI i ,COG

tie
Mj ΔP j − Ptie
ij
Δf j = Δ f COI i 2 17
M COI i 2πD
1+ Δ f COI i
M j ΔP j − Ptie
COI i ,j
While (2.15) ties each COI frequency to the COG frequency dynamics, (2.17)
connects each bus frequency (local frequency) to the COI dynamics. This means
that the frequency dynamics of each bus can be tied to the COG frequency using
simple algebraic equations. In the other words, (2.15) and (2.17) illustrate how the
frequency dynamics propagate in the system.
2.3.1.5 Simulation Results

Three test systems are employed for illustrating the capability of the modeling
procedures in this chapter: (i) a simple two-area, four-machine test system,
(ii) a 16-machine, five-area 68-bus test model of the New York New England
(NYNE) test system, and (iii) the IEEE 50-machine test system.
Two-Area System
Figure 2.3 shows a single-line diagram of this system. All the generating units are
modeled with sixth-order synchronous machine models with excitation systems
[18]. Each generator is equipped with a simple turbine-governor model of
Figure 2.4. The total system load is 2734 MW; the disturbance considered is the
shedding of 1400 MW load, i.e. 14 pu at bus 14 in Area 2.
Area 1 Area 2
1 10 20 3 101 13 120 11
Gen3
Gen1
110
Gen2 Gen4
4
2 12
14
Figure 2.3 Single-line diagram of two-area system.

ωref Porder Tmax
Speed Pmech
1 1 1 + ST3 1 + ST4
r 1 + STs 1 + STc 1 + ST5
Figure 2.4 Simple Turbine Governor model (r = 25, Ts = 0.1, T3 = 0.0, T4 = 1.25, T5 = 5.0,
Tc = 0.5, Tmax = 1.0).
Using (2.15) and (2.17), the frequency deviations at bus 1 in Area 1 can be
expressed in terms of the COG frequency as
1 06
Δf1 = Δ f COG 2 18
1 + 0 06Δ f COI 1
Similarly, for bus 12 in Area 2, one can find that
1 78
Δ f 12 = Δ f COG 2 19
1 + 0 93Δ f COI 2
The significant difference between (2.18) and (2.19) stems from the fact that the
fault takes place in Area 1. Figure 2.5 shows the transient behavior of synchronous
machines, obtained by a conventional washout filter, a frequency divider [8], and
the frequency propagation approach in (2.15) and (2.17). For comparison pur-
poses, the time constant of the washout filter Tf is set to the default value 0.01 [8].
Additional insight into the ability of the method to characterize the slow system
dynamics can be obtained from the modal analysis of the reduced-order COG rep-
resentation. Table 2.1 compares the frequency of the two slowest modes of the full
1.02
f1 [pu]
1.01
1
0 5 10 15
1.02 Actual value

Washout filter
f12 [pu]
Frequency propagation
1.01 Frequency divider
1
0 5 10 15
Time [s]
Figure 2.5 Frequency responses of two-area system following the loss of 14 pu of load at
bus 14.
Table 2.1 Comparison of two slowest modes for original model and equivalent model.
Original model frequency Equivalent model frequency Error (%)
0.1936 0.1956 1.4

0.6802 0.7088 4.2
model with those of the COG-based equivalent model. As shown, the maximum
error is less than 5%. This example suggests that simple algebraic equations of
(2.15) and (2.17) can estimate local frequencies from the COG-based model of
Figure 2.2 with high accuracy.
NYNE Test System

The NYNE test system, including five geographical regions, is used to test the
ability of coherency-based frequency propagation analysis to characterize power
and frequency deviations following large system perturbations. A diagram of the
NYNE system, including areas and their interconnections, is shown in
Figure 2.6. Detailed generating unit models and their controllers were included
in the simulations. Generators 1–12 are equipped with a fourth-order type II power
system stabilizer, tuned to provide sufficient damping [19, 20].
14 Area 3 Area 1
66 8
41 40 48 47 60
1
25 26 29
53
28
10 2 61
27
62 9
30
31 3
1
63 18 17
24
46 38 32 16
33 15 22
Area 5 9 21
34 4
14 19 58
49 35 6
42 36 5 20
8 6 12 56 23
67 64 57 4
15 45 12
51 5
59
50 44 37 11 13 7
7 54
2
65 10
52 43 13
39
55
68
3
Area 4 16 Area 2
Figure 2.6 Single line diagram of the NYNE system showing coherent areas and their
interconnections.
Numerical results comparing the estimated frequency are presented below. Four
contingency scenarios (CSs) are considered in this analysis:
•• CS1:
CS2:
Loss of 6000 MW at bus 37 in Area 2
Three-phase fault at line 31–62, cleared in 12 cycles
•• CS3:
CS4:
Three-phase fault at line 52–68 followed by the outage of generator G16
Outage of line 50–52 resulting in system instability
Figures 2.7 and 2.8 illustrate the frequency and tie-line power dynamics at key
system locations for contingency scenarios CS1 and CS2. The frequency of buses
59 (Area 1), 1, 63 (Area 2), and 47 (Area 3) and the net power flows to Areas 2 and 4
are selected for analysis.
In each case, the exact frequency behavior, obtained by a time-domain (T-D)
simulation program, is used as a basis for evaluating the accuracy of the estimation
methods. In the studies described below, (2.4) is employed to estimate the net
power flow to areas in the equivalent system; the net power flow to Area 1 and
Area 4 in the physical system is calculated by
Ptie,1 = Ptie tie tie

1,2 + P 1,27 + P 9,8 2 20
and
Ptie,4 = Ptie tie

50,51 + P 52,41 2 21
where Ptie
k,m represents the power flow across the tie-line connecting bus k, located
in the study area, to bus m, located in neighboring areas.
Careful analysis of the numerical results in Figure 2.8 reveals that the frequency
dynamics of bus 63 in response to contingency CS2 shows some discrepancy with
the actual frequency response. Such discrepancies may result from system topol-
ogy changing conditions modifying the estimated virtual reactances. In the basic
framework, the inputs to the COG system are selected as bus voltage magnitudes,
angles, and frequencies, obtained using a T-D simulation program. As a result, the
COG dynamics could be updated or renewed using various strategies. In the below
given results, three main strategies (STs) to compute the virtual reactances in (2.8)
and (2.9) are compared:
•• ST1: Calculations based on pre-fault values

ST2: Calculations based on post-fault steady-state values
• ST3: Calculations based on bus phase angles collected at a rate of 2 cycles/s
Table 2.2 compares the accuracy of the COG-based model in compliance with
the aforementioned strategies for estimating first-swing oscillations. As shown
in the table, by recalculating the virtual reactances, the accuracy of the proposed
framework can be enhanced even for system topology changing disturbances.
(a)
1.015
1.01
f1 [pu]
Actual value
Washout filter
1.005
Frequency divider
1
0 5 10 15
1.015
1.01
f59 [pu]
1.005
1
0 5 10 15
Time [s]
(b)
1.012
1.01
1.008
f59 [pu]
1.006 COI Response

Actual value
1.004 Frequency propagation
Washout filter
1
0 1 2 3 4 5 6
Time [s]
(c)
–4 Net power flow to Area 4 in the system
Net power flow to Area 4 in the equivalent system
Ptie [pu]
–4.5
–5
0 1 2 3 4 5 6 7 8 9 10
Time [s]
Figure 2.7 System response for contingency scenario CS1; (a) frequency responses of
NYNE system, and (b) detail of frequency response showing the COI frequency; (c) net power
flows to Area 4.
Figure 2.9 compares the frequency responses for the aforementioned strategies
with those of actual response and frequency divider approach [8]. The results con-
firm the high accuracy of the frequency propagation approach to estimate both the
transient and mid-term dynamics.
(a)
1.01
f63 [pu]
0.99
0 5 10 15
1.002
Actual frequency
f47 [pu]
1 Washout filter
0.998
0 5 10 15
Time [s]
(b)
1.5
1
Ptie [pu]
0.5
Net power flow to Area 1 in the system
0 Net power flow to Area 1 in the equivalent system
–0.5
0 1 2 3 4 5 6 7 8 9 10
Time [s]
Figure 2.8 (a) Frequency responses of NYNE system for contingency scenario CS2, (b) net
power flows to Area 3.
Table 2.2 Comparison of first swing amplitudes.
Strategy Actual value Frequency propagation Error (%)
ST1 1.0483 1.0101 3.64

ST2 1.0483 1.0322 1.54
ST3 1.0483 1.0446 0.35
To further verify the theoretical basis behind the computation of fictitious reac-
tances, Table 2.3 compares the ratio of the calculated impedances for different volt-
age magnitudes, including 1 pu voltage and the exact voltage magnitude after one,
two, and three cycles, following the inception of fault. The results suggest that volt-
age magnitudes may not significantly affect the ratio of reactances.
Numerical experience shows that during a fault, when bus voltage magnitudes
deviate from nominal values, the COG-based method can still exhibit satisfactory
performance. This could be justified by the fact that the COG-based equivalent
1.005
1
f63 [pu]
0.995 Actual Value

ST3
0.99 ST1
ST2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

Time [s]
Figure 2.9 Frequency responses of different strategies for contingency scenario CS2.
Table 2.3 Sensitivity of virtual reactances to voltage variations.
X tie
COI X tie X tie X tie
2 ,COG COI 3 ,COG COI 4 ,COG COI 5 ,COG
X tie
COI X tie X tie X tie
1 ,COG COI 1 ,COG COI 1 ,COG COI 1 ,COG
v = 1 pu 1.01 1.32 2.18 0.73

Snapshot 1 1.03 1.31 2.13 0.74
Snapshot 2 1.01 1.29 2.23 0.73
Snapshot 3 1.01 1.29 2.2 0.73
model mitigates the response of generators to voltage deviations, reflected as inter-

unit synchronizing oscillations between generators.
Modal analysis was further conducted to evaluate the method’s ability to assess
the propagation of low-frequency oscillations. The NYNE has four inter-area
modes of concern. Table 2.4 compares the eigenvalues of the full system and
the COG representation in Figure 2.2. The reduced model includes five areas,
five interconnecting lines, and their associated buses. As shown in Table 2.4,
Table 2.4 Comparison of modal analysis results.
Original system model COG-based equivalent model
No. Eigenvalue Frequency Eigenvalue Frequency
1 −0.7018 ± 1.9710j 0.3137 −0.7356 ± 1.7514j 0.2787

2 −0.1184 ± 3.2666j 0.5199 −0.1004 ± 3.0123j 0.4794
3 −0.3590 ± 3.7108j 0.5906 −0.3913 ± 4.0108j 0.6383
4 −0.1657 ± 4.8915j 0.7785 −0.1535 ± 4.6454j 0.7393
1
f59 [pu]
0.99
0.98
0 5 10 15
1
Actual frequency
f63 [pu]
0.99 Washout filter
0.98
0 5 10 15
1.05
V27 [pu]
0.95
0 5 10 15
1
V63 [pu]
0.9
0.8
0 5 10 15
Time [s]
Figure 2.10 Frequency and voltage responses of 16-machine for contingency

scenario CS3.
the COG-based equivalent model provides consistent results with eigen analysis
results of the original (unreduced) system. This result also suggests that the
approach can be effectively applied when characterizing long-term system
behavior.
To further confirm the validity of the model, Figures 2.10 and 2.11 show the
performance of the method for SC3 and SC4 resulting in system instability. The
method is seen to accurately characterize system instability for both voltage and
frequency signals.
During transients associated with cascading events, the stiffness of the COG-
based method mitigates the variations of the center of angles and bus voltages. This
is shown in Figure 2.11 which demonstrates that while such variations degrade the
performance and accuracy of the proposed method to characterize short-term
behavior, the overall trend of the unstable frequency oscillation is accurately
visualized.
1.01
1.005
1
f [pu]
0.995
0.99
0.985
0 0.2 0.4 0.6 0.8 1
Time [s]
1.002
Actual Local Frequency
1 Frequency Propagation
0.998
f [pu]
0.996
0.994
0.992
0.99
0 0.2 0.4 0.6 0.8 1
Time [s]
Figure 2.11 Frequency responses of 16-machine for contingency scenario CS4.
Finally, the effectiveness of the COG-based method is assessed to estimate fre-

quency dynamics of interest in control and protection applications, including
frequency nadir, and the rate of change of frequency (RoCoF) [17]. Table 2.5 com-
pares the frequency nadir and RoCoF for the washout filter and the frequency
propagation algorithm. This comparison is provided for two different scenarios:
Table 2.5 Comparison of frequency dynamics using the washout filter and proposed
method.
Nadir Nadir Error RoCoF RoCoF Error

No. Bus type (FPA) (WF) (%) (FPA) (WF) (%)
Loss of 1 Dynamic 1.022 1.020 0.2 0.022 0.0242 0.1

load load
8 Dynamic 1.018 1.017 0.1 0.018 0.0234 0.3
load
51 Dynamic 1.025 1/023 0.2 0.025 0.0350 0.4
load
53 Generator 1.017 1.016 0.1 0.017 0.0221 0.3
68 Generator 1.026 1.026 0.0 0.026 0.0286 0.1
Loss of 1 Dynamic 0.971 0.970 0.1 0.029 0.0290 0.0

generation load
8 Dynamic 0.968 0.968 0.0 0.032 0.0512 0.6
load
51 Dynamic 0.97 0.970 0.0 0.030 0.0450 0.5
load
53 Generator 0.976 0.976 0.0 0.024 0.0312 0.3
68 Generator 0.979 0.982 0.3 0.021 0.0252 0.2
FPA: Frequency propagation algorithm; WF: washout filter.
2470 MW load shedding at bus 52, and tripping of 4000 MW generation at bus 68.
Numerical results demonstrate the accuracy of the COG-based method to capture
slow system motion.
IEEE 50-Machine Test System

The 50-machine power network consists of 145 buses, 453 transmission lines,
52 transformers, and 60 loads; the total load is 2.83 GW and is used to investigate
the performance of the method for more complex systems representations.
In the first scenario, the result obtained by applying a three-phase fault on bus
67 is considered. The fault is assumed to be cleared in 10 cycles by primary pro-
tection. Results obtained by the COG-based method are compared with those of
using T-D simulation. A close examination of simulation results in Figure 2.12
shows that there is a good agreement with the results obtained by the frequency
propagation-based method.
Further, the effectiveness of the COG-based method in the presence of inertia-
less generating units is investigated. For this purpose, 10% of the system load is
assumed to be supplied by wind farms. Figure 2.13 demonstrates the efficiency
1.002 Frequency propagation

T-D simulation
f93 [pu]
1
0.998
0.996
0 5 10 15
1.002
f105 [pu]
1
0.998
0 5 10 15
1.005
f135 [pu]
0.995
0 5 10 15
Time [s]
Figure 2.12 Frequency responses of 50-machine system following a three-phase fault on

lines 67–119.
of the frequency propagation approach by comparing the estimated frequency

dynamics and T-D simulation results.
Analysis of frequency studies in Figures 2.13 indicates that unrealistic frequency
behavior associated with variations of the active power of variable generations is
mitigated using the COG-based estimation method. To further investigate the
accuracy of the modeling, the estimated frequency is numerically compared with
1.002
f60 [pu]
T-D simulation
1
0.998
0.996
0 5 10 15
1.005
f93 [pu]
0.995
0 5 10 15
1.005
f105 [pu]
0.995
0 5 10 15
Time [s]
Figure 2.13 Frequency responses of 50-machine system including wind power facing
three-phase fault.
Table 2.6 Computational burden of different bus frequency estimators.
Scenario T-D simulation Frequency propagation
With wind 535 29.34

Without wind 25 9
the actual frequency obtained using T-D simulation. In this analysis, the best fit-
ness value (BFV), a measure of modeling accuracy, is calculated as
y−y
BFV = 1 − 2 22
y−y
where y is the frequency response obtained from the T-D simulation, ŷ is the fre-
quency response obtained from equations (2.15) and (2.17), and y is the mean of y.
Simulation results reveal that while the BFV for the COG-based equivalent model
is more than 85% in Figures 2.7–2.12, it is less than 65% for the system considering
wind farms. This illustrates the necessity of modifying the COG-based model
through stochastic approaches which are discussed in the next section.
It may also be noted that the time required to estimate local frequencies in the
frequency propagation algorithm in the presence of wind power is about 30 sec-
onds, while it is more than 500 seconds for close T-D simulation. Table 2.6 com-
pares the computational burden of frequency propagation paradigm and close T-
D simulation. The significant difference between the frequency propagation tech-
nique and T-D simulation can be justified by noting that the number of state vari-
ables in the COG-model based estimation is much less than that for T-D simulation.
Finally it should be emphasized that the results of Figures 2.12 and 2.13 suggest
that the tie-line power flows and the local frequencies converge to steady state in a
short-time horizon. This finding justifies the fact that the COG-based method
could be used to rapidly and accurately determine the post-fault stable equilibrium
points following perturbations as well as to assess the effect of remedial control
schemes on frequency dynamics.
2.3.2 An Enhanced COG-Based Model

2.3.2.1 Key Concept
The concept of COG was introduced in Section 2.3.1 to study long-term power-
frequency transients following large perturbations. The COG model shows high
efficiency to capture power and frequency oscillations, and thus it could be
employed with high reliability in small-signal, voltage, and frequency stability
studies. However, simulation results in the presence of variable generations
suggest the need for modifying the deterministic COG-based model. The theory
developed in this section extends the considerations of the effect of variable gen-
erations and the associated uncertainties on the COG model formulation. Without
the loss of generality, DGs and ESSs included in a multi-MGs (MMGs) are assumed
as variable generations to realize penetrated power grids.
Using the COG concept, the given MGs integrated-power grid can be repre-
sented by a simplified equivalent model of Figure 2.14, obtained by solving a sim-
ple minimization problem of (2.8) and (2.9). Two limitations are inherent for the
representation of (2.8) and (2.9) to assess the impact of MMGs integration:
(i) uncertainties in the representation of MMGs cannot be properly incorporated
and (ii) different operating modes, i.e. grid-connected and islanded modes, cannot
be properly represented [9, 21].
60 Hz Area 2
Inertial
response
Nadir
Area 1 δCOI2,MCOI2
tie ,
PCOI 1 COG tie ,
PCOI 2 COG
δCOI1,MCOI1
COG
tie ,
PCOI
Area k j COG δCOG,MCOG
tie ,
PCOI n COG
δCOIk,MCOIk
conv. + MESS + MMGs

δCOIn,MCOIn
COG considering MMGs and ESSs.
Incorporating the dynamics of MMGs into the equivalent model representations

requires a transformation of the simple deterministic COG formulation of (2.8)
and (2.9) to a complex uncertain optimization problem. In what follows, the
use of the COG framework in the case of an MG-integrated power system with
several MMGs is reviewed in the light of a robust approach to study transient
power-frequency behavior.
To introduce the proposed formulation, assume that several MMGs are
integrated with ESSs to provide frequency control during emergency conditions
as well as to improve the grid’s resiliency. Taking the contribution of MMGs
into account, the balance constraint (2.3b) in the penetrated grid can be
represented as
COI i ,COG − ΔP i
Ptie
n n MMG
T tie
COI i ,COG = =0 2 23
i=1 i=1
ωCOG
where the virtual tie-line power flow, Ptie

COI i ,COG, has the form of (2.4). The problem
of interest again is to determine the fictitious reactance X tie
COI i ,COG in (2.23). For this
purpose, a four-step procedure based on a data-driven goal programming
approach is proposed as follows:
Step 0: Insert (2.3b) into (2.23), as

n
1 V COI i V COG
sin δCOI i − δCOG − ΔPMMG =0 2 24
ω
i = 1 COG X tie
COI i ,COG
i
Step 1: Reformulate (2.24) as a minimization problem
1 V COI i V COG
min sin δCOI i − δCOG − ΔPMMG
ω
i = 1 COG X tie
COI ,COG
i
2 25
i
st 0< X tie
COI i ,COG ≤1
Step 2: Recast (2.25) in the linear form
min ξ− − ξ +
X tie
COI ,COG
i
n n
1 V COI i V COG
st sin δCOI i − δCOG + ξ − − ξ + = ΔPMMG
ω
i = 1 COG X tie
COI ,COG i=1
i
i
−
ξ ,ξ +
≥0
COI ,COG ≤ 1
0 < X tie
i
2 26
where ξ− and ξ+ define deviations from the target value, ξ, in the negative and
positive directions. Goal programming assigns a goal value to each of the objective
measures, i.e. zero to (2.25). Undesired deviations from the target value ξ are
then minimized using (2.26). From linear programming theory, at least one of
the ξ− and ξ+ must be zero [22].
Step 3: Generate scenarios for ξ and ΔPMMG

i
Step 4: Redefine (2.26), taking the expected value of ξ into account, as
S
min
tie
ps ξs− − ξs+ 2 27
X COI s=1
i ,COG
s.t.
n n
1 V COI i V COG
sin δCOI i − δCOG + ξs− − ξs+ = ΔPMMG
ω
i = 1 COG X tie
COI i ,COG i=1
i,s
−
ξs , ξs+ ≥ 0
0 < X tie
COI i ,COG ≤1
2 28
where ps denotes the probability of each scenario and s denotes the scenario.

The efficiency of the goal programming-based equivalent model of (2.27) and
(2.28) is investigated through a comparison of frequency response features as well
as computational time with those of the GA-based equivalent model (2.8) and (2.9)
and T-D simulation results for two area power system model of Figure 2.3. For this
purpose, ξ− and ξ+ are set to 0 and 0.1, respectively. Table 2.7 demonstrates the
high efficiency of the equivalent model to estimate the frequency dynamics of
interest.
Also of interest, a simple interval approach is utilized to assess the robustness of
the method. The interval approach assumes that the uncertain parameter takes
value in a range. The aim is to find the lower and upper bounds of the objective
function in compliance with the interval [23]. It could be reinterpreted as the prob-
abilistic modeling with a uniform probability density function (PDF). Table 2.8
reports the lower and upper bounds of the objective function for 5% uncertainty
in the power of MMG. The results confirm the robustness feature of the goal pro-
gramming-based model which is a key point when dealing with highly penetrated
power grids.
To calculate the reported results of Tables 2.7 and 2.8, 10 random scenarios
of equal probability, i.e. Ps = 0.1 in (2.27), are generated. Minimization of
Table 2.7 Computational burden of different bus frequency estimators.
Scenario Frequency nadir RoCoF Computational time
T-D simulation 1.021 0.61 23

Equivalent model (2.8) and (2.9) 1.016 0.58 2.1
Stochastic model (2.27) and (2.28) 1.009 0.55 9.14
Table 2.8 Uncertainty analysis of the equivalencing.
Scenario Upper bound Lower bound
Equivalent model (2.8) and (2.9) 0.5712 0.112

Stochastic model (2.27) and (2.28) 0.461 0.432
(2.27) subject to (2.28) reveals that the reduced equivalent model can be
characterized by
X tie
COI 1 ,COG = 0 2134; X tie
COI 2 ,COG = 0 3172 2 29
2.3.3 Generalized Equivalent Model

2.3.3.1 Basic Logic
Not only the deterministic COG-based model of (2.8) and (2.9) and stochastic
COG-based model of (2.27) and (2.28) in the previous sections cause nonunique
results but also the time consumption feature affects the effectiveness of the mod-
els. Therefore, in this section, the so-far models would be modified to have an effi-
cient analytical model. Figure 2.15 schematically demonstrates the logic behind
the modified equivalencing approach. The lines between areas in Figure 2.15
are considered as the auxiliary reactances which in turn simplify the calculation
of fictitious reactance of interest in (2.6).
Indeed, Figure 2.15 is a modified version of Figure 2.2 where a combination of
star and end to end topologies are considered in the process of calculation of the
fictitious reactances. The logic behind the modeling procedure is simple: “while the
summation of applied torques to COG is zero, the summation of torques in each area
d Δω
is non-zero.” According to (2.1) and considering RoCoF = , the summation
dt
of torques in each area could be explained by
Area 2
Area 1 δCOI2,MCOI2
tie ,
PCOI1 COG tie ,
PCOI 2 COG
δCOI1,MCOI1
tie ,
COG
PCOI
Area k j COG δCOG,MCOG
tie ,
PCOIn COG
Area n
δCOIn ,MCOIn
δCOIk ,MCOIk
modified COG concept.
M × RoCoF = T m − T e 2 30
and hence, (2.6) would be rewritten for area i in Figure 2.15 as
k
T tie
COI i ,COG + T tie
COI i ,COI j = M × RoCoF 2 31
j=1
Considering (2.6) and (2.31), a set of n equations with n undefined fictitious reac-
tances would be derived.
2.3.3.2 Simulation and Results

The effectiveness of Figure 2.15 to represent the power grid equivalent model is
investigated on two test systems. In the first scenario, the outage of generator num-
ber 4 is considered in the two-area power grid of Figure 2.3. Figure 2.16 shows the
equivalent model of the system in the introduced framework.
Figure 2.16 Equivalent model of

two-area system.
X12 Area 2
δCOI2 , MCOI2
Area 1
δCOI1 , MCOI1
XCOI2 , OCOI
T1tie,COG T2tie,COG
XCOI1 ,COG
COG
δCOG , HCOG
According to (2.6) and (2.31), one could write:
0 37X COI 1 COG − 0 58X COI 2 COG = 0

0 13X COI 1 COG + 0 73X 12 = 0 45 2 32
0 51X COI 2 COG − 0 42X 12 = 0 36
Figure 2.17 compares the results of T-D simulation with those of obtained
using (2.32).
1
f1 [p.u.]
0.98
0 5 10 15
1
f2 [p.u.]
0.98
0 5 10 15
1 T-D simulation
f3 [p.u.]
COG-Based method
0.98
0.96
0 5 10 15
Time [s]
Figure 2.17 Frequency responses of two-area system for the outage of generator 4.
1.005
f1 [p.u.]
0 5 10 15
1.005
T-D simulation
f2 [p.u.]
COG-Based method
1
0 5 10 15
1.005
f3 [p.u.]
0 5 10 15
Time [s]
Figure 2.18 Frequency responses of two-area system for tripping of load.
In another attempt, the efficiency of the generalized COG-based method is

investigated for the rejection of a load. Following the same procedure as
(2.32) gives:
0 18X COI 1 COG − 0 32X COI 2 COG = 0
0 72X COI 1 COG + 0 23X 12 = 0 15 2 33
0 08X COI 2 COG − 0 12X 12 = 0 07
which in turn gives rise to the dynamic behavior of Figure 2.18.
Finally, the effectiveness of the model is further justified using NYNE test
system of Figure 2.6 for four different scenarios:
•• Scenario
Scenario
1:
2:
Outage of generator 16 (Figure 2.19)
Three-phase fault on line 49–52 (Figure 2.20)
•• Scenario
Scenario
3:
4:
Outage of generator 15 (Figure 2.21)
Tripping of load at bus 37 (Figure 2.22)
2.4 MG Equivalent Model
Developing MG equivalent models is of high importance when dealing with MGs-

integrated power system stability analysis. The equivalent model should consider
uncertainties related to output power fluctuations and different operating modes
of the sources. This section begins with a subsection devoted to the modeling of
DGs in grid-forming mode (islanded mode) and will be continued by deriving
1
f1 [p.u.]
0.995
0 1 2 3 4 5 6 7 8
1
f5 [p.u.]
0.995
0 1 2 3 4 5 6 7 8
1
f12 [p.u.]
T-D simulation
COG-Based method
0.995
0 1 2 3 4 5 6 7 8
Time [s]
Figure 2.19 Frequency responses of NYNE system for scenario 1.

f1 [p.u.]
0.999
0 1 2 3 4 5 6 7
1.001
f7 [p.u.]
1 COG-Based method
T-D simulation
0.999
0 1 2 3 4 5 6 7
f13 [p.u.]
1
0.999
0.998
0 1 2 3 4 5 6 7
Time [s]
an experimental-based equivalent model of MG in grid-supporting mode (grid-

connected mode). University of Kurdistan MG (UOK-MG), represented in
Figure 2.23, is employed to validate the results in this section.
2.4.1 Islanded Mode

2.4.1.1 Synchronous-Based DG
Synchronous-based DG, i.e. Genset, includes an internal combustion (IC) engine
driven by explosive combustion of gasoline and a wound field synchronous
machine (SM) [24–26]. In the Genset modeling, there are two main components:
the IC engine that converts the fuel to mechanical power and regulates the
1
f1 [p.u.]
0.998
0.996
0.994
0 5 10 15
1
f9 [p.u.]
0.998
0.996
0.994
0 5 10 15
1
f13 [p.u.]
0.998 COG-Based method

T-D simulation
0.996
0.994
0 5 10 15
Time [s]
1.015
fCOI, 1 [p.u.]
1.01
1.005
1
0 1 2 3 4 5 6 7
1.015
fCOI, 2 [p.u.]
1.01
1.005
1
0 1 2 3 4 5 6 7
1.015
fCOI, 3 [p.u.]
1.01
COG-Based method
1.005 T-D simulation
1
0 1 2 3 4 5 6 7
Time [s]
frequency of Genset and the exciter that regulates the terminal voltage of the
Genset [27–30]. Details of the Genset controller are shown in Figure 2.24 [26].
Fuel command FCMD is the output of the fuel controller Figure 2.25. Using this
framework, the measured speed of the synchronous machine compares with the
reference signal to produce an error signal. The error signal is fed into a propor-
tional-integral (PI) controller to produce the torque signal [27, 28, 30].
Main Grid Static Switch
SS
SS SS
AC load
SS SS
ESS
DC bus SM
IC
SS SS SS
Controllable AC load
C
C
D
D
C
C
D
+
DC Load
–
WT ESS PV
Figure 2.23 Three-phase schematic representation of the UOK-MG; WT, wind turbine; PV,
photovoltaic; SS, static switch; IC, internal combustion; SM, synchronous machine.
The limiter often implements in the block diagram of Figure 2.25 to avoid unre-
alistic commands during large load transients. The resultant torque then converts
to the fuel command signal using the torque to fuel conversion ratio Ktf. The output
fuel command signal of Figure 2.25 then applies to the simplified IC engine model
of Figure 2.26 to produce mechanical power. In this way, first, the fuel command
signal converts into the torque signal. Furthermore, engine combustion delay
affects the resultant torque. Finally, the torque converts to the mechanical output
power while losses are removed from the resulting value [27, 28, 30].
The exciter mechanism of Figure 2.27 is simply represented by a first-order
transfer function. Furthermore, the output DC field voltage multiplies by the elec-
trical speed, producing the AC voltage magnitude. The implemented limiter in
Figure 2.27 represents the saturation of the DC exciter field. The output is the field
voltage of SM [27, 28, 30].
The input signal to the exciter of Figure 2.27 comes from the voltage regulator of
Figure 2.28. It could be observed from Figure 2.28 that the error signal combines
with the feedforward value, which is the expected value required for nominal volt-
age at the terminal. Feedforwarding this value allows for quicker initial conver-
gence without the integrator having to wind up [27–31].
Vreq
Ecmd +–
Controller –
Verr
mQ
Vab, Vcd dq Vmeas

frame
Calculation of
Ia, Ic dq Control Qmeas
frame Quantities
SBASE from
Measurements Pmeas
VBASE
ωbase
Fcmd ++ – Preq
Controller + mP +
ωerr – +
ωmeas
1
1 + τ ωs
ωmeas
0
Pmax + PI
–
–Δω
+
Pmeas
+
Δω
Pmin –
+ PI
0
Figure 2.24 Genset controller scheme; MQ, slope of Q − V droop; MP, slope of P − ω droop;
KPI, integral control gain; KPP, proportional control gain; FCMD, command fuel signal; ECMD,
exciter control signal; Pmeas, real-time value of real power; Qmeas, real-time value of reactive
power; I , line current; Δω, allowable frequency change.
●
2.4.1.2 Genset Model Validation

Figure 2.29 shows the frequency dynamics of the Genset in response to turn on a
4 kW Static Load Bank (SLB) at 2 seconds. A comparison of the experimental and
simulation results reveals the high efficiency of the model.
ωref Pmeas
ωmeas ωerr Te Fcmd

– +
+ PI + Ktf
Figure 2.25 Fuel controller of Genset; Ktf, torque to fuel conversion ratio.
ω2meas Km
Fcmd Tm Pmec
ηthr Kcv Kfr Fcmd Time –
Delay ωmeas +
Figure 2.26 Simplified model of the IC engine; ηthr, thermal constant; Kcv, calorific value;
Kfr, fuel rate at rated speed; Km, mechanic losses constant.
Exciter Machine
Ecmd 1 Ef
Ecmd ωmeas
1 + τe s
Figure 2.27 Exciter model; τe, exciter machine time constant.
PI 1
Vref Kvi
S
Vmeas Verr + Ecmd

+– Kvp ++ +
Figure 2.28 Voltage regulation diagram; Kvi, integral controller gain; Kvp, proportional
controller gain.
2.4.1.3 Inverter-Based DG
Figure 2.30 shows the block diagram representation of an inverter-based DG [31].
Owing to the turn on/off effects of high-frequency switches, output signals
have a fundamental component and higher harmonics. However, a high rate of
(a)
4500
Power [W]
3000
Simulated
1500 Experimental
500
(b)
50
frequency [Hz]
49
48
0 5 10 15
Time [s]
Figure 2.29 Dynamics of Genset in response to turn on 4 kW load at 2 seconds,

(a) frequency response, (b) active power dynamics.
switching, in the range of 10 kHz, together with the implemented LC filters signif-
icantly mitigates harmonics [26, 29, 31]. Hence, the inverter, in the islanded oper-
ation mode, is represented as an ideal, balanced three-phase voltage source as
shown in Figure 2.31 [31]. Accordingly, the instantaneous three-phase bus vol-
tages can be expressed as
ab = mV DC cos ωt + θ
V ESS
bc = mV DC cos ωt + θ + 2π 3
V ESS 2 34
ca = mV DC cos ωt + θ − 2π 3
V ESS
In this model, the modulation index m in (2.34) is a scalar coefficient that con-
trols the voltage at the inverter terminals [26, 31]. Moreover, the same reasoning
that uses to represent the Genset controller in the islanded mode can be extended
to control inverter-based sources. However, the controller command signals are
PT CT
m LF3
Inverter
Controller θ(t) Firing System
+
VDC
CF3
Figure 2.30 Inverter-based DG block diagram; m, modulating index; θ(t), angle for the voltage at the inverter terminals.
Figure 2.31 Ideal source model.

Vab Vca
m
θ (t)
Vbc
inverter frequency and voltage modulation index (Figure 2.32) [32–34]. Further-
more, in Figure 2.32, conventional droop characteristics implement using fre-
quency and voltage droop coefficients [35].
2.4.1.4 Inverter-Based DG Model Validation

Figure 2.33 compares the ESS simulation results, obtained for switching on of
4 kW SLB at 1st second and switching off of 3 kW SLB at 5th seconds of the sim-
ulation, with those of experiments. Except for high-frequency noise, the simulated
waveforms for real and reactive powers follow the experiments with high accuracy.
The oscillation-free behavior of the simulated waveforms stems from the fact that
the source of power models as an ideal voltage source with the only fundamental
frequency. Furthermore, the offset between the simulated and experimental reac-
tive powers is due to the neglecting of the line and transformer models.
2.4.2 Grid-Connected Mode

2.4.2.1 Basic Logic
After successful derivation of the models for DGs in islanded mode, this
section discusses the dynamic equivalencing of a cluster of DGs and local loads,
visualized in the form of grid-connected MG, for power-frequency transient stud-
ies. The analysis framework can be summarized in four steps:
•• Step 1: Define an arbitrary number of operating points

Step 2: Different commitment of power sources in the MG to realize operating
points in Step 1
• Step 3: Metering of the injected power of MG to the host grid, i.e. PMG, for each
operating point in Step 2
• Step 4: Modal analysis (Prony analysis) of PMG, metered in Step 3, to calculate

inertia constant, i.e. M. Applying Prony analysis to PMG gives:
PI Vreq
Kvp
m + +
–
+ –
Kvi
S
Vab, Vcd Vmeas mQ

dq
frame
Calculation of
Ia, Ic
dq Control Qmeas
frame Quantities
SBASE from
Pmeas
Measurements
VBASE
ωbase
Preq
1 ++ –
θ (t) mP +
S +
0
Pmax + PI
–
–Δω
Pmeas +
+
Δω
Pmin –
+ PI
0
Figure 2.32 Inverter-based controller scheme.
f ai f 1
2 35
PMG s + ci PMG 1 ci
s+
ai ai
where a, c are constant parameters which may be defined by Prony analysis; refer
to Section 7.2 for details of Prony analysis. Equation (2.35) follows the same char-
acteristics as the classical swing equation of the form
(a)
6
Experimental
P [KW]
Simulated
4
(b)
0.5
Q [KVar]
0
–0.5
0 3 6 9 12
Time [s]
Figure 2.33 Experimental and simulated waveforms for real and reactive power output for
inverter-based DG operation in the UOK-MG. (a) active power, (b) reactive power.
f 1
2 36
PMG M i s + Di
1 ci
Hence, (2.35) with a reinterpretation of ≜ M; ≜ D could be employed to
ai ai
calculate MG inertia constant. A conceptual scheme of the adopted model is
shown in Figure 2.34.
Generally stated, Figure 2.34 suggests that MG response, from an upward point
of view, would be mapped onto the conventional synchronous generator. Accord-
ingly, the classical swing equation (2.36) represents the MG dynamics.
• Step 5: Return to Step 2, repeat until the prespecified number of operating points
are considered
• Step 6: Extract relationship, using curve fitting tools, between inertia constant
and committed DGs.
The above steps are summarized in the flowchart of Figure 2.35.
2.4.2.2 Model Validation

Experimental results of UOK-MG in grid-connected mode are utilized to demon-
strate the efficiency of the modeling procedure. Furthermore, a real-time digital
power system simulator is employed to visualize the host grid for testing purposes.
Figure 2.36 shows the dynamic responses of the MG, host grid, and constituent DG
for a sequence of events.
Diesel Generators
Advanced Energy Storage
Wind Turbines CLEAN SPARK
Solar Photovoltaics
Power grid
Figure 2.34 A conceptual scheme of the proposed MG dynamic equivalent model.
While event 1 refers to the importing of power to the MG, event 2 refers to the
islanding of the MG from the grid. It is evident that, between events 1 and 2, the
host grid provides the power of the MG with inertia constant. Following (2.36) for
the injected power of the host grid to the MG in the transient period, in
Figure 2.36, gives
ωMG 0 37
2 37
ΔPMG 0 27 + 0 63s
and in the steady state, (2.37) would be rewritten as
ωMG
= 1000 2 38
ΔPMG
where ωMGs and ΔPMGs are the frequency at the point of common coupling and the
injected power of the MG, respectively. Equations (2.37) and (2.38) suggest that the
MG mimics the behavior of a synchronous generator and a constant power source in
the transient and steady-state periods, respectively. This brings a first-order circuit
response to a pulse function in mind. Therefore, the same reasoning that uses to
assess the first-order circuit in response to a pulse function could be adopted to
approximate the effects of MG. In this way, one could represent the classical swing
equation of MG as
Figure 2.35 Flowchart representation of the

equivalencing approach. Start
Set number of iterations,

i,e. K; i = 1
Commit DGs in field

setup arbitrary; i = i + 1
Metering of field setup

injected power to grid
Calculation of M by
using Modal analysis
N
i = k?
Y
M = f(PGenset)
End
df t
M MG = Tm t − Te t u ζ − u υ 2 39
dt
where ζ and υ (ζ < υ) are DGs redispatching and islanding times, respectively.
Also of interest, event 2 refers to the islanding of the MG where droop charac-
teristic affects frequency response. The ratio of the grid power variation to the
steady-state frequency deviation of the islanded MG would be defined as the droop
characteristic. One could write this for Figure 2.36 as
0 − 1000
ΔP 5000 pu
R= = = 1 17 2 40
Δf 49 83 − 50 Hz
Finally, a relationship between inertia constant of (2.37) and the MG capacity
should be derived to facilitate assessing of impact of penetration level on the
grid dynamics. For this purpose, the MG inertia, obtained by (2.37) for several
1 2
4
2
P [KW]
1
0
–1
–2
Grid
50.2 MG
Wind
f [Hz]
50
49.8
49.6
0 6 12 18 24 30
t [s]
Figure 2.36 Experimental waveforms for real power and frequency output for MG, wind
turbine, and host grid.
operating points, maybe plotted against the ratio of the Genset to the MG capacity
Sn in Figure 2.37. In Figure 2.37, Sn, is defined as
0.8
0.7
0.6
0.5
H [s]
0.4
0.3
0.2
0.1
0.07 0.12 0.17 0.22 0.27 0.32 0.37 0.42 0.47
Sn [pu]
Figure 2.37 Relationship between MG inertia and ratio of Genset to MG capacity.

References 67
PGenset PGenset
Sn = = 2 41
PMG PGenset + PESS + PPV + PWT
Actually, Sn in (2.41) changes by the recommitment of the constituent Gensets.
For this purpose, two Gensets with the rated capacities of 5 and 10 kW and with,
respectively, 0.15-second and 0.21-second inertia constants are employed. To fur-
ther generate data sets for model derivation, an analog simulator, with inertia con-
stant of 0.1 seconds, is utilized. The figure reveals that there is a direct relationship
of form
H MG = 1 6109Sn − 0 0644 2 42
with root square (R2) of 0.954.
2.5 Summary
This chapter deals with the dynamic equivalencing of MGs-integrated power sys-
tems. Several methods are introduced to model the host grid and the distribution
network. The host grid modeling relies on the notion of COG dynamics in mechan-
ics to estimate local and global frequency behavior. Using this framework, the
power system dynamic behavior is represented by an equivalent model in which
the geographical areas interact with the COG through fictitious interconnectors
from which the inherent dynamics of power frequency transients are explained
using fundamental physics principles. The relationship between the frequency
of the COG and the motion of local centers of angle is determined and expressions
to compute local frequency deviations following major disturbances are derived.
Afterward, experimental-based models of constitute DGs of the UOK-MG in
islanded mode are derived. The basic description of the elements, as well as exper-
imental validation of the results, makes the finding of the chapter comprehensive.
A realistic model of the MG operating in grid-connected mode is derived to realize
a high penetrated grid in Chapters 3, 4, and 6.
References
1. Resende, F. and Peas Lopes, J. (2007). Development of dynamic equivalents for
microgrids using system identification theory. In: Power Tech, 2007 IEEE
Lausanne, 1033–1038. IEEE.
2. Bevrani, H., Ghosh, A., and Ledwich, G. (2010). Renewable energy sources and
frequency regulation: survey and new perspectives. IET Renewable Power
Generation 4 (5): 438–457.
3. Lasseter, R.H. (2011). Smart distribution: coupled microgrids. Proceedings of the

IEEE 99 (6): 1074–1082.
4. Bevrani, H., François, B., and Ise, T. (2017). Microgrid Dynamics and Control.
New York: Wiley.
5. Golpîra, H. and Bevrani, H. (2019). Microgrids impact on power system frequency
response. Energy Procedia 156: 417–424.
6. Golpîra, H., Atarodi, A., Amini, S. et al. (2020). Optimal energy storage system-
based virtual inertia placement: a frequency stability point of view. IEEE
Transactions on Power Systems 35 (6): 4824–4835. https://doi.org/10.1109/
TPWRS.2020.3000324.
7. Annakkage, U., Nair, N.K.C., Liang, Y. et al. (2012). Dynamic system equivalents: a
survey of available techniques. IEEE Transactions on Power Delivery 27 (1): 411–
420. https://doi.org/10.1109/TPWRD.2011.2167351.
8. Milano, F. and Ortega, A. (2017). Frequency divider. IEEE Transactions on Power
Systems 32 (2): 1493–1501. https://doi.org/10.1109/TPWRS.2016.2569563.
9. Golpîra, H. and Messina, A.R. (2018). A center-of-gravity-based approach to
estimate slow power and frequency variations. IEEE Transactions on Power Systems
33 (1): 1026–1035.
10. Benda, B.J., Riley, P.O., and Krebs, D.E. (1994). Biomechanical relationship
between center of gravity and center of pressure during standing. IEEE
Transactions on Rehabilitation Engineering 2 (1): 3–10. https://doi.org/10.1109/
86.296348.
11. Meriam, J.L. and Kraige, L.G. (2012). Engineering Mechanics: Dynamics. New
York: Wiley.
12. Ruiz-Vega, D., Messina, A.R., and Pavella, M. (2004). Online assessment and
control of transient oscillations damping. IEEE Transactions on Power Systems
19 (2): 1038–1047. https://doi.org/10.1109/TPWRS.2004.825909.
13. Kundur, P., Balu, N.J., and Lauby, M.G. (1994). Power System Stability and Control.
New York: McGraw-Hill.
14. Bevrani, H. (2009). Robust Power System Frequency Control. New York: Springer.
15. Morison, G.K., Gao, B., and Kundur, P. (1993). Voltage stability analysis using
static and dynamic approaches. IEEE Transactions on Power Systems 8 (3): 1159–
1171. https://doi.org/10.1109/59.260881.
16. Golpîra, H., Bevrani, H., and Naghshbandy, A.H. (2012). An approach for
coordinated automatic voltage regulator and power system stabiliser design in
large-scale interconnected power systems considering wind power penetration.
IET Generation, Transmission & Distribution 6 (1): 39–49. https://doi.org/10.1049/
iet-gtd.2011.0411.
17. Golpîra, H., Seifi, H., Messina, A.R., and Haghifam, M.R. (2016). Maximum
penetration level of micro-grids in large-scale power systems: frequency stability
References 69
viewpoint. IEEE Transactions on Power Systems 31 (6): 5163–5171. https://doi.org/

10.1109/TPWRS.2016.2538083.
18. Rogers, G. (2000). Power System Oscillations. Norwell, MA: Kluwer.
19. Rogers, G. (1999). Power system structure and oscillations. IEEE Computer
Applications in Power 12 (2): 14–21. https://doi.org/10.1109/67.755641.
20. Wilson, D.H., Hay, K., and Rogers, G.J. (2003). Dynamic model verification using a
continuous modal parameter estimator. In: Power Tech Conference Proceedings,
2003 IEEE Bologna, vol. 2, 6. IEEE.
21. Golpîra, H., Messina, A.R., and Bevrani, H. (2019). Emulation of virtual inertia to
accommodate higher penetration levels of distributed generation in power grids.
22. Wagner, H.M. (1969). Principles of operations research: with applications to
managerial decisions. In: Principles of Operations Research: With Applications to
Managerial Decisions. Englewood Cliffs, NJ: Prentice-Hall.
23. Aien, M., Hajebrahimi, A., and Fotuhi-Firuzabad, M. (2016). A comprehensive
review on uncertainty modeling techniques in power system studies. Renewable
and Sustainable Energy Reviews 57: 1077–1089.
24. Golpîra, H., Haghifam, M.R., and Seifi, H. (2015). Dynamic power system
equivalence considering distributed energy resources using Prony analysis.
International Transactions on Electrical Energy Systems 25 (8): 1539–1551.
25. Nikkhajoei, H. and Lasseter, R.H. (2009). Distributed generation interface to the
CERTS microgrid. IEEE Transactions on Power Delivery 24 (3): 1598–1608.
26. Golpîra, H. (2019). Bulk power system frequency stability assessment in presence
of microgrids. Electric Power Systems Research 174: 105863.
27. Krishnamurthy, S., Jahns, T., and Lasseter, R. (2008). The operation of diesel
gensets in a CERTS microgrid. In: Power and Energy Society General Meeting-
Conversion and Delivery of Electrical Energy in the 21st Century, 2008 IEEE,
1–8. IEEE.
28. Renjit, A., Illindala, M., Lasseter, R. et al. (2013). Modeling and control of a natural
gas generator set in the CERTS microgrid. In: Energy Conversion Congress and
Exposition (ECCE), 2013 IEEE, 1640–1646. IEEE.
29. Lemmon, M.D. (2009). Advanced Distribution and Control for Hybrid Intelligent
Power Systems. Final Project Report, University of Notre Dame: 1–23. Report
Number: UWM-11012009-03.
30. Krishnamurthy, S. and Lasseter, R. (2009). Control of Wound Field Synchronous
Machine Gensets for Operation in a CERTS Microgrid. Consortium for Electric
Reliability Technology Solutions (CERTS), Final Report Task 5. Value and
Technology Assessment to Enhance the Business Case for the CERTS Microgrid.
31. Lemmon, M. (2010). Comparison of Hardware Tests with SIMULINK Models of
UW Microgrid. Technical Report. Univ. Notre Dame, 2010.
32. Gao, Y. and Ai, Q. (2018). A distributed coordinated economic droop control
scheme for islanded AC microgrid considering communication system. Electric
Power Systems Research 160: 109–118.
33. Karami, M., Seifi, H., and Mohammadian, M. (2016). Seamless control scheme for
distributed energy resources in microgrids. IET Generation, Transmission &
Distribution 10 (11): 2756–2763.
34. Venkataramanan, G., Illindala, M., Houle, C., and Lasseter, R. (2002). Hardware
Development of a Laboratory-Scale Microgrid Phase 1: Single Inverter in Island
Mode Operation. NREL Report No. SR-560-32527 Golden, CO. National
Renewable Energy Laboratory.
35. Chang, C.-C., Gorinevsky, D., and Lall, S. (2014). Dynamical and voltage profile
stability of inverter-connected distributed power generation. IEEE Transactions on
Smart Grid 5 (4): 2093–2105.
71
Stability Assessment of Power Grids with High Microgrid

Penetration
The increasing complexity and operation of the electric power grids with high
levels of microgrids (MGs) penetration together with higher loading conditions
require the development of efficient control and analysis techniques with the abil-
ity to extract the relevant system behavior, identify abnormal behavior, and design
corrective measures. This need is due to the distributed nature of diverse renew-
able energy resources in the system and more complex dispatching strategies.
In this chapter, the analysis of the impact of high levels of MG penetration on
power system stability is discussed from various points of view such as frequency
stability, small-signal stability and voltage performance. Methods for interpreting
system dynamics in terms of simplified system representations are developed, and
criteria to determine maximum penetration levels are given using deterministic
and statistical sensitivity analyzes. Extensions to this framework to capture the
impact of large-scale wind and solar photovoltaic (PV) farms are discussed in later
sections of this book.
3.1 Introduction
3.1.1 Motivation
penetration of distributed energy resources (DERs) and distributed generation
(DG) sources, visualized through the MG concept. While low penetration of
MGs has a negligible influence on the stability of the host power system, high pen-
etration levels of MGs may significantly affect system stability and raise several
reliability concerns at the transmission level [1–4]. The increased demand for elec-
trical power, as well as environmental concerns and pressing need to reduce
dependence on fossil fuel, on the other hand, has forced many power system uti-
lities to set an ambitious target for the deployment of DERs/DGs.
72 3 Stability Assessment of Power Grids with High Microgrid Penetration
The integration of multiple MGs into the bulk power system requires the devel-
opment of new analysis tools and methodologies, to assess various aspects of
power system dynamic behavior and transmission limitations. These include
(i) determining the maximum levels of DGs/MGs generation in the system,
(ii) identifying the best system locations to deploy new DERs, and (iii) assessing
MGs’ control and dispatching strategies to improve system reliability and stability.
The first issue is addressed in this chapter. Discussion of issues two and three are
deferred until Chapters 4, 6, and 8.
Because modern DGs utilize inverter-based systems to interconnect with the
grid, their interactions with the bulk power system are different from those of con-
ventional synchronous generating units. One critical issue is the reduction of sys-
tem inertia, especially associated with inverter-based DERs [5–7]. Inertia
reduction resulting from significant penetration of utility-scale MGs and wind
and solar PV farms renders system dynamics faster and thus jeopardizes system
stability and reliability [8, 9].
In this chapter, a systematic methodology for the analysis of the impact of
distributed MGs on power system dynamic behavior is proposed. The analysis
of system dynamics is facilitated by the adoption of a simplified power system
model, in which frequency dynamics are represented explicitly through sensitivity
relationships.
First, a brief review of the subject is presented. Some important definitions asso-
ciated with the system frequency response are introduced.
3.1.2 Relations with Previous Literature

A general overview regarding the impact of low inertia on power system stability
and operation is provided in [7, 9]. In [10, 11], a trial-and-error based method-
ology to determine the maximum allowable penetration of wind generation is
presented. The transient stability assessment of power systems with high levels
of renewable generation is discussed in [12, 13]. On the area of wind penetration,
the scenario-based approach of [14] focuses on system inertia and primary
reserve values. A simplified frequency model is employed in [15] to study the fre-
quency behavior of an MG-integrated power system. Another scenario-based
approach, applied to a part of the Australian grid, is discussed in [16]. In this
respect, a framework for assessing renewable integration limits using a sce-
nario-based approach is discussed in [17]. Keyhani and Chatterjee [18], advo-
cated the use of a new automatic generation control (AGC) structure, which
tackles intermittency drawbacks associated with high penetration levels of
DGs. In yet another approach, a frequency response model is proposed in
[19], which can be used to evaluate AGC performance under wind power
uncertainty.
A common limitation of approaches that assess the impacts of high MGs/

DERs/DGs penetration levels on system stability is that energy penetration
levels are based on heuristic considerations. This requires the Transmission Sys-
tem Operators (TSOs) to define a set of expected operating conditions to be
assessed during the time-domain simulation procedure. Due to the highly uncer-
tain nature of MGs/DERs/DGs and the high dimensions of modern intercon-
nected power systems, the number of scenarios to be analyzed can become
daunting, and this presents a problem to determine a valid set of operating
points. As a consequence of this assumption, many recent research works quan-
tify system inertia for the MGs scenarios based on static unit commitment and
dispatch modeling studies. Other studies focus on inertia reduction rather than
the impact of the MG’s representations, and hence the effects of MG dynamics
and structure are neglected. This, in turn, causes rendering results inaccurate,
thereby offering limited insight into the impact of MGs on system dynamic
performance.
3.2 Frequency Stability Assessment
3.2.1 Background on Frequency Indices

Frequency response, as a measure of an interconnection’s ability to stabilize fre-
quency following a disturbance, can be assessed using a combination of physi-
cally motivated metrics. Typically, these include the rate of change of
frequency (RoCoF), the frequency response nadir, and the post-disturbance fre-
quency fluctuation over a given time window [20, 21]. These indices are typically
employed by TSOs for online monitoring and control systems, to initiate protec-
tion or remedial measures such as load shedding, or dispatch reserve power.
Accurate and timely estimation of system dynamics helps the associated entities
to procure ancillary systems to support the reliable operation of the power
grid [22].
3.2.1.1 Rate of Change of Frequency

The RoCoF is the time derivative of the frequency (df/dt). The initial RoCoF fol-
lowing a sudden torque imbalance is determined by the amount of stored rota-
tional kinetic energy on the system [23]. As discussed in more detail below, the
classical swing equation shows that the RoCoF is inversely proportional to the
rotational inertia. In typical applications, a RoCoF standard of 0.5–1 Hz/s is
utilized.
3.2.1.2 Frequency Nadir

The Frequency nadir is another metric of interest to examine the impact of high
levels of inverter-connected renewable on frequency fluctuations following loss
of generation events. Formally, the frequency nadir measures the maximum
post-contingency frequency deviation after a generator trip and is the point at
which frequency is arrested [24]. The Union for the Coordination of the Transmis-
sion of Electricity (UCTE) establishes 49.2 Hz as the minimum allowable post-
contingency frequency [20].
3.2.1.3 Delta Frequency Detection

Based on the North American Electric Reliability Corporation (NERC) resource sub-
committee, a frequency event is detected and captured if, during a 15-second roll-
ing time window, the frequency deviation exceeds a threshold. While the threshold
is system dependent, a threshold value of 300 mHz is adopted based on [25, 26].
The delta frequency detection criterion is a measure of the ability of the generating
units to respond to a disturbance over a given time interval.
3.2.2 Frequency Stability Assessment Under High MG

Penetration Levels
Conventional power system stability and dynamics studies are mature topics on
which a large amount of work has been done during the last two decades. There-
fore, it is advisable to represent renewable-integrated power systems stability and
control problems based on the well-studied conventional system behavior. The
analysis framework can be summarized in four steps:
• Step 1: Perform time domain (T-D) simulation of a base conventional system

with specific parameters.
• Step 2: Extract the frequency dynamics of interest, including RoCoF, frequency

nadir, and frequency evolution, from the T-D simulation results of Step 1.
• Step 3: Calculate the sensitivity of the MGs-integrated power grid dynamics with
respect to the base conventional system dynamics.
• Step 4: Represent the MGs-integrated power grid dynamics based on the base
conventional system behavior using sensitivity factors of Step 3. The above steps
are summarized in the flowchart in Figure 3.1.
3.2.3 Sensitivity Factors

3.2.3.1 Frequency Response
From Chapter 2, any change in the instantaneous system frequency, caused by a
system perturbation, can be presented in the COI formulation as [27, 28]:
Figure 3.1 Flowchart representation of the proposed

analytical approach to estimate frequency dynamics. Start
Time domain simulation

of original system
Extracting of RoCoF,
frequency nadir,
and frequency evolution
Sensitivity factors
calculation
fpenetrated = g(foriginal)
End
2M COI d2 θ
= ωCOI T mech − T elec 31
ω0 dt 2
in which the rotor angle, θ, is defined as
θ = ωCOI t − ω0 t + θ0 32
and, therefore,
d Δ f COI
2M COI = 2π f COI T mech − T elec 33
dt
Defining now
d Δ f COI
RoCoF = 34
dt
and
MiΔ f i
i
Δ f COI = 35
Mi
the COI frequency response (3.1) can be represented in the form
2M COI RoCoF COI = 2π f COI T mech − T elec 36

Using the framework shown in Figure 3.1, two power system models are con-
sidered next. The first model referred to as the base conventional system, charac-
terized by its inertia constant M1 and damping coefficient D1, represents the main
(original) system. The second model referred to as the reduced system, character-
ized by M2, D2, is the modified base model after the integration of MGs; in devel-
oping this model, it is assumed that M describes the aggregated representation of
different generators and load nodes [29].
The frequency response of the base system is then given by
d Δf1
2M 1 = ω1 T mech − T elec 37
dt
and for the MG-integrated power system
d Δf2
2M 2 = ω2 T mech − T elec 38
dt
Dividing (3.8) by (3.7) gives
M 2 RoCoF 2 2π f 2
= 39
M 1 RoCoF 1 2π f 1
Rearranging the terms in (3.9) yields
f 2 M1
RoCoF 2 = RoCoF 1 3 10
f 1 M2
Equation (3.10) describes the frequency response of the reduced system as a
function of the frequency response and equivalent inertia constants of the base
system.
Furthermore, introducing the effect of tie-line power, Ptie,ij, and the load dynam-
ics, Di, (3.3) can be rewritten in the following more useful form
1
fi = ΔPi − 2πDi f i − Ptie,ij ; i, j = 1, …, n, i j 3 11
2πM i j
for area i [9]. In (3.11), Ptie,ij and n are the tie-line power between areas i and j and
the number of areas, respectively.
According to the NERC guideline, the RoCoF is defined as the rate of frequency
deviation during 500 ms after the inception of a fault. Using this guideline and
assuming nominal frequency as 50 Hz, one can rewrite (3.11) as
f i − 50 1
= ΔPi − 2πDi f i − Ptie,ij 3 12
05 2πM i j
and hence,
05
f i = 50 + ΔPi − 2πDi f i − Ptie,ij 3 13
2πM i j
More precisely, Ptie,ij in (3.13) deviates from the pre-fault value and can be
explained using Taylor approximations, namely
Ptie,0 + ΔPtie = Ptie,0 + 2πPmax f i − f j cos δi0 − δ j0 δi − δ j − δi0 − δ j0

+ 2πPmax f i − f j cos δi0 − δ j0
2
− 2πPmax f i − f j sin δi0 − δ j0 δi − δ j − δi0 − δ j0
3 14
Generally, the term (δi − δj − δi0 − δj0) gets a small value and hence, one can
2
rewrite (3.14) as
Ptie,0 + ΔPtie = Ptie,0 + 2πPmax fi − f j cos δi0 − δ j0 Δδ 3 15
Extension of (3.11)–(3.15) to all the areas in an n-area interconnected power sys-

tem yields to the n − 1 equations, with n unknown frequencies, as
Fi f 1 , f 2 , …, f j + C i = 0; j = 1, 2, …, n, i = 1, 2, …, n − 1 3 16
The nth equation to complete the set of (3.16) is formulated based on the COI
response (3.1). The frequencies of the areas can then be computed by solving (3.16)
together with (3.1) for the MG-integrated power system. Furthermore, the RoCoF
for the MG-integrated power system can be calculated using (3.10).
3.2.3.2 Delta Frequency Detection

To further proceed with the MG-integrated power system stability assessment, the
frequency response of area i in the Laplace domain form of (3.17) is employed [30].
2πT ij
ΔPL −Δf j s
Δfi s = − s 3 17
2πT ij
M i s + βi +
s
In (3.17), βi is defined as the frequency bias of area i.
By considering ΔPLi in the form of a step function, one could rewrite (3.17) as
ΔPL − 2πT ij Δ f j s
Δfi s = − 3 18
M i s2 + βi s + 2πT ij
in the frequency domain and as

ΔPL − 2πT ij Δ f 1
e − p 2 t − e − p1 t
j
Δfi t = − 3 19
Mi p1 − p2
in the time domain. In (3.19), p1 and p2 are the poles of (3.18). Furthermore, at the
end of the time interval of interest, the frequency deviation is represented by
e − p2 t + ΔT
− e − p1 t + ΔT
j
Δ f i t + ΔT = −
Mi p1 − p2
3 20
Delta frequency detection criterion imposes that the difference between (3.20)
and (3.19) for the main grid as well as the MG-integrated power grid must fulfill
ΔPL − 2πT ij Δ f j 1
e − p2 t − e − p1 t
Mi p1 − p2
− e − p2 t + ΔT − e − p1 t + ΔT < 0 3 3 21
Mi p1 − p2
Following the same procedure as that of (3.10) specifies the frequency deviation
over a given time window for the MG-integrated power grid. For this purpose, all
the variables have specific values except for βi. The steady-state frequency devia-
tions for both the base and reduced systems are employed to calculate βi. In this
way, one could write
Δ f steady-state
β2i = β1i × 1i
3 22
Δ f steady-state
2i
Equation (3.22) describes the frequency bias of the MG-integrated power system
as a function of the main system characteristics.
3.2.4 Simulation and Results

The effectiveness of (3.1)–(3.22) is investigated on the New York New England
(NYNE) test system [31]. This system is a reduced-order equivalent of the intercon-
nected New England Test System (NETS) and New York Power System (NYPS),
with five geographical regions [32]. Figure 3.2 shows the single line diagram of this
system. The study considers two types of loads: constant impedance and induction
motors. The test system data are taken from [33].
Eight scenarios, including the rejection of loads and tripping of the generating
units, are considered to assess the efficiency of the formulation. As the first exam-
ple, the accuracy of the formulation is assessed by solving (3.16) for the tripping
generator number 13. Observe that the system frequency starts to decline from
the nominal value and reaches 49.77 Hz. Figure 3.3 shows the COI frequency
responses of the original and the MG-integrated systems, obtained by T-D
14 Area 3 Area 1
66 8
41 40 48 47 60
1
25 26 29
53
28
10 1 2 61
27
62 9
30
31 3
11
63 18 17
24
46 38 32 16
33 15 22
Area 5 9 21
34 4
19 58
49 14
35 6
42 36 5 20
8 6 12 56 23
67 64
12
57 4
15
51 45 5
59
50 44 37 11 13 7
7 54
2
65 10
52 43 13
39
55
68
3
Area 4 16 Area 2
Figure 3.2 Single line diagram of the 68-bus system showing coherent areas and their
interconnections.
simulation, for various penetration levels, including 5.6% and 8%. The MG-
integrated system is realized by utilizing the MG equivalent model (2.39), devel-
oped in Chapter 2, i.e.
df t
M MG = Tm t − Te t u ζ − u υ 3 23
dt
Also of interest, the center of gravity (COG) concept, introduced in Chapter 2,
can be employed to estimate the local frequencies shown in Figure 3.3.
1
1 5.6% MG penetration
Without MG
0.999 0.999
8% MG penetration
0.998
0.998
f [Hz]
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
0.997
0.996
0.995
0 5 10 15
Time [s]
Figure 3.3 Area 2 frequency in response to reduction of COI.

According to (3.16) and for the penetration levels of 5.6%, one could write:
0 11 f 1 + 0 33 f 2 − 0 02 f 3 − 0 15 f 4 + 0 32 f 5 = 0 092 3 24
0 01 f 1 + 0 43 f 2 − 0 17 f 3 − 0 27 f 4 + 0 71 f 5 = 0 312 3 25
0 67 f 1 + 0 03 f 2 + 0 07 f 3 − 0 83 f 4 + 0 21 f 5 = 0 762 3 26
0 85 f 1 + 0 12 f 2 + 0 37 f 3 − 0 29 f 4 + 0 54 f 5 = 0 341 3 27
To complete the set of equations (3.24)–(3.27), the COI frequency dynamics
MiΔ f i
d
Mi
2M COI = 2π f COI T mech − T elec 3 28a
dt
is employed. Further manipulation of (3.28a) leads to
Mif i
2M COI = 2π f COI T mech − T elec 3 28b
i
Mi
i
By rearranging the terms in (3.28b) with using the definition of RoCoF,

Δfi
i.e. f i = , one obtains
05
Mi f i
2M COI = π f COI T mech − T elec 3 29
i
Mi
i
Initializing (3.29) for the given disturbance leads to

0 32 f 1 + 0 42 f 2 + 0 54 f 3 + 0 76 f 4 + 0 52 f 5 = 0 12 3 30
Solving the set of (3.24)–(3.27) and (3.30) leads to the frequency nadir of 0.9975
pu in Area 2. This indicates a 1.15% error in comparison with those of T-D simu-
lation result. Table 3.1 reports the estimation errors of (3.16) for the penetration
levels of 5.6%.
Furthermore, the estimation errors for the penetration levels of 8% are reported
in Table 3.2. The results of Tables 3.1 and 3.2 suggest that (3.1)–(3.22) can be effec-
tively used to assess the impact of high penetration levels of MGs on the host grid
frequency dynamics.
3.3 Maximum Penetration Level: Frequency Stability
3.3.1 Basic Principle

The frequency metrics in Section 3.2.1 should explicitly follow associated stan-
dards to ensure a reliable supply of electricity [20]. In this section, these metrics
Table 3.1 Frequency dynamics error for NYNE test system under 5.6% penetration level.
Scenario Disturbance Nadir error (%) RoCoF error (%) 15-seconds error (%)
1 G12 ~0 1.99 ~0
2 G13 0.3 1.15 0.11
3 L14 0.32 1.22 0.23
4 L15 0.68 0.46 0.18
5 G16 1.00 0.15 0.10
6 L37 ~0 ~0 0.17
7 L42 1.34 1.04 0.27
8 L52 1.18 0.46 0.12
Table 3.2 Frequency dynamics error for NYNE test system under 8% penetration level.
Scenario Disturbance Nadir error (%) RoCoF error (%) 15-seconds error (%)
1 G12 ~0 0.81 ~0
2 G13 0.13 ~0 ~0
3 L14 ~0 0.38 ~0
4 L15 0.68 0.69 0.41
5 G16 ~0 ~0 0.03
6 L37 ~0 0.09 0.07
7 L42 0.74 0.21 0.07
8 L52 1.98 0.12 0.72
are incorporated into (3.1)–(3.22) to establish a systematic methodology for the cal-
culation of the maximum allowable penetration levels of MGs. A schematic depic-
tion of the calculation process is given in Figure 3.4.
3.3.2 Background on MG Modeling

In what follows, the reduced system is realized by employing the developed equiv-
alent model in Chapter 2. More precisely, the model in (3.23) is replaced by some of
the conventional synchronous generators.
In Chapter 2 and through (2.42), it was shown experimentally that there is a
direct relationship between the inertia constant and the kinetic energy of GENSets
divided by the rated VA, as:
Given Fault
Base System Reduced System
fmin = g1(M) fsteady-state = g2(M) frolling window = g3(M) fRoCoF = g4(M)
Min Instantaneous Min steady state Rolling window RoCoF
M1min M2min M3min M4min
Maximum Allowance Penetration Level
Figure 3.4 Schematic description for the maximum MG penetration level.
M MG = 1 6109Sn − 0 0644 3 31
PGensets
where Sn = . Here, MMG, PGenset, and PMGs are the MGs inertia, the gener-
PMGs
ation of Genset units, and the generation of the MGs, respectively. To realize high
penetration levels of MGs in the power system, multi-MGs (MMGs) structure is
considered. It is assumed that each MG in the MMGs configuration has the same
structure as that of Figure 2.23, but with the different commitments of the loads
and sources. Simulation results suggest that yet there is a direct relationship, with
Root Mean Square Error (RMSE) equal to 0.2165, between the MMGs inertia and
the ratio of the GENSets to the MMGs capacities. It should be noted that power
systems are robust enough so that small variation of the MMGs parameters, caused
by the aforementioned RMSE, may not significantly affect the system perfor-
mance. Therefore, the direct relationship of the form (3.31) can be employed to
quantify the penetration levels of MGs.
3.3.3 Minimum Inertia Related to Frequency Nadir

The first impediment to decreasing the rotational inertia arises from the minimum
instantaneous frequency which the system can experience during the normal oper-
ation. It is noteworthy that a similar criterion, named point C criterion, is intro-
duced by NERC in [34]. As stated in Section 3.2.1, the minimum permissible
frequency is 49.2 Hz.
To introduce the proposed methodology, assume that the same fault is applied to
both, the main and the reduced systems. Accordingly, dividing (3.8) by (3.7) gives:
M2 d Δ f 2 f
= 2 3 32
M1 d Δ f 1 f1
Multiplying both sides of (3.32) by Δf1 results in
M2 d Δ f 2 Δf1
Δf1 = f 3 33
M1 d Δ f 1 f1 2
d Δf2
where the term describes the sensitivity of frequency deviation for the MG-
d Δf1
integrated power system to the frequency deviation for the original system.
The minimum permissible inertia can be calculated by utilizing (3.33). Formally,
d Δf2
Δ f 1 in (3.33), as the frequency change of the MG-integrated power system,
d Δf1
is set to the maximum permissible frequency deviation, i.e. 800 mHz. Further, Δf1
and f1 are the post-fault frequency deviation and frequency for the original system,
respectively. Moreover, the frequency f2 is set to 49.2 Hz. Initializing (3.33) by the
aforementioned values gives:
M2 Δf1
08 = × 49 2 3 34
M1 f1
and then
Δf1
M2 = M1 × 61 5 3 35
f1
Equation (3.35) calculates the minimum rotational inertia which beyond it, the
system frequency tends to deviate from the permissible value. It is noteworthy that
in the case of frequency increment, (3.35) can be rewritten as the form
M2 Δf1
08 = × 50 8 3 36
M1 f1
then
Δf1
M2 = M1 × 63 5 3 37
f1
In deriving (3.35) and (3.37), the amplitude of the disturbance and the kinetic
energy of the rotating masses, as two important factors affecting frequency dynam-
ics [35], are taken into account.
Here, it should be noted that (3.35) and (3.37) are derived for a single-area power
system or the aggregated inertia for all the areas. However, to calculate the min-
imum permissible rotational inertia in each area, (3.11) and (3.12) should be used.
Under these considerations, all the parameters in (3.12) are set to the permissible
values. More precisely, (δi − δj − δi0 − δj0) is set to the maximum permissible phase
difference between areas, fi − fj is set to the maximum permissible frequency devi-
ation and f i − f j will be replaced by an aggregated model of noncoherent gen-
erators, namely
1 Mi + M j Mi Mj
fi −f j = Pmi − Pei − Pmj − Pej 3 38
2π 2M i M j 2M i M j 2M i M j
Afterward, the same procedure as those of (3.35) and (3.37) can be followed to
calculate the maximum permissible inertia reduction.
3.3.4 Minimum Inertia Related to Delta Frequency Detection

To calculate the minimum rotational inertia in compliance with the rolling win-
dow, consider the frequency response of the original system as those of (3.19) and
(3.20). By setting ΔT in (3.20) to 15 seconds, one could write
e − p2 − e − p1
j
Δfi = − t + 15 t + 15
3 39
Mi p1 − p2
According to the delta frequency detection criterion, the difference between
(3.39) and (3.19) must fulfill (3.40) for both the original and the MG-integrated
power systems, i.e.
e − p 2 t − e − p1 t
M 1i p1 − p2
1
− e − p1 t + 15 − e − p2 t + 15 ≤03 3 40
p1 − p2
Following the same procedure as done for (3.35) gives
ΔPL −2πT ij Δ f
M 1i
j 1
p1 −p2 e −p2 t −e −p1 t − 1
p1 −p2 e −p2 t + 15
−e −p1 t + 15
ΔPL −2πT ij Δ f
≤1
M 2i
j 1
p'1 −p'2 e −p'2 t −e −p'1 t − 1
p'1 −p'2 e −p'2 t + 15 −e −p'1 t + 15
3 41
To calculate the minimum rotational inertia, (3.41) is tied to the associated
standard. In this way, Δfj is set to the maximum allowable frequency deviation
in the viewpoint of the transient or the steady-state value (whichever is the larger).
Setting Δfj to the maximum permissible value is a pessimistic assumption that
guarantees the reliable and secure operation of the system for any operating con-
dition. Moreover, (3.41) suggests that except for M and D, the droop characteristic
R significantly affects the maximum penetration levels calculation. The droop

characteristic is defined by (2.40) in Chapter 2, i.e.
Δf Δf Δt
R= = × 3 42
Δp Δt Δp
The DER power-frequency (P/f) grid code determines the permissible value of R
in (3.42). Based on the NERC standard [36], a change of 10% per second for the rate
of response to a step command to reduce power output is reasonable for a variable
generation. This rate is tied to the 15-second-rolling window to calculate the per-
missible droop value, namely
0 3 + 0 5 Hz 1 second
R= × = 0 53 3 43
15 second 0 1 pu
By substituting R in (3.22) for the MG-integrated power system, and considering
the actual value of R for the original system, one can calculate D in the MG-
integrated system. More precisely, while the exact steady-state frequency deviation
for the original system is taken from T-D simulation data, it is set to the maximum
permissible value for the MG-integrated power system. It is noteworthy that all the
parameters in (3.41) have specific values except for M2 and t. Statistical approaches
should be used to calculate M2. For this purpose, a relationship between the rota-
tional inertias of the original and the MG-integrated power systems is derived.
3.3.5 Minimum Inertia Related to RoCoF

The RoCoF, in response to a system perturbation ΔPL, can be represented by
(3.6), as
2M RoCoF = ΔPL 3 44
The TSO’s grid code suggests that a requirement for generation units to remain
connected to the transmission system during frequency excursions involves a max-
imum RoCoF of 0.5–1 Hz/s. More precisely, one can write
ΔPL
= RoCoF ≤ 1 3 45
2M
Rearranging the terms in (3.45) gives
ΔPL
M> 3 46
2
Physically, (3.46) describes the minimum permissible inertia as a function of
the RoCoF.
3.3.6 Maximum Penetration Level

The minimum permissible rotational inertia is defined based on the maximum of
(3.35), (3.41), and (3.46) as
15 − sec
M MMGs = max M nadir
min , M min , M RoCoF
min 3 47
15 − sec
where M nadir
min , M min , M RoCoF
min are the minimum permissible inertia in compliance
with the frequency nadir, 15-second rolling window, and RoCoF, respectively. For
calculation of the maximum permissible penetration levels of MGs, MMMGs in
(3.47) is tied to the equivalent model (3.23). In this way, the same procedure as
that of (3.31) is performed to represent the MMGs system inertia, on the base of
200 MVA, based on the Gensets capacity as
M MMGs = 1 89Sn 3 48
PGensets
where Sn = . Equation (3.48) suggests that each MMGs with 200 MVA gen-
PMMGs
erations, including 200 × Sn MVA Gensets generations, represents
1 89 × Sn second inertia. Accordingly, the maximum penetration levels of MGs
are defined by
PMMGs
P limit = 3 49
Sn
where Plimit, PMMGs, and Sn are the maximum allowable penetration level, the gen-
eration of the MGs, and the apparent power of the system, respectively.

The effectiveness of (3.49) is investigated on two interconnected power systems:
NYNE test system of Figure 3.2 and the IEEE 50-machine test system.
3.3.7.1 Analysis Tools

Four tools are utilized to calculate the maximum penetration level: Matlab 2018b
including power system toolbox (PST) [31] and PLECS [37] are used to perform the
dynamical simulation; MAPLE 18 [38] and DSA tool, including the transient sta-
bility assessment (TSA) and the small-signal stability assessment (SSSA) [39] soft-
wares, are used to assess stability performance.
3.3.7.2 Dynamical Simulation Results

Figure 3.5 shows the frequency responses of area 2, for different rotational inertia.
As shown in this plot, the original system frequency starts to decline following the
outage of generator 14 from the nominal value and reaches the minimum value of
50
49.98
f2 (Hz)
49.96
49.94 COI = 4
COI = 3.90, with MG in area 2 only
COI = 3.80, with MG in all areas
49.92 COI = 3.70, with MG in all areas
0 1 2 3 4 5 6 7 8 9 10
Time (s)
Figure 3.5 Area 2 frequency response in response to reduction of COI.
about 49.96 Hz. Also of interest, the maximum phase differences between areas in
(3.12) are calculated using the DSA tool, as shown in Table 3.3.
Moreover, the term Pmj − Pej in (3.38) is set to zero which satisfies the acceptable
operation in the worst-case condition. On the other hand, fi − fj in the MG-
integrated power system is set to the maximum permissible value, i.e. 1 Hz.
Accordingly, one can write:
0 15M 1 M 2 + 0 07M 1 M 3 − 0 1M 2 M 4 − 0 03M 1 M 5 = 0 3 50
0 06M 2 M 3 + 0 01M 1 M 4 − 0 12M 3 M 4 − 0 05M 3 M 5 = 0 891 3 51
0 08M 3 M 1 + 0 06M 3 M 2 − 0 07M 1 M 4 − 0 09M 4 M 5 = 0 711 3 52
0 13M 4 M 2 + 0 16M 1 M 4 − 0 04M 2 M 5 − 0 03M 1 M 2 = 0 951 3 53
Furthermore, the minimum permissible COI constant, COIoverall, defines the last
equation, namely
50 − 49 59
COI overall = M 1 61 5 3 54
49 59
or in the final form of
Table 3.3 Maximum phase angle difference in deg.
Area 1 Area 2 Area 3 Area 4 Area 5
Area 1 — 13.5 31.5 9.8 11.6

Area 2 13.5 — 16.5 17.2 13
Area 3 31.5 16.5 — 15.6 23.4
Area 4 9.8 17.2 15.6 — 21
Area 5 11.6 13 23.4 21 —
Table 3.4 Minimum permissible rotational inertia in response to tripping G14.
Case study Area 1 Area 2 Area 3 Area 4 Area 5
Base system 3.66 4 3 3 4.45

Reduced system 1.73 1.89 2.14 1.08 2.06
COI overall = 0 51 M 1 = 0 51 3 66 = 1 86 3 55
Solving the set of (3.50)–(3.55), we obtain the minimum inertia on each area of
the system, as reported in Table 3.4.
The minimum permissible inertia in compliance with the 15-second rolling win-
dow, on the other hand, is calculated. Representing the MG-integrated power sys-
tem based on the original system characteristics is done using (3.22), namely
1
D2i +
R2i 0 58
= = 1 16 3 56
1 05
D1i +
R1i
then
1 1
D2i + = 1 16 D1i + 3 57
R2i R1i
where 0.58 and 0.50 are the steady-state frequency deviations and the maximum
allowable deviation, respectively. Substituting (3.57) in (3.41) leads to a time-
variant equation with one degree of freedom, M2. The extracted relationship
between the inertias of the original and the MG-integrated power systems in
the form of
M 2 = 0 94 ln M 1 − 0 23 3 58
is employed to calculate M2. Initializing (3.58) with the actual values gives
M 2 = 0 94 ln 4 − 0 23 = 1 07 3 59
The minimum permissible rotational inertia of area 2 is then defined using
(3.47), namely
M 2,Critical = max 1 89, 1 07, 0 4 = 1 89 3 60
Note that 0.4 in (3.60) comes from (3.46). Equation (3.60) suggests that the fre-
quency nadir imposes a lower limit on the rotational inertia. Making use of (3.60)
into (3.48) and (3.49) leads to the maximum penetration levels of 11.34% expressed
in percentage in terms of the overall system capacity.
The same procedure as those of (3.50)–(3.60) is redone for eight scenarios related
to different disturbances. The numerical results are reported in Table 3.5. The
Table 3.5 Maximum penetration level for 16-machine test system.
Minimum 15-seconds rolling Maximum

Scenario Fault instantaneously window penetration (%)
1 G12 1.76 0.97 14.23

2 G13 1.20 1.07 12.29
3 G14 1.89 1.07 11.34
4 L15 2.18 0.2 10.94
5 G16 2.49 0.6 10.87
6 L37 3.11 1.53 10.56
7 L42 2.01 1.13 12.01
8 L52 2.12 1.65 12.09
Gi: outage of generator number i; Li: outage of load number i.
results suggest that the 15-second rolling window could not affect the maximum
penetration levels of MGs, as the frequency reaches steady state in less than
10 seconds.
To further confirm the validity of the model, the IEEE 50-machine test system is
used. The system is an approximate model of an actual power system. Table 3.6
reports the minimum permissible rotational inertia and the associated maximum
penetration levels of MGs for eight different scenarios. In Table 3.6, the main factor
that imposes a lower limit on the penetration levels of MGs is underlined.
Table 3.6 Maximum inertia reduction for 50-machine test system.
Minimum 15-seconds rolling Maximum

Scenario Fault instantaneously window penetration (%)
1 G2 65.13 66.16 11.44

2 G3 63.12 67.41 12.26
3 G4 71.03 68.86 12.03
4 G6 60.61 64.71 12.98
5 G23 68.39 70.81 15.71
6 L105 56.61 55.29 19.32
7 L144 58.81 54.13 20.27
8 L145 61.12 63.27 17.3
Of note that M RoCoF

min in (3.47) could not generally affect the penetration level, as
the size of disturbance in per unit is negligible in comparison with those of
15 − sec
M nadir
min , M min .
3.4 Small-Signal Stability Assessment
3.4.1 Basic Definition

Following the given definition by Kundur et al. [40], small-signal stability refers to
the ability of synchronous machines of an interconnected power system to remain
in synchronism after being subjected to a small disturbance. Generally, there are
two major types of small-signal instability: (i) increase in rotor angle through a
non-oscillatory or aperiodic mode due to lack of synchronizing torque and
(ii) rotor oscillations of increasing amplitude due to lack of sufficient damping
torque.
Generator-turbine rotational inertia plays an important role in providing syn-
chronizing capability whenever a disturbance causes a mismatch between the
mechanical input power and the electrical output power [40]. Therefore, decreas-
ing overall system inertia in response to increasing penetration levels of MGs can
lead to potential small-signal stability problems, which is addressed in this section.
Moreover, it is advantageous to determine whether a particular generator’s inertia
has a significant impact on a particular inertial oscillation mode. This
section addresses this issue using sensitivity analysis for generators.
3.4.2 Key Concept

Derivation of a linear model of a nonlinear system, around a certain operating
point, is the key factor in small-signal stability analysis. While the stability of a
linearized system is assessed using the eigenvalues, λ, of the state matrix, A, the
participation of each state in a given eigenvalue is determined using the right,
φ, and the left, ψ, eigenvectors [41, 42]. Following the nomenclature in [41],
the ith eigenvalue of the system matrix A, and its corresponding eigenvectors
are defined as
Aφi = λi φi
3 61
ψ i A = λi ψ i
For a complex eigenvalue that corresponds to an oscillatory mode, the fre-

quency, f, of the oscillation and the damping ratio ς are expressed by [43]:
λi = σ i ± jωi 3 62
3.4 Small-Signal Stability Assessment 91
ωi
fi = 3 63
2π
σi
ςi = 3 64
σ 2i + ω2i
In (3.62), a positive real part, i.e. σ i > 0, corresponds to an oscillation with the
increasing amplitude. As the complex pole (3.62) moves towards the right half-
plane (RHP), the damping of the system worsens [41]. Here, the damping ratio
determines the decline rate of a specific mode.
The problem of interest is to study how the critical modes of a system are affected
by penetration levels of MGs. Critical modes of the system are those within the
frequency range of 0.01–2 Hz and damping of less than 10% [41]. For analysis pur-
poses, the simple yet efficient method of [41] is applied to the MG-penetrated
system.
The analysis framework can be summarized as follows:
• Step 1: Calculate the critical modes of the conventional power grid, with no pen-
etration of MGs, using eigenvalue analysis;
• Step 2: Repeat the procedure as Step 1 for the system with different MGs pene-
tration level;
• Step 3: Compare the results to assess the impact of MGs penetration level on the
small-signal stability;
• Step 4: Conduct sensitivity analyses with respect to the displaced generators’

inertia.

To numerically assess the impact of the penetration levels of MGs on small-
signal stability, the NYNE test system of Figure 3.2 is used. The base system
with no MGs penetration levels exhibits four inter-area modes of interest, as
reported in Table 3.7. Refer to Chapter 2 for more details on how to calculate
the modes.
The analysis is reconducted for various MGs penetration levels, and the results
are reported in Table 3.8 for the mode which is detrimentally affected by increasing
MGs penetration levels, i.e. 0.3590 + 3.7108j. The results suggest that the oscilla-
tory mode has been adversely affected by increasing MGs penetration levels. This
could be justified through the fact that the real part of the eigenvalue moves closer
to the RHP as the MGs penetration levels increase. Such justification could be also
observed from the reduction in the damping ratio. Of note, the significant reduc-
tion of the damping ratio from 20% to 30% MGs penetration levels is mainly caused
by displacing some large conventional generators that were in service before the
Table 3.7 Critical modes of the base system.
Mode Real part Imaginary part Frequency Damping ratio
1 −0.7018 1.9710j 0.3137 33.543

2 −0.1184 3.2666j 0.5199 3.6236
3 −0.3590 3.7108j 0.5906 9.6290
4 −0.1657 4.8915j 0.7785 3.3851
Table 3.8 Critical mode detrimentally affected by high MGs.
Penetration level (%) Real part Imaginary part Frequency Damping ratio
0 −0.3590 3.7108j 0.5906 9.6290

10 −0.2816 3.6513j 0.5811 7.0690
20 −0.2201 3.5891j 0.5712 6.1210
30 −0.1305 3.5705j 0.5683 3.6500
30% MGs penetration levels. By displacing those generators, the overall system
inertia dramatically reduces, and hence the damping ratio of the adversely
impacted mode significantly decreases [41].
To further confirm the validity of the framework, sensitivity analysis is carried
out corresponding to the reported mode in Table 3.8. The sensitivity analysis is
used to analytically demonstrate the detrimental impacts of the MGs penetration
levels on the small-signal stability. The ith eigenvalue sensitivity analysis with
respect to the inertia of the jth generator is expressed as
∂A T
∂λi ψ i ∂H j
ϕi
= 3 65
∂H j ψ i ϕi
T
Table 3.9 reports a summary of the sensitivity of the critical mode to the inertia
variations of the conventional generators that are being displaced with the MGs in
the 20% penetration levels. As the damping of the system modes is determined by
the real part of the eigenvalues, the real part is presented in Table 3.9.
The negative real part sensitivity of the eigenvalues in Table 3.9 shows the
adverse impact of high MGs penetration levels on system damping. The sensitivity
analysis results corroborate the results derived from the full eigenvalue analysis of
the system under various MGs penetration levels.
3.5 Maximum Penetration Level: Small-Signal Stability 93
Table 3.9 Eigenvalue sensitivity analysis in the 20% MGs penetration.
Bus number Inertia Sensitivity factor
55 4.96 −0.0213
56 4.16 −0.0189
59 4.32 −0.0200
62 2.91 −0.0103
63 2.00 0.0065
64 1.17 0.0094
3.5 Maximum Penetration Level: Small-Signal Stability

3.5.1 Basic Idea
As discussed above, the critical modes of the system are characterized by frequen-
cies in the range of 0.01–2 Hz and damping of less than 10%. Conceptually, the
zero damping ratio determines the maximum permissible penetration levels of
MGs in the system. The analysis framework to determine the maximum penetra-
tion levels of MGs can be summarized in six steps:
• Step 1: Extract oscillation modes of the conventional power grid with no pene-
tration of MGs, using eigenvalue analysis.
•• Step 2: Compute the critical modes in Step 1;

Step 3: Repeat the procedure in Step 1 for systems with different MGs penetra-
tion levels;
•• Step 4: Determine the damping ratios for different MGs penetration levels;
Step 5: Extract a relationship between the MGs penetration levels and the damp-
ing ratio of the form
ςi = g PMGs 3 66
• Step 6: Set the left-hand side of (3.66) equal to zero and solving for PMGs, one can
find the maximum penetration levels of MGs.

To numerically assess the effectiveness of (3.66) to calculate the maximum pene-
tration levels of MGs, the NYNE test system of Figure 3.2 is employed. Table 3.8
10
9
Damping ratio [%]
8
7
6
5
4
3
0 5 10 15 20 25 30
Penetration level [%]
Figure 3.6 Extracted relationship between damping ratio and penetration level.
describes the variations of the damping ratio as a function of penetration levels of

MGs. Accordingly, one could derive a relationship in the form of
ςi = − 0 1889PMGs + 9 45 3 67
2
between the damping ratio and the MGs penetration level, with R of 0.9731.
Figure 3.6 visualizes the data reported in Table 3.8 as well as the extracted
relationship.
Maximum penetration levels of MGs characterize with zero damping ratio as
0 = − 0 1889PMGs + 9 45 3 68
By rearranging the terms in (3.68), one has
9 45
PMGs = = 50 02 3 69
0 1889
Equation (3.69) exhibits that with the 50% penetration levels of MGs, the damp-
ing ratio of the mode is zero. To further confirm the validity of (3.69), eigenvalue
analysis is conducted for the system with the 50% penetration level of MGs.
A comparison of the results reveals that the error of 1.89% is small enough to justify
the efficiency of the method.
3.6 Voltage-Based Realization of the MG-Integrated

Power Grid
It is well known that MGs are realized in the low and medium voltage levels.
Therefore, following the calculation of maximum penetration levels of MGs in
Sections 3.5 and 3.6, this section tries to determine the maximum active power that
MGs can inject into each bus of the distribution system without causing steady-
state voltage violations.
3.6.1 Key Concepts

The method introduced by Ayres et al. [44] is used to calculate the maximum pen-
etration levels of MGs in the distribution system. The method relies on the Jaco-
bian sensitivities for direct estimation of the maximum amount of active power
that MGs can inject into each bus of the system.
Following the nomenclature of [44], assume that a typical bus in the system has
the initial voltage of V 0. Installing new MGs in the studied bus affects the voltage
which could be represented by
V = V 0 + ΔV P + ΔV Q 3 70
where ΔVP and ΔVQ are voltage changes in response to the variations of the
injected active and reactive powers, respectively. The first step to calculate ΔVP
and ΔVQ in (3.70) is to derive the voltage sensitivities related to the active and reac-
tive power injections.
3.6.2 Jacobian Sensitivities

The voltage sensitivity factors can be calculated using the linearized model of
the form
ΔP J Pθ J PV Δθ
= 3 71
ΔQ J Qθ J QV ΔV
The elements of the Jacobian matrix (J) represent the sensitivities of the power
variations, i.e. ΔP, ΔQ, to the voltage variations, i.e. ΔV, Δθ.
3.6.2.1 V-P Sensitivity

The sensitivity of the voltage to the active power variations in a typical bus, i.e. V-P
sensitivity, can be calculated by setting ΔQ = 0 in (3.71). In this way, one
could write
−1
ΔP = J PV − J Pθ J Qθ J QV ΔV P = J RPV ΔV P 3 72
By rearranging the terms in (3.72) with the use of the definition of the reduced
Jacobian matrix JRPV, one obtains
−1
ΔV P = J RPV ΔP 3 73
It should be noted that inversion of JQθ is feasible only if all the buses are mod-
eled as PQ buses. This is the case for distribution systems, where the slack bus is the
only voltage control bus [44]. Moreover, MGs in the steady-state studies are for-
mally represented by PQ buses since they do not contribute to the voltage control
of the system.
3.6.2.2 V-Q Sensitivity

Equation (3.73) is derived assuming the unity power factor. To tackle the power
factor into the problem formulation (3.73) should consider V-Q sensitivities. Fol-
lowing the same procedure as that of (3.72) gives
−1
ΔQ = J QV − J Qθ J Pθ J PV ΔV Q = J RQV ΔV Q 3 74
Rearranging the terms in (3.74) gives the final form of

−1
ΔV Q = J RQV ΔQ 3 75
where JRQV is the reduced Jacobian matrix, which represents the sensitivity of
voltage magnitude with respect to the reactive power injection variations. As
the penetration levels of MGs express in the term of active power, one could
substitute
ΔQ = ΔP × tan cos − 1 pf 3 76
in (3.75) to derive
−1
ΔV Q = J RQV ΔP × tan cos − 1 pf 3 77
Considering (3.73) and (3.77), the bus voltage V in (3.70) is rewritten as

−1 −1
V = V 0 + J RPV ΔP + J RQV ΔP × tan cos − 1 pf 3 78
and then
−1 −1
V = V 0 + J RPV + J RQV × tan cos − 1 pf ΔP 3 79
By defining
−1 −1
SPQ = J RPV + J RQV × tan cos − 1 pf 3 80
Equation (3.79) gets the final form of
V = V 0 + SPQ ΔP 3 81
Considering the acceptable steady-state voltage, the maximum permissible MGs
penetration level in each bus of the distribution system is defined by
ΔV
ΔP = 3 82
SPQ
Indeed (3.82) specifies the capacity of MGs which could be installed in the stud-
ied bus without violating the steady-state voltage.

The practical applicability of the above framework is assessed using the IEEE
70-Bus test system of Figure 3.7. The system data are taken from [45]. The network
active and reactive power demands are 3802.2 kW and 2694.6 kVAr, respectively.
In the studies described below, the transformer taps are adjusted to 1.04 pu to
29 30 31 32 33 34 35 36
48 49 50 51
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28
67 68 69 70
54 55 56 57 58 59 60 61 62 63 64 65 66
52 53
37 38 39 40 41 42 43 44 45 46 47
Figure 3.7 Single-line diagram of the test system.

(a)
1
Maximum active power injection per bus [pu]
0.8
0.6
0.4
0.2
0
0 10 20 30 40 50 60 70
Bus number
(b)
0.98
0.96
V [pu]
0.94
0.92
0.9
0 10 20 30 40 50 60 70
Bus number
Figure 3.8 (a) Maximum active power injection per bus and (b) voltage magnitude per bus.
3.7 Summary 99
pf = 1
pf = 0.95
1 pf = 0.90
Maximum active power injection per bus [pu]
0.8
0.6
0.4
0.2
0
0 10 20 30 40 50 60 70
Bus number
Figure 3.9 Maximum active power injection per bus for various power factor.
maintain the voltage magnitude of every bus within the allowable range of
0.95–1.05 pu, for the case without DGs.
Considering the unity power factor, Figure 3.8a demonstrates the maximum
active power injection per bus in the system. For this purpose, the minimum per-
missible voltage magnitude in (3.80) considers being 0.9 which is an acceptable
value for an emergency condition. Figure 3.8b shows the voltage magnitude per
bus after installing the maximum capacity on each bus.
Also of interest, Figure 3.9 depicts the effect of power factor on the maximum
active power injection per bus. It appropriately demonstrates the effectiveness
of (3.80).
3.7 Summary
MG-integrated power system stability analysis from frequency, small-signal, and

voltage points of view is discussed in this chapter. Furthermore, some mathemat-
ical-based approaches have been discussed to determine the maximum penetra-
tion level of MGs in the grid. The methods explicitly rely on the basic power
system equations which, in turn, make the proposed frameworks completely inde-
pendent of the test cases.
References
1. Hatziargyriou, N., Asano, H., Iravani, R., and Marnay, C. (2007). Microgrids: an
overview of ongoing research, development, and demonstration projects. IEEE
Power Energy Magazine 5 (4): 78–94. https://doi.org/10.1109/MPAE.2007.376583.
New York: Wiley.
3. Lasseter, R.H. (2002). Microgrids. In: 2002 IEEE Power Engineering Society Winter
Meeting. Conference Proceedings (Cat. No. 02CH37309), vol. 1, 305–308. IEEE.
4. Lasseter, R.H. and Paigi, P. (2004). Microgrid: a conceptual solution. In: Power
Electronics Specialists Conference, 2004. PESC 04. 2004 IEEE 35th Annual, vol. 6,
4285–4290. IEEE.
33 (1): 1026–1035.
7. Milano, F., Dörfler, F., Hug, G. et al. (2018). Foundations and challenges of
low-inertia systems. 2018 Power Systems Computation Conference (PSCC): IEEE,
pp. 1–25.
8. Golpîra, H., Seifi, H., Messina, A.R., and Haghifam, M.R. (2016). Maximum
viewpoint. IEEE Transactions on Power Systems 31 (6): 5163–5171. https://doi.org/
10.1109/TPWRS.2016.2538083.
on power system stability and operation. IFAC Proceedings 47 (3): 7290–7297.
10. Vittal, E., Keane, A., and O’Malley, M. (2008). Varying penetration ratios of wind
turbine technologies for voltage and frequency stability. In: Power and Energy
Society General Meeting-Conversion and Delivery of Electrical Energy in the
21st Century, 2008 IEEE, 1–6. IEEE.
11. Lalor, G., Mullane, A., and O’Malley, M. (2005). Frequency control and wind
turbine technologies. IEEE Transactions on Power Systems 20 (4): 1905–1913.
12. Folly, K. and Sheetekela, S. (2009). Impact of fixed and variable speed wind
generators on the transient stability of a power system network. In: Power Systems
Conference and Exposition, 2009. PSCE’09. IEEE/PES, 1–7. IEEE.
References 101
13. Mitra, A. and Chatterjee, D. (2013). A sensitivity based approach to assess the
impacts of integration of variable speed wind farms on the transient stability of
power systems. Renewable Energy 60: 662–671.
14. Fernandez-Bernal, F., Egido, I., and Lobato, E. (2014). Maximum wind power
generation in a power system imposed by system inertia and primary reserve
requirements. Wind Energy 18 (8): 1501–1514.
15. Doherty, R., Mullane, A., Nolan, G. et al. (2010). An assessment of the impact of
wind generation on system frequency control. IEEE Transactions on Power Systems
25 (1): 452–460.
16. Yan, R., Saha, T.K., Modi, N. et al. (2015). The combined effects of high penetration
of wind and PV on power system frequency response. Applied Energy 145: 320–330.
17. Ahmadyar, A.S., Riaz, S., Verbič, G. et al. (2018). A framework for assessing
renewable integration limits with respect to frequency performance. IEEE
18. Keyhani, A. and Chatterjee, A. (2012). Automatic generation control structure for
smart power grids. IEEE Transactions on Smart Grid 3 (3): 1310–1316.
19. Chen, X., Lin, J., Wan, C. et al. (2019). A unified frequency-domain model for
automatic generation control assessment under wind power uncertainty. IEEE
Transactions on Smart Grid 10 (3): 2936–2947. https://doi.org/10.1109/
TSG.2018.2815543.
20. U. O. Handbook (2009). P1–Policy 1: load-frequency control and performance [C],
Version: v3. 0 rev, vol. 15, no. 01.04.
23. Karimi, H., Karimi-Ghartemani, M., and Iravani, M.R. (2004). Estimation of
frequency and its rate of change for applications in power systems. IEEE
Transactions on Power Delivery 19 (2): 472–480.
25. Illian, H.F. (2011). Frequency Control Performance Measurement and Requirements.
Berkeley, CA: Lawrence Berkeley National Laboratory.
26. NERC (2013). Standard BAL-003-1 – frequency response and frequency bias
setting.
27. Bevrani, H. (2008). Robust Power System Frequency Control. New York: Springer.
28. Golpira, H. and Bevrani, H. (2011). Application of GA optimization for automatic
generation control design in an interconnected power system. Energy Conversion
and Management 52 (5): 2247–2255.
29. Golpira, H., Bevrani, H., and Naghshbandy, A.H. (2012). An approach for
coordinated automatic voltage regulator–power system stabiliser design in large-
scale interconnected power systems considering wind power penetration. IET
Generation, Transmission & Distribution 6 (1): 39–49.
30. Bevrani, H. and Hiyama, T. (2017). Intelligent Automatic Generation Control. USA:
CRC Press.
31. Chow, J. and Rogers, G. (2000). Power system toolbox. Cherry Tree Scientific
Software. https://www.ecse.rpi.edu/~chowj/PST_2020_Aug_10.zip (accessed
November 2020).
32. Wilson, D.H., Hay, K., and Rogers, G.J. (2003). Dynamic model verification using a
continuous modal parameter estimator. In: Power Tech Conference Proceedings,
2003 IEEE Bologna, vol. 2, 6. IEEE.
33. Rogers, G. (2000). Power System Oscillations. Norwell, MA: Kluwer.
34. Miller, N., Shao, M., Venkataraman, S. et al. (2012). Frequency response of
California and WECC under high wind and solar conditions. In: Power and Energy
Society General Meeting, 2012 IEEE, 1–8. IEEE.
35. Baggini, A.B. (2008). Handbook of Power Quality. New York: Wiley Online Library.
36. Abraham Ellis, R.W., Zavadil, B., Jacobson, D., and Piwko, R. (2012). 2012 Special
Assessment Interconnection Requirements for Variable Generation. NERC Report,
pp. 6–17.
37. PLECS (n.d.). Web address. http://www.plexim.com (accessed November 2020).
38. MAPLE (2018). Web address. https://maplesoft.com (accessed November 2020).
39. DSA Tools (n.d.). Dynamic security assessment software. Powertech Labs Inc.
http://www.dsatools.com/ (accessed November 2020).
40. Kundur, P., Balu, N.J., and Lauby, M.G. (1994). Power System Stability and Control.
New York: McGraw-Hill.
41. Eftekharnejad, S., Vittal, V., Heydt, G.T. et al. (2013). Small signal stability
assessment of power systems with increased penetration of photovoltaic
generation: a case study. IEEE Transactions on Sustainable Energy 4 (4): 960–967.
42. Messina, A.R. (2009). Inter-area Oscillations in Power Systems: A Nonlinear and
Nonstationary Perspective. Germany: Springer Science & Business Media.
43. Ugwuanyi, N.S., Kestelyn, X., Thomas, O. et al. (2020). A new fast track to
nonlinear modal analysis of power system using normal form. IEEE Transactions
on Power Systems 35 (4): 3247–3257. https://doi.org/10.1109/
TPWRS.2020.2967452.
44. Ayres, H., Freitas, W., De Almeida, M., and Da Silva, L. (2010). Method for
determining the maximum allowable penetration level of distributed generation
without steady-state voltage violations. IET Generation, Transmission &
Distribution 4 (4): 495–508.
45. Baran, M.E. and Wu, F.F. (1989). Optimal capacitor placement on radial
distribution systems. IEEE Transactions on Power Delivery 4 (1): 725–734.
103
Advanced Virtual Inertia Control and Optimal Placement
Power grids worldwide have experienced a significant transformation, which has

been predominantly characterized by increased penetration of power electronic-
based technologies. Among these new technologies are wind and photovoltaic
(PV) generations, various storage technologies, flexible AC transmission systems
(FACTS) devices, distributed generations (DGs), high-voltage direct current lines,
and responsible loads. With significant integration of power electronic-based tech-
nologies, the dynamic behavior of power systems has progressively become more
dependent on fast-response power electronic devices, thus, altering the power sys-
tem dynamic behavior. Accordingly, new stability concerns have arisen which
need to be appropriately addressed.
In Chapter 3, a framework for stability assessment of renewable integrated
power systems was presented. The impediments to realizing high renewable inte-
grated power systems were also discussed in small-signal and frequency stabilities
points of view. This chapter discusses special attributes of the specific power elec-
tronic-based technologies to enhance system dynamics. Using this framework, the
special attributes of the technologies are modeled using the well-known synchro-
nous generator (SG). Optimal placement of the technologies, as the main step
toward the successful implementation of advanced control schemes in Chapters
6 and 8, is also discussed in detail.
4.1 Introduction
penetration of power-electronic-interfaced loads, DG, and microgrids (MGs). Par-
amount among these is the inertia requirement challenge, as inverter-connected
renewable sources are increasingly replacing SGs. Unlike the conventional SGs,
inverter-based DGs do not provide physical inertia to support the grid. Therefore,
104 4 Advanced Virtual Inertia Control and Optimal Placement
since SGs are gradually replaced by DGs, transmission system operators (TSOs) are
faced with the issue of lack of inertia, which intrinsically leads to a large rate of
change of frequency (RoCoF) and power variations in the grid. As a result, the
power system is prone to frequency and power fluctuations, and the design of
relays, including RoCoF-based relays, should be reconfigured.
As suggested by analytical results in Chapter 3, rotational inertia reduction in
the grid may adversely affect frequency dynamics, small-signal stability, and sys-
tem control and leads to degrade the performance of traditional control schemes.
This, in turn, may result in large dynamic deviations and, potentially, load shed-
ding, and instability. Accordingly, the increasing penetration of inverter-
interfaced DGs motivates the need to develop additional ancillary control services
to improve system dynamics and stability.
To address this issue, the concept of the virtual synchronous generator (VSG),
virtual synchronous machine, or synchronverter has been proposed [1, 2]. It is
shown that by adding short-term energy storage to emulate the kinetic energy
of a rotating mass and mimic the swing equation of an SG in the control scheme,
inverters can also provide inertia support for the grid to restrain its frequency fluc-
tuation, in the same way as an SG. Since the principle of these concepts is similar,
for convenience sake, all these inverters are referred to as VSG in this chapter.
Manipulation of the inertia constant by the VSGs could enhance system dynam-
ics and stability. This could be realized using inertia-based control schemes.
Indeed, inertia provision plays an important role in designing advanced control
schemes in modern power grids. However, the successful implementation of
advanced control schemes not only depends on the nature of the designed scheme
but also the adequacy of the inertia provision sources. In this way, after a discus-
sion on the application of VSGs in power system stability and performance
improvement, optimal placement of dispatchable inertia besides the associated
realization mechanisms would be emphasized in this chapter and afterward,
the inertia-based advanced control schemes are addressed in Chapter 6. Without
the loss of generality, while DGs (such as wind farms), high-voltage direct current
(HVDC) systems, and energy storage systems (ESSs) could emulate virtual inertia,
the ESSs are considered as the source of virtual inertia in this chapter.
4.2 Virtual Synchronous Generator
Power grid control must provide the ability of electric power systems to regain a
state of operating equilibrium after being subjected to a physical disturbance, with
most system variables, i.e. frequency, voltage, and angle, bounded so that
practically the entire system remains intact. Thus, the main control loops are
known as frequency control, voltage control, and power oscillation damping
control.
Additional flexibility may be required from various control levels so that the sys-
tem operator can continue to balance supply and demand on the modern power
grids. The contribution of DGs and renewable energy sources (RESs) in regulation
task refers to the ability of these resources to regulate their power output, by appro-
priate control action. This can be regarded as adding virtual inertia to the grid and
considered a solution. Virtual inertia emulation system requires the inverter to be
able to store or release an amount of energy depending on the grid frequency’s
deviation from its nominal value, analogous to the inertia of a conventional gen-
erator. This system is known as VSG and will then operate to emulate desirable
dynamics, such as inertia and damping properties, by flexible shaping of its output
active and reactive powers.
4.2.1 Concept and Structure

The VSG concept was introduced as a promising solution toward the grid stability
and performance issues caused by high penetration of RESs. A VSG can be estab-
lished by using a DG or short-term ESS coupled with the power electronics
inverter/converter with an appropriate control mechanism [2].
The general architecture of a VSG is presented in Figure 4.1. In this scheme, a
primary power source such as a DG unit is connected to the grid via a power
inverter (VSG system). It is expected to operate as a conventional SG by providing
inertia and damping property virtually and by displaying the same reaction as
the conventional SG when there is a sudden change of load or disturbance in
the system. The VSG control block is expected to regulate the output of the
inverter based on the RoCoF and the difference between the reference frequency
and grid frequency like the way conventional SGs are governed by the swing
equation:
d2 d
ΔPVSG = Pm − Pe = JΔω' + DΔω = J 2 δ t + D dt δ t 41
dt
where J is the inertia coefficient, D is the damping coefficient, and δ is the rotor
angular position difference from its reference position.
The swing Equation (4.1) describes the relative motion of the rotor in respect to
the stator field as a function of time. The generator inertia reacts to the disturbance
and plays a significant role in power system stability, which is the most important
property of a synchronous machine.
Pin VSG Pout Power

Grid
Primary DC/DC
Source
DC/AC
Inverter
AC/DC
VSG Control
Supervisory command
Figure 4.1 Conceptual and general structure of VSG.
4.2.2 Basic Control Scheme and Applications

A typical control scheme of a VSG system is shown in Figure 4.2 [1]. It is an RMS-
value-based control scheme without using an additional inner voltage or current
control loop. Different from VSG control schemes with an inner voltage loop, this
Lls Lf Zline
Primary Storage Cf
Source Device Vout(abc) Iout(abc)
BUS
Q0
Qref + +
Vbus Q Droop PI
^
E
– +
Vbase Virtual Vpwm
Impedance PWM
P0 Control θpwm abc/αβ
Virtual
Governor and 1/s θm V out(αβ) Iout(αβ)
ωm
Inertia
Pout
Power
Qout
LPF Meter
Vbus
Estimator
Figure 4.2 Basic control scheme of a VSG.

type of VSG control regulates the output voltage through the reactive power con-
trol loop. Owing to the absence of an inner voltage loop, the active power control
loop becomes quite simple and robust, and bus voltage deviations become smaller
and insensitive to the output impedance [3].
In the power generation part, the “Virtual Governor and Inertia” block is the
core of VSG control. In the literature, various approaches using different damping
technologies are proposed for this part. A virtual voltage drop over virtual induc-
tance Lls is generated to adjust the equivalent output reactance X of the inverter as
shown in (4.2), given by
X = ω0 LLS + L f + Lline 42
The block “Vbus estimator” estimates the bus voltage from the measurement of
output voltage and current to provide a common reference for the block “Q
Droop,” which makes a simple linear function between voltage and reactive
power [1]. To emulate the steady-state operation of an SG, the governor model
is usually simulated in a dispatchable VSG-based power source, as shown in (4.3):
Pin = P0 − k p ωm − ω0 43
To mimic the dynamics of an SG, the swing equation should be emulated. If the
effect of damper windings is omitted, the swing equation of an SG can be
expressed as:
dωm
Pin − Pout = Jωm 44
dt
Combining (4.3) and (4.4) yields
dωm
P0 − Pout = Jω0 + k p ωm − ω0 45
dt
In interpreting the control structure in Figure 4.2, note that the VSG control
emulates Equation (4.5) through the “Virtual Governor and Inertia” block.
Apart from the VSG control topology shown in Figure 4.2, several different VSG
controls exist in the literature. The VSG control schemes can be categorized into
three main groups based on the nature of the output reference from the VSG [4].
1) Current references-based VSG control: This group utilizes current references

from VSG and allow a quite natural implementation of high-order electrical
models of the SG since inverters are controlled to generate the currents that
would result from a real SG.
2) Voltage references-based VSG control: This group is another possible approach
to utilize the voltage references from VSG. The group can be referred to as “volt-
age references from VSG.” Although only a reduced-order model of the SG can
be applied in this approach, but unlike the first group, the utilizing voltage
command allows this VSG to function in stand-alone mode.
3) Power references-based VSG control: This group emulates the inertia response
by tracking the grid frequency without implementing any SG model. The group
is called “power references from VSG,” as the current reference corresponding
to a given power reference is used to control the inverter.
Some applications of VSGs in power systems have been reported in [5–10]. For
example, the VSG concept was successfully implemented in [5] for multiterminal
HVDC systems to suppress low-frequency oscillation and enhance the power oscil-
lation damping performance of the AC/DC system. A dual VSG-based modular
multilevel matrix converter (M3C) control scheme for frequency regulation sup-
port of a remote AC grid via low-frequency AC transmission system is proposed
in [6]. In [7], a VSG is used to enhance the performance of a stand-alone gas engine
generator.
Furthermore, several studies considering the application of VSG in PV systems
[8], wind power generation [9], permanent magnet synchronous generators
(PMSGs) [10, 11], and doubly fed induction generators (DFIGs) have been
reported. Besides, the VSG control can be applied to other grid-tied inverters, such
as those in the ESSs [12], bidirectional battery chargers of electrical vehicles (EVs)
providing vehicle-to-grid (V2G) services [13], and voltage source converters (VSCs)
in high-voltage DC transmission system [5, 14–17]. Generally, the VSG-based DC–
AC converter becomes a standard interface for smart grid integration [18].
As emphasized above, the VSGs with emulated inertia capabilities may play a
critical role in reducing the RoCoF and the frequency nadir of the power system
[19]. Owing to this promising feature, it attracts a lot of attention in recent years.
These works are mainly targeted at improving the damping of a VSG via fixed
parameter or adaptive parameter methods [1], improving its parameter tuning
[20], fault ride-through ability, power quality, and applying this concept in various
types of grid-interfaced power electronics devices [6, 7, 21, 22].
4.2.3 Application in Power System Dynamic Enhancement

Although the applications of VSGs in an MG have been fully studied, the grid inte-
gration performance of VSG is not well understood. In previous VSG studies, the
grid is usually modeled as an infinite bus or a single SG. Several works have inves-
tigated the interaction between one VSG and one SG in an isolated MG. However,
these studies overlook the most interesting case of interaction between the VSGs
and multiple SGs.
In this subsection, the impact of high penetration of inverter-based DGs with
and without VSG on system dynamic performance is illustrated on the IEEE 9-
bus system shown in Figure 4.3. Detailed parameters of generators, governors,
Figure 4.3 IEEE 9-bus system. 2 T2 7 8 9 T3 3

Line7_8 Line8_9
G2 G3
Line5_7 L3 Line6_9 L4
5 6
L1 Line4_5 Line4_6 L2
4
T1
1
G1
steam turbines, and exciters are taken from [23], except minor changes mentioned
in [24]. Power system stabilizers (PSSs) are not included in the system, and the
rated power of G1, G2, and G3 are 512, 270, and 125 MVA, respectively.
Three case studies to evaluate the grid integration performance of VSG are con-
sidered. In Case 1, it is assumed that all three generators in Figure 4.3 are SGs. Case 2
is the same as Case 1, except that a cluster of coherent grid-following inverter-based
DGs with the same overall rated power replaces G3. For the clarity of illustration,
the DGs are aggregated into a single DG of 125 MVA. Finally, Case 3 is similar to
Case 2, except that the VSG control is used to control the inverter-based DGs.
In these studies, the VSG is designed to have the same inertia and droop coef-
ficient as G3 in Case 1, while the equivalent reactance from the transient emf to the
adjacent bus of the G3 node is 0.3 pu, and the damping ratio of VSG is set to 0.9.
4.2.3.1 Scenario 1: 10-MW Load Increase at Bus 9

Figure 4.4 shows the time-domain results, including the frequency and output
active power of G1 and G2, when a 10-MW load suddenly connected to Bus 9.
The grid integration performance of DGs can be evaluated based on Figure 4.4.
Comparison of Cases 1 and 2 shows that increased penetration of inverter-based
DG leads to more severe frequency nadir and higher RoCoF after the disturbance.
The results of Case 3 show that these issues can be addressed by the VSG control
with a considerable mitigation in RoCoF and frequency nadir. Moreover, Case 3
has a better performance than Case 1, in terms of the improved frequency nadir.
4.2.3.2 Scenario 2: 20-MW Power Command Decrease of G3

The system performance is also investigated following a 20-MW step decrease in
power control input of G3. The system responses including the frequency and out-
put active power of G1 and G2 are shown in Figure 4.5. Comparison between the
ωm1 (rad/s)
377
376.8
Case 1
376.6 Case 2
Case 3
376.4
377
ωm2 (rad/s)
376.8
376.6
376.4
Po1 (MW)
78
76
74
72
Po2 (MW)
168
166
164
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Time (s)
Figure 4.4 System response for a 10-MW load increase at Bus 9.
studied cases further verifies the conclusion obtained with Scenario 1. The VSG
improves frequency nadir and reduced transient RoCoF. On the other hand, Case
1 shows completely different dynamics, because changing the power command of
an SG is delayed by its slow governor and turbine responses.
4.2.4 Application to Power Grids with HVDC Systems

The large penetration of RESs has added a large power density to the existing
power system. This increased power density and efficiency have brought many
challenges such as power quality issues, power delivery, and transmission. Numer-
ous topologies and schemes have been suggested to deal with power control and
transmission issues of the renewable integrated power grids.
As an important example, consider the test system shown in Figure 4.6, in which
the wind farms are located far from the load centers and thus require a long trans-
mission lines. For farther wind farms installations, current trends in research and
practice point toward the use of HVDC transmission with VSC-based HVDC
transmission being the preferred approach as it displays distinct control and design
advantages over traditional line-commutated converter technology [25]. In such
installations, the wind farm collector network typically operates at 50 or 60 Hz,
377
ωm1 (rad/s)
376.5 Case 1
Case 2
376 Case 3
377
ωm2 (rad/s)
376.5
376
Po1 (MW)
80
Po2 (MW)
170
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Time (s)
Figure 4.5 System response for a 20-MW step decrease in G3 power command.
which is then converted to HVDC by a wind farm converter station for transmission
to a load center converter station (Figure 4.6).
The VSC-HVDC is currently considered as the market leader for wind integra-
tion at distances greater than 60–80 km largely due to its established use in point-
to-point bulk power transfer [26]. The HVDC system is broadly applied for
Main Grid
Power
converter
HVDC System
Wind farms
Small power
stations Rural/remote
Load centers
Figure 4.6 Power grid with HVDC system.

point-to-point interconnection of two power systems. It is also well used for inter-
connecting the offshore wind farm with the onshore stations using a subma-
rine cable.
An HVDC link may improve the system stability of a wide area power system by
decoupling of some areas. For instance, consider two interconnected AC grids by
an HVDC link. This system exhibits a natural decoupling between the AC grids in
terms of both voltage and frequency (the AC grids are interconnected in an asyn-
chronous manner). These kinds of interconnections are primarily aimed at pre-
venting the excursions of oscillations between AC systems, for instance,
between a stronger and a weaker electric network [27]. Furthermore, the HVDC
transmission technology is currently deemed a desirable solution for long-distance
power transmission and offshore RESs.
In contrast, the VSC-HVDC systems have some weak points. The HVDC system
is sensitive to faults on the DC lines. When a fault occurs on the DC side of a
VSC-HVDC system, the insulated-gate bipolar transistor (IGBT) cannot control
and freewheeling diodes work as a bridge rectifier. It is not able to withstand large
surge currents and may be damaged before the fault is cleared. Some solutions are
proposed; however, additional control and switching devices are needed.
An HVDC system uses the VSC stations that, from a physical inertia point of
view, their dynamic behavior is quite different from that of the SGs. Thus, it makes
the frequency stability problem in the connected AC grid more serious than the
past. For this reason, the HVDC systems are expected to participate in the grid
frequency support. This issue is more significant when it is known that the inte-
gration of RESs instead of SGs reduces system inertia. Consequently, the system
frequency stability and dynamic performance are affected. We can consider this
issue from a different perspective. One typical application of HVDC transmission
systems is to supply a weak or passive/islanded grid far away from the main grid.
With constant power control, the output power from the HVDC-VSC station is
constant, and no frequency support is provided to the weak grid. When a signif-
icant part of the power in a local gird is supplied by this station, the system inertia
becomes lower, and changes in generation or load may cause large-frequency
deviations possibly leading to system instability [14]. In response to this challenge,
it could be a promising solution to use the VSC station to provide frequency sup-
port to the local grid.
There are several works investigating system frequency support through HVDC
systems [5, 14, 27–29]. The main idea is to use the VSG concept for grid-tied HVDC
converters to emulate the classical SG, thus providing virtual inertia and system
frequency support, as shown in Figure 4.7.
HVDC System VSG Grid
VDC
DC/AC
Inverter
VSG Control
Supervisory command
Figure 4.7 HVDC system as a primary source for grid frequency support.
4.3 Dispatchable Inertia Placement
Not only an insufficient level of inertia but also its heterogeneous distribution may
render system dynamics faster. This fact, along with the need to economically
keeps the system secure, and makes the optimal placement of virtual inertia as
a key factor [30]. Accordingly, optimal placement of virtual inertia can be studied
in two perspectives: (i) frequency stability and (ii) small-signal stability.
4.3.1 Frequency Dynamics Enhancement

The ever-growing number of frequency incidents, in response to fluctuations of
renewable power sources, accompanied by low-level inertia feature jeopardize fre-
quency stability [1, 30]. Following a disturbance, inertial response may positively
affect the power imbalance before the activation of traditional frequency control
loops. This means that efficient system inertia has a major influence on the fre-
quency stability and performance characteristics such as the RoCoF and frequency
nadir [31].
4.3.1.1 Background: Literature Review

The effects of ESSs on frequency regulation have been studied in recent research
works [32–34]. These works deal with the problem of balancing of generation and
load to maintain a constant system frequency and to keep tie-line power flows
within some prespecified tolerance. However, these studies consider long-term
frequency response as well as steady-state metrics as the main metrics utilized for
system resilience analysis, neglecting the impact of inertial response from RESs on
system stability. In this direction, some recent studies have investigated the effect
of reduced inertia on frequency stability and transient stability of the power system
[35–42]. In parallel with these efforts, some recent works, [43–50], investigate ways
in which virtual inertia could be emulated, including appropriate control of wind
turbines and ESSs. In [51], a framework that addresses various aspects of inertia
emulation and control, including how virtual inertia emulation and its location in
the system impacts system stability, is proposed. Some questions about the heter-
ogeneous inertial profiles and how the associated negative impacts are reduced by
inertia emulation have been raised in [38]. Further, Poolla et al. [30] propose an
H2-based performance metric to determine the optimal placement of virtual
inertia. The determination of the optimum size of ESS to provide the primary
frequency control is addressed in [52, 53]. However, due to the lifetime concerns,
the ESSs cannot effectively participate in the primary regulation. In practice, the
ESSs are dispatched using an optimal control strategy, designed to optimize the
state-of-charge (SOC) range and the lifetime constraints. Some research works,
such as in [54], deal with the optimal placement of virtual inertia in power systems
considering network structure. These approaches utilize DC power flow to incor-
porate network structure into the model.
The main limitation of these approaches is their reliance on static considerations.
It is well known, however, that dynamic frequency indices, such as the RoCoF and
frequency nadir, are important parameters to assess frequency performance and the
development of protection schemes. Based on these considerations, the following
subsections formulate the problem of finding virtual inertia locations in terms of
dynamic as well as static metrics. The main step toward the optimal placement
of virtual inertia in a power grid is to analyze its effects on the frequency behavior;
and this could be realized by the appropriate modeling of the virtual inertia.
4.3.1.2 Virtual Inertia Modeling

Main Idea
As discussed in Section 4.2, the VSGs can be used to provide virtual inertia; the
VSG-based virtual inertia emulation strategy is based on a similar power-bal-
ance-based synchronization mechanism to that defined by the swing equation
of a conventional SG [55–57]. This means that the effects of virtual inertia on
the frequency indices can be assessed using a dynamical equivalent model
obtained by mapping the electromechanical behavior of the ESSs onto the sec-
ond-order SG model.
To illustrate this idea, let the swing equation of a conventional SG be
expressed as:
s
δ = ωs t − ω0
46
Mωs = T m t − T e t
s
where δ , ωs, and ω0 are the mechanical rotor angle, the mechanical rotor angular
speed, and the initial angular speed, respectively; M, Tm(t), and Te(t) are the inertia
constant, the mechanical input torque, and the electrical output torque, respec-
tively [58]. Taking the slow electromechanical behavior of the ESS into account,
the associated dynamics could be represented by (4.6). The problem of interest,
however, is to calculate the equivalent inertia constant and the mechanical input
torque.
In what follows, a data-driven approach in which the uncertain behavior of the
ESS is accounted for in the swing Eq. (4.6) is proposed.
MUSIC Analysis: Methodology and Application

To introduce the adopted model, assume that the injected power of the ESS to the
host grid is a discrete-time signal P(n) of length L. Let the time-varying signal P(n)
be decomposed into a summation of K sinusoidal components and noise, as
[59, 60]:
K
P n = ak cos nωk + Φk + w n 47
k=1
where, ak, Φk, ωk, and w(n) are the magnitude and the initial phase angle, har-
monic frequency in radius, and additive white noise, respectively. In the model,
ak and ωk are assumed to be deterministic and unknown, and Φk is unknown
and assumed to be random and uniformly distributed in [−π, π]. Alternatively,
the model (4.7) can be expressed in the form of noisy complex exponentials as [59]:
K
P n = Ak e jnωk + w n 48
k=1
where Ak = |Ak|ejΦk is the complex magnitude of the kth-harmonic (noise) signal

component. As the MUltiple Signal Classification (MUSIC) algorithm is a noise
subspace-based method, it is a good tool to deal with experimental noisy measured
signals. Using this framework, the dimensional space is divided into the signal and
noise components, which is of high importance to accurately calculate M and Tm
(t) in (4.6).
The MUSIC method employs a harmonic model and estimates the frequencies
and powers of the signal harmonics. Application of the MUSIC method to the data
sequence P(n), gives
1
PMUSIC ejω = z 2 49
i = K+1 eU si
where si is the eigenvectors associated with the noise subspace that is orthogonal
to the signal eigenvector e = [1 ejω ej2ω ej(z−1)ω]T, and eU denotes the complex-
conjugate transpose; z is the dimension of space spanned by P(n). It is worth
emphasizing that PMUSIC(ejω) in (4.9) does not relate to any real power spectrum;
rather, the only purpose of this pseudo-spectrum is to generate peaks whose fre-
quencies correspond to those of the dominant frequency components. This feature
makes the MUSIC approach interesting to develop an equivalent model based on
dominant modes.
For a given signal of interest and using (4.7)–(4.9), the model eigenvalues can be
calculated. By knowing the eigenvalues and because the impulse response is the
inverse Laplace transform of eigenvalues, one could represent a signal of interest
with a predefined model of (4.6). For this purpose, suppose the impulse response of
system is I(t); for the input signal x(t), i.e. x(t) = Tm(t) in (4.6), one could write y(t),
i.e. ω, as:
∞
y t =x t I t ejωt dt 4 10a
t=0
where
I t = y t ∗PS 4 10b
and PS in (4.10b) is the pseudo-spectrum of the signal. Equation (4.10b) reveals
that I(t) is obtained from the convolution of y(t) and PS. In the modeling procedure
and by measuring the output response of the system y(t) and by knowing I(t), the
problem of interest is to calculate x(t) in (4.10a). By calculating Tm (t), the ESS
could be replaced by the SG model of (4.6). Using this framework, angular speed
ω in (4.10a) is defined based on the dominant frequency components of the
pseudo-spectrum in (4.9). Figure 4.8 gives a schematic illustration of this model.
In this plot, Figure 4.8a illustrates the process of virtual inertia emulation using the
battery ESS, while Figure 4.8b describes a simplified block diagram representation
of the equivalent model. The input to the control algorithm is the frequency at the
connection point of the inverter fcp, and Pin represents the grid injected power.
4.3.1.3 Experimental Verification

The effectiveness of (4.6)–(4.10b) to represent the ESS behavior is illustrated using
the existing battery ESS at the University of Kurdistan Micro-Grid (UOK-MG).
Figure 4.9 shows a three-phase diagram representation of the UOK-MG.
Figure 4.10 shows the ESS and the main grid power variation behavior, i.e. Pin in
Figure 4.8, recorded for the UOK-MG. As shown in Figure 4.10, event 1 triggers the
charging process of the ESS in response to deviations from the minimum SOC.
(a) Inverter
Pin
DC
Energy Storage
DC
Power grid
+ fCP
Control
–
Algorithm
(b)
y(t)
Pseudo- I(t) Tm(t)

spectrum (PS) Eqn. (4.10b) Eqn. (4.10a)
Pin H
ejωt
Max
ω
(PS)
Figure 4.8 Block diagram representation of the proposed modeling process: (a) virtual
inertia emulation mechanism and (b) the proposed equivalent model; Pin, y(t), and fcp
represent the grid injected power, the frequency deviation of the ESS, and the frequency at
the point of connection of the ESS, respectively.
The main grid power deviation during the charging process in Figure 4.10 is uti-
lized to calculate the pseudo-spectrum of Figure 4.11 which, in turn, is used to
estimate the dominant frequency components in (4.9). Setting the frequency devi-
ation of the ESS in Figure 4.12 as y(t) and the pseudo-spectrum of Figure 4.11 as PS
for (4.10b) gives
s
δ = ωs − ω0
4 11
0 53ωs = 1 − e − 0 38t C − T e t
where C is a constant value equal to the DC term in (4.9).

In (4.11), note that the constant 0.38 represents the dominant frequency of the
pseudo-spectrum of Figure 4.11, which can be introduced into (4.10b) to calculate
the 0.53 s inertia constant. Figure 4.12 shows the effectiveness of the equivalent
Main Grid Static Switch
SS
SS SS
AC load
SS SS
ESS
DC bus SM
IC
SS SS SS
Controllable AC load
C
C
D
D
C
C
D
+
DC Load
–
WT ESS PV
Figure 4.9 Three-phase schematic representation of the UOK-MG.
1
Grid
3 ESS #1
P [KW]
1.5
0
–1.5
0 3 6 9 12 15
Time [s]
Figure 4.10 The ESS and grid experimental dynamic responses.
model (4.11) to approximate the inertial response behavior of the ESS. To exactly
mimic the frequency behavior of ESS using (4.11), the oscillatory behavior of the
adopted model can be removed using a 20-sample rolling-averaging window. This
approach averages the long-term oscillations, and hence, mitigates the oscillatory
behavior beyond the inertial response horizon. Results in Figure 4.12 show that the
dynamic behavior of the ESS, especially in the inertial response horizon, can be
10
Power [dB]
–10
–20
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalized Frequency
Figure 4.11 The pseudo-spectrum estimation via MUSIC.
ESS frequency response

Equivalent model response of (4.11)
50.15 Equivalent averaged-based model response of (4.11)
f [Hz]
50
49.85
0 6 12 18 24 30
Time [s]
Figure 4.12 Comparison of frequency response computed from the experiment and the
equivalent frequency response models.
approximated by the SG model. It should be emphasized that while a conventional

SG is slower and less flexible compared to the ESS, the developed dynamical equiv-
alent model of the ESS in the grid-connected mode is not only affected by the fast-
inherent features of the ESS but also it is significantly influenced by the external
zone, i.e. the host grid, features.
4.3.1.4 Economic Modeling

This section formulates the ESS placement problem as an optimal techno-
economic problem.
Costs of Energy Storages and Technologies

There are two main approaches for assessing the cost of ESS technologies: (i) total
capital cost (TCC), and (ii) life cycle cost (LCC) [61]. In the TCC approach, all
terms associated with the purchase, installation, and delivery of the ESS units,
including the power conversion system (PCS) costs (CPCS), costs of ESS (Cstor),
and balance of plant (BOP) cost (CBOP) can be represented as:
Ccap = CPCS + C BOP + Cstor × t ch $ kW 4 12
where tch is the charging/discharging time. The balance of the ESS, known as the
BOP, includes site wiring, interconnecting transformers, and other additional
ancillary equipment and is measured on a $/kW basis [62]. However, the LCC
is a more common metric to evaluate and compare different ESS technologies.
The annualized LCC is formulated according to (4.13) which considers operation
and maintenance costs (Co&M,a), replacement cost (CR,a), and annualized TCC:
CLCC,a = Ccap,a + COM,a + CR,a $ kW yr 4 13
in which:
T
i 1+i
CRF = T
4 14
1+i −1
Ccap,a = TCC × CRF $ kW yr 4 15
COM,a = CFOM,a + C VOM,a × ncycle × t ch $ kW yr 4 16
r − kt CR × t ch
CR,a = CRF × 1+i × $ kW yr 4 17
k=1 ηsys
where CRF, i, T, r, t, and ηsys are the capital recovery factor, interest rate and the
lifetime, the number of substitutions in a lifetime, the replacement period, and the
overall efficiency, respectively; CFOM,a and CVOM,a define the fixed and variable
operation and maintenance costs. The subscript “a” stands for “annualized” costs.
Formulation of the Objective Function and Constraints

Equation (4.13) specifies the annual cost per kilowatt of the installed ESS in compli-
ance with the lifetime. However, for optimal placement of virtual inertia, it is neces-
sary to rewrite the cost function in (4.13) according to the amount of virtual inertia.
The synchronous inertia constant M is defined as the ratio of stored kinetic
energy to the rated apparent power of the system as:
0 5J VI ω2
M= 4 18
Sbase
where JVI, ω, and Sbase are the moment of inertia, angular velocity, and rated
apparent power, respectively. Since the stored energy in ESSs is usually expressed
in volt-ampere hour (VAhESS), it is needed to express the associated value in terms
of Joule. Considering a unity power factor, (4.18) can be rewritten as:
M ESS Sbase 3600 VAsecESS

KW ESS = KVAESS = M ESS = 4 19
3600 s Sbase
This equation gives the average hourly power that can be injected/absorbed to/
from the grid by the ESS. By substituting (4.19) into (4.13), one could write the
optimization problem as:
nESS
M ESS Sbase
minimize F M ESS
i = CLCC,a i × i
4 20a
M ESS
i i=1
3600 sec
st RoCoF i ≤ RoCoF max 4 20b
Δ f nadir i ≤ Δ f nadir max 4 20c
SOC min ≤ SOC i ≤ SOC max 4 20d
where nESS is the number of ESSs. Moreover, the SOC should remain within an
appropriate range which is addressed in (4.20d). The SOC can be calculated as fol-
lows [63, 64]:
Δt
0 ζp t dt
SOC Δt = SOC 0 − 4 21a
E ESS,rated
where
ζc p t <0
ζ= 1 4 21b
p t >0
ζd
and p(t) is battery power which gets negative values for the charging procedure
and positive values for the discharging period; EESS,rated, Δt, ζ c, and ζ d are the nom-
inal energy capacity, charge/discharge time, and charging and discharging effi-
ciencies of the battery, respectively.
Determining the Bounds of Constraints

Constraints (4.20b) and (4.20c) explain that the optimization problem (4.20a)
enforces the RoCoF and frequency nadir in all areas to be less than the standard
values. These terms make the optimization problem difficult to deal with as it
depends on the dynamical indices. Generally, it is common to specify the
lower/upper bounds based on different criteria, including the capacity of equip-
ment or budget. Therefore, the problem of interest here is to rewrite the upper
and lower bounds of (4.20b) and (4.20c) in terms of the emulated inertia, i.e. MiESS.
Rate of Change of Frequency The RoCoF is a meaningful criterion that measures

the rate of frequency change following a large loss-of-generation event. Large
RoCoF values indicate that less time is available for a system operator to arrest
frequency decline. Typically, a time interval from 100 ms to 2 s is defined to
measure the RoCoF [65, 66]. The ENTSO standard [65] explains that RoCoF is
allowed to get a value between 0.5 and 1 Hz/s.
To obtain dynamic frequency indices based on lower bounds of inequality con-
straints in (4.20b), the RoCoF can be defined in terms of the classical swing equa-
tion of (4.6) as [67]:
dΔf t
2M = ΔPm t − ΔPL t − ΔPtie t 4 22
dt
where ΔPm(t), ΔPL(t), and ΔPtie(t) represent mechanical power, electrical power,
and tie-line power changes, respectively. Considering the definition of RoCoF, one
could write
ΔPm t − ΔPL t − ΔPtie t
RoCoF = 4 23
2M
The Taylor series expansion of (4.23) about the independent variables of H, ΔPm,
ΔPL, and ΔPtie gives
∂RoCoF i ∂RoCoF i ∂RoCoF i ∂RoCoF i
ΔRoCoF i = ΔΔPmi + ΔΔPLi + ΔΔPtiei + ΔM i
∂ΔPmi ∂ΔPLi ∂ΔPtiei ∂ΔM i
1 −1 −1 − ΔPmi − ΔPLi − ΔPtiei
= ΔΔPmi + ΔΔPLi + ΔΔPtiei + ΔM i
2M i 2M i 2M i 2M i 2
4 24
Since the slow inherent dynamics of interest is only given by the last term in
(4.24), the other terms can be neglected. Inserting (4.23) into (4.24) gives
ΔM i
RoCoF i − = ΔRoCoF i 4 25
Mi
It follows that the minimum inertia which guarantees that the RoCoF remains
within the permitted range can be calculated as:
ΔRoCoF i, max
i, min = ΔM i, min = M i −
M 'ESS
RoCoF i
4 26
RoCoF i, max − RoCoF i
M 'ESS
i, min = Mi −
RoCoF i
where RoCoFi,max, is the maximum allowable RoCoF, and M ESSi,min represents the
minimum required inertia that should be emulated by the battery ESS in area i,
which complies with the RoCof. It equals to the difference between the desired
inertia to enforce the system to follow the standards and the present inertia con-
stant, i.e. ΔMi,min.
Frequency Nadir The minimum instantaneous frequency following a distur-

bance, known as frequency nadir, is mainly dependent on the system total inertia
and the capability of the power resources to provide primary frequency response.
According to the NERC and the Union for the Coordination of the Transmission of
Electricity (UCTE) standards [68, 69], the allowed minimum frequency in a power
system during normal operation is 800 mHz. Taking the time dependence of the
governor response into account, one can approximate the frequency nadir after
a system event as [43]:
ΔPL + ΔPtie 2 T d
Δ f nadir = 4 27
4MR
where R is the extra power received from the governor and Td is the response time
of the governor.
In deriving (4.27), it is assumed that the mechanical power through the governor
increases as a linear function of time with the steady gradient R/Td [70, 71]. While
this is a conservative assumption, Great Britain and Ireland practices show that
this is the case for the power increment within 5 and 10 seconds (Td), respectively,
following a contingency [72]. Applying Taylor’s expansion to (4.27) gives
∂Δ f nadir,i ∂Δ f nadir,i ∂Δ f nadir,i
ΔΔ f nadir,i = ΔΔPLi + ΔΔPtie,i + ΔM i
∂ΔPLi ∂ΔPtie,i ∂ΔM i
ΔPLi + ΔPtiei T di ΔPLi + ΔPtiei T di
= ΔΔPLi + ΔΔPtiei
2M i Ri 2M i Ri
− ΔPLi + ΔPtiei 2 T di
+ ΔM i
4M 2i Ri
4 28
Following the same procedure as that in (4.26), one could rewrite (4.28) in
the form:
ΔM i
Δ f nadir,i − = ΔΔ f nadir,i = Δ f nadir,i − f 0 4 29
Mi
The minimum inertia, i.e. M ''ESS

i, min , which guarantees frequency nadir to remain
within the permitted range is calculated by:
ΔΔ f nadir i, max f nadir max − Δ f nadir i
i, min = ΔM i, min = M i −
M ''ESS = Mi −
Δ f nadir i Δ f nadir i
4 30
To simultaneously satisfy both the frequency nadir and the RoCoF standards,
the lower bound for virtual inertia in the optimization problem for each area is
selected as the maximum value of (4.26) and (4.30), namely:
M ESS 'ESS ''ESS

i, min = max M i, min , M i, min 4 31
Moreover, the overall system inertia has a direct impact on the frequency indi-
ces. This means that some considerations should be made regarding overall system
inertia and, consequently (4.26) and (4.30) will be completed by adding a new
equality constraint. For this purpose, the frequency of the overall center of inertia
(COI), which should satisfy strict frequency standards, can be used to determine
the overall amount of inertia in the system as:
Δ f COI
M COI = Q = M 61 5 4 32
f COI
where ΔfCOI and fCOI represent the frequency deviation and frequency of the sys-
tem, without ESS, after the fault, respectively. Details on how to derive (4.32) are
given in Chapter 3. Formally, Equation (4.32) gives the required amount of inertia
constant which guarantees acceptable frequency dynamics of the COI. Of note that
Q would be realized by adding the emulated inertia of ESSs to the conventional
SGs inertia. Accordingly, the optimization problem (4.20) can be rewritten as:
nESS
M ESS Sbase
minimize F M ESS
i = C LCC,a i i
4 33a
M ESS
i i=1
3600 sec
st M COI = Q 4 33b
M ESS
i, min ≤ M ESS
i ≤ M ESS
i, max 4 33c
SOC min ≤ SOCi ≤ SOCmax 4 33d
where the dynamic inequality constraints (4.20b) and (4.20c) are reformulated as
the algebraic inequality constraint (4.33c) in terms of the inertia constant. This
increases the simplicity and speed of the calculations.
4.3.1.5 Simulation and Results

Linear System
As the first motivating example, a linearized model of a three-area power system is
used to assess the efficiency of the proposed formulation. The block diagram of
each area is shown in Figure 4.13.
Firstly, a 0.2 per unit load disturbance is applied in areas 1 and 3. As the first
scenario, the required virtual inertia is calculated only based on (4.33b) and arbi-
trary realized through the model of (4.11) in area 1. For simulation purposes, it is
assumed that the ESS would be triggered upon the inception of the fault. Compar-
ison of frequency dynamics for the system with and without virtual inertia reveals
that while inertia emulation improves the frequency nadirs of areas 1 and 2, the
frequency performance of area 3 deteriorates. This, in turn, numerically justifies
the need for optimal inertia placement. Within this framework, (4.33a)–(4.33d)
may be written as:
1 ΔPLi
βi
Ri
ΔPgi ΔPmi
+ – –
+ 1 1 + 1 Δfi
K (s)
+ ACEi ΔPCi 1 + sTgi 1 + sTti – Di + 2(Hi + hi)s
narea
Controller Governor Turbine Rotating mass
and load ∑ Tij
j=1
j≠i
ΔPtie,i 2π/
+
s
narea –
∑ TijΔfj
j=1
j≠i
Figure 4.13 Block diagram representation of control area i. βi, Ri, Tgi, Tti, and Di are
frequency bias, droop characteristic, governor time constant, turbine time constant, and
damping property, respectively.
nESS
M ESS Sbase
minimize F M ESS
i = CLCC,a i i
4 34a
M ESS
i i=1
3600 sec
st M COI = 0 053 4 34b
0 0129 ≤ M ESS
1 4 34c
0≤ M ESS
2 4 34d
0 0225 ≤ M ESS
3 4 34e
30 ≤ SOCi ≤ 80 4 34f
where for instance, the minimum inertia of area 1 in (4.34c) is calculated based on
(4.26), (4.30), and (4.31) as:
1 + 1 1870
M 'ESS
1, min = 0 08335 × − = 0 0129
− 1 1870
M ''ESS
1, min = 0 4 35
M ESS
1, min = max 0, 0 0129 = 0 0129
The parameters used in (4.34a) are summarized in Tables 4.1 and 4.2.
The results obtained, using a simple genetic algorithm (GA) with 0.05 and 0.8
mutation and crossover coefficients, respectively, from the optimization of
(4.34a) are shown in Table 4.3. To further assess the efficiency of the formulation,
Table 4.3 compares the results with those obtained in [30, 73]. Comparison results
justify the fact that the dynamic behavior of ESS can significantly affect the
optimal placement problem.
Table 4.1 Economical parameters related to the optimization problem [42, 43].
Parameter Value Parameter Value
i (%) 8 CFOM,a ($/kw-yr) 10

CPSC ($/kw) 200 CVOM,a ($/kwh) 5
CBOP ($/kw) 50 R 2
Cstor ($/kw) 300 t (yr) 6
CR ($/kw) 300 ηsys (%) 75
Table 4.2 Technical parameters related to the optimization problem [41].
Parameter Value Parameter Value
Sbase (MVA) 1000 PESS (MW) 1

RoCoFmax (Hz/s) 1 EESS, rated (MVAh) 0.25
Δfnadir,max (Hz) 0.8 ηd = ηc(%) 75
SOC(0) 0.5 t (year) 15
SOCmin 0.3 tch (hours) 0.25
SOCmax 0.8 ncycle 1000
Table 4.3 Optimization results in three-area system.
Method M1ESS M2ESS M3ESS F(MiESS)
Formulation (4.33) 0.016 0 0.037 2.5916

Ref. [30] 0.023 0.016 0.022 3.0942
Ref. [73] 0.012 0.009 0.038 2.8735
Further, Figure 4.14 compares the frequency behavior and RoCoF of generators 1,
2, and 3 for three cases of interest: (i) with virtual inertia and according to the for-
mulation of (4.33a)–(4.33d), (ii) with virtual inertia and according to [73], and
(iii) with virtual inertia and according to Ref. [30]. It can be seen that while
frequency traces of Refs. [30, 73], and the formulation of (4.33a)–(4.33d) meet the
RoCoF and frequency nadir standards, (4.33a)–(4.33d) results in a less ESS capacity.
Also of interest, modal analysis of the results, as explained in Table 4.4, shows the
efficiency of (4.33a)–(4.33d) in comparison with that of Refs. [30, 73].
(a)
0
Δf1[Hz]
–0.2 Ref. [73]

Proposed method
Ref. [30]
–0.4
–0.6
0 2 4 6 8 10 12 14 16 18 20
0
Δf2[Hz]
–0.2
–0.4
–0.6
0 2 4 6 8 10 12 14 16 18 20
0.2
0
Δf3[Hz]
–0.2
–0.4
–0.6
0 2 4 6 8 10 12 14 16 18 20
Time (s)
(b)
1
RoCoF1 [Hz/sec]
0.5
0
–0.5
–1
0 2 4 6 8 10 12 14 16 18 20
1
RoCoF2 [Hz/sec]
0.5
0
–0.5
–1
0 2 4 6 8 10 12 14 16 18 20
1
RoCoF3 [Hz/sec]
0.5
0
–0.5
–1
0 2 4 6 8 10 12 14 16 18 20
Time (s)
Figure 4.14 (a) Frequency behaviors and (b) RoCoF of generators 1 to 3 of three-area power
system for three cases: proposed formulation [30, 73].
Table 4.4 Modal analysis of the systems with virtual inertia.
Method Proposed method Ref. [30] Ref. [73]
Damping 0.2762 0.1265 0.2634
Results show that a lower emulated virtual inertia based on (4.33a)–(4.33d) not
only decreases the cost function but also provides better performance in terms of
enhanced damping.
Tables 4.3 and 4.4 and Figure 4.14 suggest that with virtual inertia, the results of
Refs. [30, 73] seem to fulfill the constraints and are almost the same as the results
in (4.33a)–(4.33d). This could be justified through the fact that the set of generator
buses for small systems includes a few members to be considered as candidates for
ESS installation. Therefore, different algorithms may differ a bit from the capacity
point of view rather than the location which in turn causes a negligible difference
in the results. To further assess the effects of virtual inertia on the frequency and
transient stabilities, two nonlinear systems are used in what follows.
Two-Area Power System In this section, the two-area power system, shown in
Figure 4.15, is considered to further demonstrate the efficiency of the optimal
placement formulation. Modeling considerations are essentially those described
in [74]; all the generating units are modeled with sixth-order synchronous
machine models with excitation systems.
Area 1 Area 2
1 10 20 3 101 13 120 11
Gen3
Gen1
110
Gen2 Gen4
4 12
2
14

The test scenario of interest is the outage of generator G4 in the first second of
the simulation. The lower bounds of virtual inertia are calculated according to
(4.34a)–(4.34f) and (4.35) as:
nESS
M ESS Sbase
minimize F M ESS
i = CLCC,a i i
4 36a
M ESS
i i=1
3600 sec
st M COI = 4 973 4 36b
0 3151 ≤ M ESS
1 4 36c
0 3881 ≤ M ESS
2 4 36d
0 ≤ M ESS
3 4 36e
30 ≤ SOCi ≤ 80 4 36f
in which the equality constraint (4.36b) reveals that
M COI + M ESS = 4 973 M ESS = 4 973 − 4 = 0 973 4 37

Solving (4.36) leads to the optimum results of Table 4.5. The results are also
compared with those of Refs. [30, 73]. Also of interest, Table 4.6 compares the
frequency stability indices for different approaches. The results demonstrate the
high efficiency of (4.33a)–(4.33d) to optimally allocate virtual inertia in the system.
The results for the system without virtual inertia suggest that while the fre-
quency nadirs exhibit acceptable performance, the RoCoFs exceed the standard
value for some of the generators. The optimal placement of virtual inertia returns
the generators with undesired frequency dynamics to the normal region. The
efficiency of (4.33a)–(4.33d) is further assessed by time-domain (T-D) simulations
of Figure 4.16.
The effect of emulated virtual inertia on transient stability can be assessed using
a simple power angle-based stability index η [46]:
360 − δmax
η= 4 38
360 + δmax
Table 4.5 Optimization results in two-area system.
Method M1ESS M2ESS M3ESS F(MiESS)
Proposed method 0.402 0.567 0.004 4.6705

Ref. [30] 0.116 0.332 0.376 6.3104
Ref. [73] 0.212 0.315 0.316 5.7891
Table 4.6 Frequency indicators of two-area system before and after the application of
optimal inertia values.
With VI Proposed
Without VI method With VI [30] With VI [73]
RoCoF Δfnadir RoCoF Δfnadir RoCoF Δfnadir RoCoF Δfnadir

Gi (Hz/s) (Hz) (Hz/s) (Hz) (Hz/s) (Hz) (Hz/s) (Hz)
1 1.187 0.172 0.989 0.143 0.988 0.126 0.973 0.116

2 1.240 0.160 0.989 0.134 0.981 0.112 0.961 0.162
3 0.706 0.267 0.713 0.254 0.730 0.200 0.786 0.198
Ref. [30]
1.002 Proposed method
Δf1[Hz]
Ref. [73]
0.998
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
1.002
Δf2[Hz]
0.998
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
1.004
1.002
Δf3[Hz]
1
0.998
0.996
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time [s]
Figure 4.16 Frequency response of generators 1, 2, and 3 of the two-area power system for
three cases: proposed formulation [30, 73].
where δmax is the maximum angle separation of any two generators in the system.
It is noteworthy that during severe faults, most ESSs, if remain connected and con-
tinue to inject active power, get saturated and cannot follow the frequency prop-
erly. It is noted here that (4.38) relies on a conservative assumption that the fault is
not a severe one, which causes ESSs to be disconnected from the grid. Moreover,
Table 4.7 Transients stability assessment.
Method δmax η
Proposed method 11.02 0.94

Ref. [30] 20.8 0.89
Ref. [73] 16.63 0.91
the saturation of ESSs in response to sever faults is neglected. Table 4.7 demon-
strates the better performance of the proposed method in comparison with those
of Refs. [30, 73].
New York New England System The NYNE test system is used to further illustrate
the efficiency of the proposed algorithm for large-scale power systems. A single-
line diagram of the system, showing major coherent areas and their interconnec-
tions, is shown in Figure 4.17.
Five different contingency scenarios, including tripping of major generating
units and load rejection, are considered. Table 4.8 compares the results of the
optimization problem, given by (4.33a)–(4.33d), with those of Refs. [30] and
14 Area 3 Area 1
66 8
41 40 48 47 60
1
25 26 29
53
28
10 1 2 61
27
62 9
30
31 3
11
63 18 17
16 24
46 38 32
33 15 22
Area 5 34
9
4
21
19 58
49 14
35 6
42 36 5 20
8 6 12 56 23
67 64
12
57 4
15 45
51 5
59
50 44 37 11 13 7
7 54
2
65 10
52 43 13
39
55
68
3
Area 4 16 Area 2
Figure 4.17 Single-line diagram of NYNE test system showing coherent areas and their
interconnections.
Table 4.8 Optimization results in NYNE test system.
M1ESS M2ESS M3ESS M4ESS M5ESS F(MiESS)
1 Proposed method 0.412 0.432 0.313 0.092 0.111 9.2141

Ref. [30] 0.506 0.332 0.376 0.201 0.112 12.4031
Ref. [73] 0.378 0.453 0.306 0.115 0.098 11.0817
2 Proposed method 0.341 0.513 0.209 0.101 0.098 10.8601
Ref. [30] 0.340 0.712 0.301 0.113 0.160 13.0012
Ref. [73] 0.300 0.798 0.251 0.098 0.161 12.3140
3 Proposed method 0.474 0.261 0.160 0.261 0.007 9.7516
Ref. [30] 0.596 0.298 0.267 0.271 0.088 11.2113
Ref. [73] 0.314 0.351 0.294 0.314 ~0 10.0087
4 Proposed method 0.169 0.203 0.617 0.135 0.072 9.8617
Ref. [30] 0.132 0.512 0.694 0.196 0.209 10.5103
Ref. [73] 0.100 0.374 0.687 0.096 0.101 10.0102
5 Proposed method 0.613 0.032 0.116 0.076 ~0 9.0412
Ref. [30] 0.743 0.215 0.402 0.031 ~0 9.9731
Ref. [73] 0.412 0.354 0.391 0.116 ~0 9.4019
[73]. The results demonstrate the efficiency of (4.33a)–(4.33d) to enhance the

frequency dynamics with minimum cost. Furthermore, Figure 4.18 shows the allo-
cation of virtual inertia among the PV buses of the system.
Results show that the proposed optimization scheme works better and more effi-
ciently for larger areas. This can be understood by noting that the set of generator
buses for large areas includes many members to be considered as candidates for
the ESS placement. As a result, there are many possibilities for placing of the ESSs.
Figure 4.19 compares the frequency dynamics of the system, in response to the
outage of generator 1 (Scenario 1), for (4.33a)–(4.33d), and Refs. [30, 73]. It should
be noted that while there are negligible deviations between the traces in
Figure 4.19, significant differences between the cost functions justify the efficiency
of (4.33a)–(4.33d).
Also of interest, the efficiency of (4.33a)–(4.33d) to enhance transient stability is
shown in Table 4.9, using (4.38). Table 4.9 shows that the appropriate placement of
virtual inertia in the system, considering the dynamical behavior of ESS, could also
improve transient stability. This could be justified noting that recent research, for
example in Ref. [30], relies on the quasi-steady-state phasors for voltages and cur-
rents in transient stability assessment. In other words, they consider constant
0.7
Scenario 1
0.6
Scenario 2
0.5 Scenario 3
Virtual Inertia
Scenario 4
0.4
Scenario 5
0.3
0.2
0.1
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Generator number
Figure 4.18 Virtual inertia allocation for NYNE test system.
Table 4.9 Optimization results in New England system.
Method η12 η13 η14 η15 η25 η35
1 Proposed method 0.92 0.87 0.91 0.89 0.90 0.89

Ref. [30] 0.85 0.84 0.91 0.90 0.85 0.87
Ref. [73] 0.87 0.86 0.91 0.89 0.88 0.88
2 Proposed method 0.98 0.83 0.92 0.93 0.95 0.95
Ref. [73] 0.94 0.82 0.91 0.90 0.91 0.92
Ref. [30] 0.91 0.84 0.90 0.88 0.89 0.95
3 Proposed method 0.87 0.90 0.90 0.89 0.92 0.92
Ref. [73] 0.85 0.88 0.90 0.88 0.88 0.88
Ref. [30] 0.88 0.85 0.87 0.87 0.87 0.88
4 Proposed method 0.96 0.93 0. 93 0.93 0.93 0.93
Ref. [73] 0.96 0.90 0.91 0.93 0.92 0.89
Ref. [30] 0.95 0.94 0.93 0.93 0.91 0.85
5 Proposed method 0.94 0.85 0.92 0.92 0.91 0.96
Ref. [73] 0.89 0.86 0.90 0.90 0.89 0.94
Ref. [30] 0.93 0.87 0.87 0.87 0.88 0.91
1
Ref. [30]
Proposed method
0.998 Ref. [73]
f[Hz]
0.996
0.994
0 5 10 15
Time [s]
Figure 4.19 Frequency responses of NYNE test system for different approaches.
(nominal) frequency in defining the system impedances which is far from realistic
for systems with high penetration of inverter-based ESSs. This point is successfully
addressed in (4.33a)–(4.33d) by explicitly representing the dynamic behavior of
ESSs in the problem formulation.
4.3.1.6 Sensitivity Analysis

In this section, sensitivity analyses are conducted to understand the effect of oper-
ation conditions, including variations of fault magnitudes, operating point, and
annualized LCC on the optimization problem. For this purpose, (4.33a) is used
to calculate the sensitivity of the cost function to operation condition as:
nESS
M ESS Sbase ∂F M ESS
i
F M ESS + ΔF M ESS = i
CLCC,a i + ΔΔPL
i i
i=1
3600 sec ∂ΔPL
∂F M ESS ∂F M i
ESS
+ i
ΔΔPm + ΔCLCC,a i 4 39
∂ΔPm ∂CLCC,a i
According to (4.24) and (4.28), one could write the sensitivity matrix as:
∂RoCoF i ∂RoCoF i
∂M 'ESS
i, min ∂ΔPLi ∂ΔPmi ΔΔPLi
= 4 40
∂M ''ESS
i, min
∂Δ f nadir,i ΔΔPmi
0
∂ΔPLi
Substituting (4.40) in (4.39) gives
∂F M ESS ∂M ESS ∂F M ESS ∂M ESS

ΔF M ESS = i i
ΔΔPL + i i
ΔΔPm
i
∂M i
ESS ∂ΔPL ∂M i
ESS ∂ΔPm
nESS
Sbase
= CLCC,a i
i=1
3600 sec
∂RoCoF i ∂RoCoF i ∂ΔF nadir,i
× max ΔΔPL + ΔΔPm , ΔΔPLi
∂ΔPLi ∂ΔPmi ∂ΔPLi
nESS
M ESS
i Sbase
+ ΔCLCC,a i 4 41
i=1
3600 sec
Table 4.10 Sensitivity analysis.
Scenario Cost function of (4.33a)–(4.33d) Cost function of (4.41)
2 9.7516 9.5913
3 9.8617 9.9302
The effectiveness of (4.41) is now assessed for the NYNE system. For this pur-
pose, the outage of generator 1 in area 1, i.e. Scenario 1, is considered as the base
case for sensitivity analysis. The cost functions for the outage of generator 7 in area
1, i.e. Scenario 2, and generator 11 in area 2, i.e. Scenario 3, are calculated using
(4.41). Table 4.10 compares the exact results of (4.33a)–(4.33d) with those of (4.41)
which justify the effectiveness of the sensitivity analysis (4.41).
For uncertainty analysis, the equality constraint (4.33b) is represented in the
objective function (4.33a) as:
nESS
M ESS
i Sbase
minimize F M ESS
i = CLCC,a i + β M COI − Q 4 42
M ESS
i i=1
3600 sec
where β is arbitrary chosen high to enforce the results to follow the equality con-
straint (4.33b). Considering parametric uncertainty for the inertia constant MCOI,
one could write (4.42) as:
nESS
M ESS Sbase
minimize F M ESS
i = CLCC,a i i
+ β M COI + γ − Q
M ESS
i i=1
3600 sec
nESS
M ESS Sbase
minimize F M ESS
i = CLCC,a i i
+ β M COI − Q + βγ
M ESS
i i=1
3600 sec
4 43
where γ is expressed as a percentage of MCOI. To deal with uncertainty analysis, a

simple interval approach is utilized. This approach assumes that the uncertain
parameters take value in a specified interval. It could be reinterpreted as the prob-
abilistic modeling with a uniform probability density function (PDF). In this
method, the upper and lower bounds of the uncertain inertia parameter are
defined. The aim is to find the lower and upper bounds of the objective function
[75]. Using the proposed framework, assume that the maximum variation of the
inertia constant is limited to ±5%. This means that the interval of interest can be
defined as:
γ = M COI − 0 05 M COI , M COI + 0 05 M COI 4 44

ESS
which, in turn, requires that F(Mi ) takes the values:
F M ESS
i = 3 8298, 5 5112 4 45
with a uniform PDF.
4.3.2 Small-Signal Stability

4.3.2.1 Objective Function
The optimal placement of ESS to mitigate undesired frequency dynamics in low-
inertia power grids is discussed in Section 4.3.1. In this section, same formulation
would be derived from small-signal stability point of view.
The conducted analysis in Chapter 3 indicates that the damping ratio of critical
modes of system are detrimentally affected by the increasing of renewable sources
penetration, i.e. decreasing inertia. Critical modes of a system are those within the
frequency range of 0.01–2 Hz and damping of less than 10%.
To optimally place ESSs in the system according to small-signal stability consid-
erations, the same formulation as in (4.33a)–(4.33d) with some modifications given
in (4.46a)–(4.46c) is utilized:
nESS
M ESS Sbase
minimize F M ESS
i = C LCC,a i i
4 46a
M ESS
i i=1
3600 sec
st M ss
COI = K 4 46b
SOCmin ≤ SOCi ≤ SOCmax 4 46c
where superscript “ss” stands for small signal. Of note that, M ss COI in (4.46b)
includes two terms: (i) overall system inertia (M) and (ii) provided virtual inertia
by ESS (MESS). In (4.46b), K is determined on which the critical mode of interest be
characterized by the minimum damping ratio of 10%. For this purpose, a relation-
ship between the inertia and the damping ratio of the desired mode would be
derived. In this way, the trend of changing the damping ratio in response to the
variation of the inertia is used as the input for the curve fitting tool. Accordingly,
one could write
ςi = f M 4 47
By setting the left-hand side of (4.47) to 10%, the required inertia, i.e. K in (4.46b),
which guarantees the desired damping ratio would be calculated.
To further proceed with the derivation of the objective function, the equality
constraint (4.46b) is represented in the objective function (4.46a) as:
nESS
M ESS
i Sbase
minimize F M ESS
i = CLCC,a i COI − K
+ α M ss 4 48
M ESS
i i=1
3600 sec
where α is arbitrary chosen high to enforce the results to follow the equality
constraint (4.46b). On the other hand, as the damping ratio of a specific mode
is influenced by the generating units according to their participation factors, it
seems that the participation factors should be reflected in the objective function
(4.48). More precisely, the machines with higher participation factors in the stud-
ied mode should contribute to the minimization problem (4.48) with lower
weights. In this way, one could rewrite (4.48) as:
nESS
M ESS Sbase 1 ESS
minimize F M ESS
i = C LCC,a i i
+α M+ M −K
M ESS
i i=1
3600 sec i
pf i i
4 49
where pfi is the participation factor of generator i in the critical mode of interest.

The efficiency of (4.49) is examined on the NYNE benchmark of Figure 4.17.
Table 4.11 reports the eigenvalues of the system with no MGs penetration level
(details on how to obtain the results are given in Chapter 3). This analysis is per-
formed using the small-signal analysis toolbox (SSAT) which is a part of the DSA
Tools software package.
Simulation results reveal that by penetration of 52% MGs in the system, the
damping ratio of the third mode, as the worst mode detrimentally affected, reaches
to 0.42%. Therefore, the main aim of (4.49) is to allocate ESSs in the generator
busses of the system on which the damping ratio increases to 10%. To deal with
(4.49), the derivation of participation factors is mandatory. The participation fac-
tors of various machines participating in the third studied mode, i.e. −0.3590 +
3.7108j, are presented in Figure 4.20. As seen from the figure, the generator num-
ber 15 has the highest participation factor in this mode.
Table 4.11 Critical modes of the base system.
Mode Real part Imaginary part Frequency Damping ratio
1 −0.7018 1.9710j 0.3137 33.543

2 −0.1184 3.2666j 0.5199 3.6236
3 −0.3590 3.7108j 0.5906 9.6290
4 −0.1657 4.8915j 0.7785 3.3851
0.9
0.8
0.7
Participation Factor
0.6
0.5
0.4
0.3
0.2
0.1
0
0 2 4 6 8 10 12 14 16
Generator Number
Figure 4.20 Participation factor of the machines participating in the critical mode
detrimentally affected by high MGs penetration level.
According to (4.49) and Figure 4.20, it is expected that buses related to genera-
tors number 15 and 16 should require a larger ESSs capacity. Figure 4.21 shows the
allocation of virtual inertia among the generator buses of the system which in turn
justify the effects of the participation factor on the considered objective function.
0.35
0.3
0.25
Virtual Inertia
0.2
0.15
0.1
0.05
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Generator Number

References 139
4.4 Summary
The advanced control schemes in modern power grids rely on inertia manipulation
in the system to enhance stability. The successful development of advanced control
schemes highly depends on the adequacy of the inertia provision sources. In the
present chapter, first, the VSG concept and its applications in power grids are
explained. Afterward, dispatchable inertia is optimally placed in the system accord-
ing to dynamical metrics to enhance system stability and dynamic performance.
References
New York: Wiley.
2. Bevrani, H., Ise, T., and Miura, Y. (2014). Virtual synchronous generators: a survey
and new perspectives. International Journal of Electrical Power & Energy Systems
54: 244–254.
3. Liu, J., Miura, Y., Bevrani, H., and Ise, T. (2016). Enhanced virtual synchronous
generator control for parallel inverters in microgrids. IEEE Transactions on Smart
Grid 8 (5): 2268–2277.
4. D’Arco, S. and Suul, J.A. (2013). Virtual synchronous machines—classification of
implementations and analysis of equivalence to droop controllers for microgrids.
In: Proceedings of the 2013 IEEE Grenoble Conference, 1–7. IEEE.
5. Cao, Y. et al. (2017). A virtual synchronous generator control strategy for VSC-
MTDC systems. IEEE Transactions on Energy Conversion 33 (2): 750–761.
6. Al-Tameemi, M., Liu, J., Bevrani, H., and Ise, T. (2020). A dual VSG-based M3C
control scheme for frequency regulation support of a remote AC grid via low-
frequency AC transmission system. IEEE Access 8: 66085–66094.
7. Hlaing, H.S., Liu, J., Miura, Y. et al. (2019). Enhanced performance of a stand-alone
gas-engine generator using virtual synchronous generator and energy storage
system. IEEE Access 7: 176960–176970.
8. Mishra, S., Pullaguram, D., Buragappu, S.A., and Ramasubramanian, D. (2016).
Single-phase synchronverter for a grid-connected roof top photovoltaic system. IET
Renewable Power Generation 10 (8): 1187–1194.
9. Ma, Y., Cao, W., Yang, L. et al. (2017). Virtual synchronous generator control of full
converter wind turbines with short-term energy storage. IEEE Transactions on
Industrial Electronics 64 (11): 8821–8831.
10. Wang, Y., Meng, J., Zhang, X., and Xu, L. (2015). Control of PMSG-based wind
turbines for system inertial response and power oscillation damping. IEEE
11. Li, Y., Xu, Z., and Wong, K.P. (2016). Advanced control strategies of PMSG-based
wind turbines for system inertia support. IEEE Transactions on Power Systems
32 (4): 3027–3037.
12. Lopes, L.A. (2014). Self-tuning virtual synchronous machine: a control strategy for
energy storage systems to support dynamic frequency control. IEEE Transactions
on Energy Conversion 29 (4): 833–840.
13. Suul, J.A., D’Arco, S., and Guidi, G. (2016). Virtual synchronous machine-based
control of a single-phase bi-directional battery charger for providing vehicle-to-grid
services. IEEE Transactions on Industry Applications 52 (4): 3234–3244.
14. Guan, M., Pan, W., Zhang, J. et al. (2015). Synchronous generator emulation
control strategy for voltage source converter (VSC) stations. IEEE Transactions on
Power Systems 30 (6): 3093–3101.
15. Aouini, R., Marinescu, B., Kilani, K.B., and Elleuch, M. (2015). Synchronverter-
based emulation and control of HVDC transmission. IEEE Transactions on Power
Systems 31 (1): 278–286.
16. Dong, S., Chi, Y., and Li, Y. (2015). Active voltage feedback control for hybrid
multiterminal HVDC system adopting improved synchronverters. IEEE
Transactions on Power Delivery 31 (2): 445–455.
17. Li, C., Li, Y., Cao, Y. et al. (2018). Virtual synchronous generator control for
damping DC-side resonance of VSC-MTDC system. IEEE Journal of Emerging and
Selected Topics in Power Electronics 6 (3): 1054–1064.
18. Zhong, Q.-C. (2016). Virtual synchronous machines: a unified interface for grid
integration. IEEE Power Electronics Magazine 3 (4): 18–27.
19. Bevrani, H. (2009). Robust Power System Frequency Control. Springer.
20. Fathi, A., Shafiee, Q., and Bevrani, H. (2018). Robust frequency control of
microgrids using an extended virtual synchronous generator. IEEE Transactions on
Power Systems 33 (6): 6289–6297.
21. Jongudomkarn, J., Liu, J., Yanagisawa, Y. et al. (2020). Model predictive control for
indirect boost matrix converter based on virtual synchronous generator. IEEE
Access 8: 60364–60381.
22. Terazono, D., Liu, J., Miura, Y. et al. (2020). Grid frequency regulation support
from back-to-back motor drive system with virtual-synchronous-generator-based
coordinated control. IEEE Transactions on Power Electronics: 1. https://doi.org/
10.1109/TPEL.2020.3015806.
23. Demetriou, P., Asprou, M., Quiros-Tortos, J., and Kyriakides, E. (2017). IEEE 9-bus
modified test system. https://www2.kios.ucy.ac.cy/testsystems/index.php/ieee-9-
bus-modified-test-system/ (accessed November 2020).
24. Bevrani, H., Imanaka, M., and Kato, T. (2020). Grid dynamics shaping using
flexible controlled power converters. Energy Reports 6: 1490–1495.
References 141
25. Madariaga, A., Martín, J., Zamora, I. et al. (2013). Technological trends in electric
topologies for offshore wind power plants. Renewable and Sustainable Energy
Reviews 24: 32–44.
26. Rodrigues, S., Restrepo, C., Kontos, E. et al. (2015). Trends of offshore wind
projects. Renewable and Sustainable Energy Reviews 49: 1114–1135.
27. Castro, L.M. and Acha, E. (2015). On the provision of frequency regulation in low
inertia AC grids using HVDC systems. IEEE Transactions on Smart Grid 7 (6):
2680–2690.
28. Leon, A.E. (2017). Short-term frequency regulation and inertia emulation using an
MMC-based MTDC system. IEEE Transactions on Power Systems 33 (3): 2854–2863.
29. Rakhshani, E., Remon, D., Cantarellas, A.M. et al. (2016). Virtual synchronous
power strategy for multiple HVDC interconnections of multi-area AGC power
systems. IEEE Transactions on Power Systems 32 (3): 1665–1677.
30. Poolla, B.K., Bolognani, S., and Dörfler, F. (2017). Optimal placement of virtual
inertia in power grids. IEEE Transactions on Automatic Control 62 (12): 6209–6220.
32. Lian, B., Sims, A., Yu, D. et al. (2016). Optimizing LiFePO4 battery energy storage
systems for frequency response in the UK system. IEEE Transactions on
33. Wu, Z., Gao, D.W., Zhang, H. et al. (2017). Coordinated control strategy of battery
energy storage system and PMSG-WTG to enhance system frequency regulation
capability. IEEE Transactions on Sustainable Energy 8 (3): 1330–1343.
34. Zhang, F., Hu, Z., Xie, X. et al. (2017). Assessment of the effectiveness of energy
storage resources in the frequency regulation of a single-area power system. IEEE
36. Golpîra, H., Seifi, H., Messina, A.R., and Haghifam, M.-R. (2016). Maximum
penetration level of micro-grids in large-scale power systems: Frequency stability
37. Spahic, E., Varma, D., Beck, G. et al. (2016). Impact of reduced system inertia on
stable power system operation and an overview of possible solutions. In:
Proceedings of 2016 IEEE Power and Energy Society General Meeting (PESGM), 1–5.
New York: IEEE.
on power system stability and operation. IFAC Proceedings Volumes 47 (3):
7290–7297.
39. Wang, Y., Bayem, H., Giralt-Devant, M. et al. (2014). Methods for assessing
available wind primary power reserve. IEEE Transactions on Sustainable Energy
6 (1): 272–280.
40. Wang, Y., Delille, G., Bayem, H. et al. (2013). High wind power penetration in
isolated power systems—assessment of wind inertial and primary frequency
responses. IEEE Transactions on Power Systems 28 (3): 2412–2420.
41. Chu, Z., Markovic, U., Hug, G., and Teng, F. (2020). Towards optimal system
scheduling with synthetic inertia provision from wind turbines. IEEE Transactions
on Power Systems 99: 1-1.
42. Golpîra, H., Haghifam, M.R., and Seifi, H. (2015). Dynamic power system
equivalence considering distributed energy resources using Prony analysis.
International Transactions on Electrical Energy Systems 25 (8): 1539–1551.
44. D’Arco, S. and Suul, J.A. (2014). Equivalence of virtual synchronous machines and
frequency-droops for converter-based microgrids. IEEE Transactions on Smart
Grid 5 (1): 394–395.
45. Rakhshani, E., Remon, D., Cantarellas, A.M., and Rodriguez, P. (2016). Analysis of
derivative control based virtual inertia in multi-area high-voltage direct current
interconnected power systems. IET Generation, Transmission & Distribution 10 (6):
1458–1469.
46. Hammad, E., Farraj, A., and Kundur, D. (2019). On effective virtual inertia of
storage-based distributed control for transient stability. IEEE Transactions on
Smart Grid 10 (1): 327–336.
47. Farmer, W.J. and Rix, A. (2019). Optimising power system frequency stability using
virtual inertia from inverter-based renewable energy generation. In: 2019
International Conference on Clean Electrical Power (ICCEP), 394–404. IEEE.
48. Attya, A., Anaya-Lara, O., and Leithead, W. (2018). Novel concept of renewables
association with synchronous generation for enhancing the provision of ancillary
services. Applied Energy 229: 1035–1047.
49. Poolla, B.K., Groß, D., and Dörfler, F. (2019). Placement and implementation of
grid-forming and grid-following virtual inertia and fast frequency response. IEEE
50. Esmaili, M., Ghamsari-Yazdel, M., Amjady, N., and Chung, C.Y. (2020). Convex
model for controlled islanding in transmission expansion planning to improve
frequency stability. IEEE Transactions on Power Systems https://doi.org/10.1109/
TPWRS.2020.3009435.
51. Borsche, T.S., Liu, T., and Hill, D.J. (2015). Effects of rotational inertia on power
system damping and frequency transients. In: 2015 54th IEEE Conference on
Decision and Control (CDC), 5940–5946. IEEE.
References 143
52. Fini, M.H. and Golshan, M.E.H. (2018). Determining optimal virtual inertia and
frequency control parameters to preserve the frequency stability in islanded
microgrids with high penetration of renewables. Electric Power Systems Research
154: 13–22.
53. Oudalov, A., Chartouni, D., and Ohler, C. (2007). Optimizing a battery energy
storage system for primary frequency control. IEEE Transactions on Power Systems
22 (3): 1259–1266.
54. Wogrin, S. and Gayme, D.F. (2015). Optimizing storage siting, sizing, and
technology portfolios in transmission-constrained networks. IEEE Transactions on
Power Systems 30 (6): 3304–3313.
55. Mo, O., D’Arco, S., and Suul, J.A. (2016). Evaluation of virtual synchronous
machines with dynamic or quasi-stationary machine models. IEEE Transactions on
Industrial Electronics 64 (7): 5952–5962.
57. Golpîra, H., Seifi, H., and Haghifam, M.R. (2015). Dynamic equivalencing of
an active distribution network for large-scale power system frequency
stability studies. IET Generation, Transmission & Distribution 9 (15):
2245–2254.
58. Ajala, O., Dominguez-Garcia, A., Sauer, P., and Liberzon, D. (2018). A library of
second-order models for synchronous machines. arXiv preprint arXiv:1803.09707.
59. Bollen, M.H. and Gu, I.Y. (2006). Signal Processing of Power Quality
Disturbances. Wiley.
60. Manolakis, D.G., Ingle, V.K., and Kogon, S.M. (2000). Statistical and Adaptive
Signal Processing: Spectral Estimation, Signal Modeling, Adaptive Filtering, and
Array Processing. Boston: McGraw-Hill.
61. Zakeri, B. and Syri, S. (2015). Electrical energy storage systems: a comparative life
cycle cost analysis. Renewable and Sustainable Energy Reviews 42: 569–596.
62. Mongird, K., Viswanathan, V.V., Balducci, P.J. et al. (2019). Energy Storage
Technology and Cost Characterization Report. Pacific Northwest National Lab
(PNNL), Richland, WA (United States).
63. Teruo, I. (2005). State of charge calculation device and state of charge calculation
method. US Patents, US6845332B2.
64. Tang, Z.X. and Lim, Y.S. (2016). Frequency regulation mechanism of energy
storage system for the power grid. 4th IET Clean Energy and Technology
Conference (CEAT 2016), Kuala Lumpur, 1–8. doi: https://doi.org/10.1049/
cp.2016.1272.
65. European Network of Transmission System Operators for Electricity (ENTSOE)
(2016). Frequency Stability Evaluation Criteria for the Synchronous Zone of
Continental Europe – Requirements and Impacting Factors, March.
66. Eto, J.H. (2011). Use of Frequency Response Metrics to Assess the Planning and
Operating Requirements for Reliable Integration of Variable Renewable Generation.
Berkeley, CA: Ernest Orlando Lawrence Berkeley National Laboratory Tech. Rep.
LBNL-4142E.
68. Ekwue, A. and Cory, B. (1984). Transmission system expansion planning by
interactive methods. IEEE Transactions on Power Apparatus and Systems 7:
1583–1591.
69. C. E. O. H. ENTSO-E (2009). P1-Policy 1: Load-Frequency Control and
Performance, ed: Tech. Rep. 2000-130-003, May 2000.
70. Chávez, H., Baldick, R., and Sharma, S. (2014). Governor rate-constrained OPF for
primary frequency control adequacy. IEEE Transactions on Power Systems 29 (3):
1473–1480.
71. Teng, F., Trovato, V., and Strbac, G. (2016). Stochastic scheduling with inertia-
dependent fast frequency response requirements. IEEE Transactions on Power
Systems 31 (2): 1557–1566.
72. Grid, N. (2019) Security and quality of supply standards. https://www.
nationalgrideso.com/industry-information/codes/security-and-quality-supply-
standards (accessed November 2020).
73. Borsche, T. and Dörfler, F. (2017). On placement of synthetic inertia with explicit
time-domain constraints. arXiv preprint arXiv:1705.03244.
74. Rogers, G. (2012). Power System Oscillations. Berlin: Springer Science &
Business Media.
75. Aien, M., Hajebrahimi, A., and Fotuhi-Firuzabad, M. (2016). A comprehensive
review on uncertainty modeling techniques in power system studies. Renewable
and Sustainable Energy Reviews 57: 1077–1089.
145
Wide-Area Voltage Monitoring in High-Renewable

Integrated Power Systems
The development of synchronized phasor measurement units (PMUs) technology

has given utilities the ability to implement wide-area voltage stability monitoring
systems, which provide time-stamped data in near real time. In this sense, selected
voltage control areas can be used to implement wide-area voltage–volt–ampere
reactive (VAR) control schemes, develop voltage monitoring and prediction
schemes, monitor reactive power reserve requirements, and develop (localized)
reactive power markets, just to mention a few potential applications.
In recent years, several analysis tools and techniques have emerged which are
suitable for application to voltage and VAR monitoring and control at both local
and global levels. At a local level, these techniques can be used to develop inde-
pendent voltage monitoring and control strategies for microgrids (MGs) and dis-
tributed generators (DGs) such as those described in Chapter 2; at a global level,
they have the potential to be integrated to wide-area voltage control schemes using
synchrophasor measurements.
In this chapter, a novel use of measurement-based analysis techniques to iden-
tify weakly interacting voltage control areas in power systems with high penetra-
tion of distributed energy resources is presented. The proposed technique
combines the inherent abilities of graph–theoretical techniques with spectral clus-
tering and visualization methods to identify voltage control areas and reconstruct
system behavior using selected measurements. Attention is focused on three main
aspects, namely the identification of critical system zones showing a coherent
behavior and the associated reactive power sources, the use of spectral analysis
to identify the critical buses, and the computation of reduced-order models.
The evaluation of voltage stability problems is done using both static and
dynamic techniques. To facilitate comparison between both simulation
approaches and allow full comparison between static and dynamic approaches,
statistical techniques are being developed to extract modal characteristics directly
from time-domain simulations. Connections with other modal identification
methods are also investigated and numerical issues are discussed.
146 5 Wide-Area Voltage Monitoring in High-Renewable Integrated Power Systems
5.1 Introduction
Wide-area area voltage control of large interconnected systems has attracted con-
siderable interest in the last few decades [1–6]. Network voltage control at genera-
tors, and dynamic reactive power compensation devices placed at key system
locations, among other measures, can support voltage regulation and enhance
system transient stability and operating flexibility [7]. The increasing size and
complexity of power systems with high penetration levels of MGs/DGs distributed
along a large geographical area or located in remote zones, however, make voltage
control challenging.
As the number of modern renewable energy sources (RESs) with improved static
and dynamic reactive power capability grows, a significant challenge is to integrate
them into existing voltage control schemes as well as to develop effective coordi-
nation schemes [8–10]. Better detection and forecasting techniques through the
concept of wide-area voltage monitoring are also needed to allow full use of control
capabilities and realize wide-area monitoring structures.
In the recent past, there has been renewed interest in the use of automatic volt-
age regulation schemes for wind and solar photovoltaic (PV) farms connected to
the bulk power systems [9–11]. Studies show that static VAR compensators (SVCs)
located adjacent to large wind and PV farms can be used to maximize reactive
power reserves and improve voltage profiles. Voltage–VAR control can contribute
to the overall power system angle and voltage stability and result in improved sys-
tem operation and security. This is a subject that is receiving increasing attention.
The integration of advanced voltage–VAR controls in the wind and solar genera-
tors, on the other hand, raises several complex issues [8, 11]. First, farm-level volt-
age control introduces a hierarchical control system that needs to be optimized to
improve the overall system response to system perturbations. Further, coordina-
tion with other nearby generators or network reactive power compensation
devices may be needed as in many cases, wind farms require additional reactive
power support, especially during transient conditions.
One of the critical issues in modeling the inverter-based generators is reactive
power capability. Some interesting phenomena of voltage and reactive power
responses from solar PV generators have been observed and investigated, such as
high voltages under normal conditions, high transient voltages, and sustained oscil-
lations following a fault or change in the control characteristics of wind and solar PV
farms [1]. These problems may cause further reliability concerns such as overload of
subtransmission and distribution facilities, unexpected generation tripping for
overvoltage or under-excitation, and even system-wide transient instability.
Experience with the application of primary and secondary voltage control in
power systems in European countries shows that coordination of reactive sources
may result in enhanced system wide-area control and reliability [2, 12–14]. To
avoid undesirable interactions, voltage control at the various levels should tempo-
rally and spatially independent. This requires splitting the system into non-
interacting zones in which voltage is controlled individually.
Large-scale coordination of reactive power sources is challenging due to a large
number of control characteristics, the location and type of controllers, and the
characteristics of each device. Issues such as reserve capacity control and the effi-
cient utilization and coordination of reactive power sources must be addressed to
achieve fast automatic voltage control and keep the capacitive output margin
against system contingencies.
This chapter discusses the experience in the development of data-driven analysis
techniques to identify and update voltage control zones and the associated reactive
power resources. A systematic methodology for the identification of voltage con-
trol zones is first introduced. The proposed procedure consists of three main steps:
(i) the identification of strongly connected buses showing coherent behavior,
(ii) the identification of generators and SVCs participating in the critical zones,
and (iii) the determination of distance measures indicating relationships between
bus voltage magnitudes and reactive power sources. These methods are suitable for
large-scale applications and can be used to coordinate multiple available reactive
compensation devices, including SVCs, synchronous condensers, generator exci-
tation systems, and modern RESs equipped with closed-loop voltage control
schemes.
The design methodology is demonstrated on a complex test system with signif-
icant wind penetration in which several SVCs are used to control system voltage.
Results show that properly coordinated reactive power sources may have an
important impact on system dynamic behavior.
5.2 Voltage Control Areas: A Background
Recent years have witnessed the development and application of wide-area voltage
control systems with the ability to monitor voltage deviations at key transmission
buses, update set points of major closed-loop controllers, and coordinate reactive
power sources to regulate network voltages. Due to the local nature of voltage
behavior, a major issue in these hierarchical control schemes pertains to the iden-
tification of nearly independent voltage control zones.
A general review of the voltage control structures is presented in Chapter 1. In
this chapter, a brief overview of these methods, in the context of voltage monitor-
ing techniques is given. The discussion begins with a review of fundamental
concepts in the development of practical wide-area voltage control of power
systems. Then, the need for data-driven approaches to identify voltage control
zones is established.
5.2.1 Voltage Sensitivities

A widely used criterion to determine voltage control areas is the notion of voltage
and reactive power sensitivities. Following Kundur [15] and adopting the nomen-
clature in Chapter 3, the linearized steady-state system power voltage equations
can be written as:
ΔP J Pθ J PV Δθ
= 51
ΔQ J Qθ J QV ΔV
where JPθ, JPV, JQθ, and JQV are blocks of the Jacobian matrix and represent sen-
sitivities between power flow and bus voltage changes; ΔP and ΔQ are the incre-
mental changes in bus real power and bus reactive power injection; Δθ and ΔV are
the incremental changes in bus voltage angle and bus voltage magnitude.
Making ΔP = 0 in (5.1), one has that
−1 ∂Q
ΔQ = J QV − J Qθ J Pθ J PV ΔV = ΔV = J QV R ΔV 52
∂V Vo
and, therefore,
−1 ∂V
ΔV = J QV R ΔQ = ΔQ = J VQR ΔQ
∂Q V o ,Qo
where matrix J QV R is called the reduced steady-state Jacobian matrix of the system
and J VQR is a sensitivity matrix with coefficients:
∂V 1 ∂V 1 ∂V 1
∂Q1 ∂Q2 ∂Qn
∂V 2 ∂V 2 ∂V 2
∂Q1 ∂Q2 ∂Qn
J VQR = J VQRij =
∂V n ∂V n ∂V n
∂Q1 ∂Q2 ∂Qn
where n is the total number of nodes, and in the interest of simplicity, all buses are
assumed to be PQ buses.
Simplified approaches for calculating sensitivity relations are described in
Ref. [15]. Further, Ref. [16] describes alternative approaches to calculate
sensitivity relationships based on the active–reactive decoupling characteristics

in the power flow computations.
In the same vein as Lagonotte et al. [4] assume now that a dominant node (a pilot
point or pilot node), V pk is determined that represents the average or dominant
system behavior for area k. Valuable insight into the influence of changes in
the magnitude of V pk on reactive power deviations of nearby generators, SVCs,
and RES in area k can be obtained by writing the sensitivity of bus reactive power
injection at generator i to changes in the bus voltage deviations of the pilot bus at
bus k as:
ΔQgi ∂Qgi ∂V HV i
= ,i k 53
ΔV pk ∂V HV i V o ,Qo ∂V pk Qo
where V HV i is the high side bus of the machine at bus i and Qgi represents reactive
∂Qgi ∂V HV i
power injections at bus i; the terms ∂V HV i , ∂V pk represent sensitivity coefficients.
∂Qgi
Numerically, the coefficients ∂V HV i can be obtained directly from the load flow
∂V HV i
Jacobian matrix; the second term, ∂V pk , represents an electrical distance and is
a subproduct of the calculation of electrical distances in Section 5.2.2. Sensitivity
matrices are real and nonsymmetrical and reflect the propagation of voltage var-
iation following reactive power injection at a bus [17]. Variations of this model
using other system formulations are described in [4, 15–17], and references
therein.
5.2.2 Electrical Distances

In Ref. [4], an approach based upon the application of electrical distance sensitiv-
ities was introduced to estimate the magnitude of coupling between two nodes as
well as to identify non-overlapping voltage control areas.
Using the same notation as in Eq. (5.3), the electrical attenuation coefficient, αij,
between nodes i and j can be defined as:
ΔV i ∂V i ∂Q j ∂V i ∂V j
αij = = = 54
ΔV j ∂Q j ∂V j ∂Q j ∂Q j
or ΔVi = αijΔVj, where, in general, αij αji. Physically, αij ≈ 1, when buses i, j are
electrically close and has a small value when they are electrically distant. Other
approaches to determining approximate sensitivity relations based on the load
flow equations are described in [4, 17].
An attenuation matrix can then be obtained from the notion of entropy or infor-
mation theory. The electrical pairwise distance, dij, between buses i and j can be
defined as dij = dji = − log(αij. αji) that has two important properties, positivity and
symmetry.
Extending this approach to the multidimensional case, one can define the matrix
of attenuation between all the buses, A, as:
α11 α12 α1n
α21 α22
A = αij = α2n 55
αn1 αn2 αnn

As discussed in the practical application of the method in Section 5.5, the atten-
uation matrix is numerically sparse and a threshold value can be defined to con-
struct a truncated attenuation matrix A (i.e. not all sites are connected).
In the context indicated later, the attenuation matrix A can be reinterpreted in
terms of a connected, weighted graph, whose nodes or vertices represent physical
nodes or sensor locations and the edges represent distance relationships. Edges
with a value above a predefined threshold can be retained to reduce the compu-
tational burden and improve understanding of the underlying connectivity or the
required communication structure.
This interpretation provides the opportunity to obtain distance or attenuation
measures, directly from observational or measured data as explained in
Section 5.4, and makes the algorithms well suited for efficiently applying graph-
based clustering procedures. A second desired application is the evaluation of hier-
archical or decentralized communication structures.
In the case of power networks, several graph-based system representations have
been proposed and used to divide data into clusters based on concepts such as
electrical distance [18] or spectral characteristics [19]. A limitation of these formu-
lations is their inability to represent the time-ordering of the data, especially from
PMU data recordings, which makes them more appropriate to represent and char-
acterize static structures rather than dynamic information. Further many graph-
based representations are based on several simplifying assumptions, many of
which make them unsuitable for representing dynamic data.
These observations motivate the view that a dynamic framework is needed to
capture both structural and temporal information in measured data.
5.2.3 Reactive Control Zones and Pilot Nodes

Most existing wide-area voltage control schemes rely on the decomposition or divi-
sion of a large power system into loosely interacting voltage control areas, in which
the voltages within each area are independently controlled using large or domi-
nant reactive power resources in an automated fashion [2].
The approach involves three main steps:
1) Subdivide the attenuation data A into subsets or clusters which are pairwise
disjoint and connected;
2) Select a robust node or bus that represents the average or dominant system
behavior;
3) Determine sensitivity relations between dominant bus voltage behavior and
reactive power sources.
These approaches are usually hierarchical; first, a voltage control zone associ-
ated with the largest sensitivities is determined. Then, the remaining zones are
determined using an iterative approach [1, 2, 4].
5.2.3.1 Selection of Optimal Pilot Buses

The determination of reactive zones and pilot nodes in voltage control schemes
can be formulated as a problem of clustering distance measures. In this sense,
the determination of pilot bus voltages for area k, V pk can be posed as the solution
of an optimization problem (refer to Figure 5.3):
pilot nodeareak = min αij 56

i, jϵk
where several constrained optimization techniques can be utilized.

Thus, for instance, with reference to Figure 5.1, node 1 is selected as a candidate
pilot bus for area i, if αtot1 = 5k = 1 α1k < αtot2 < … < αtot5 subject to the constraint
that the electrical distance with other pilot buses should be small to avoid
interactions.
From a practical standpoint, two main criteria are often used to select bus pilot
buses [2, 4, 12]:
1) Pilot buses should be chosen among the strongest nodes in the system using
criteria such as the maximum short-circuit current.
2) Coupling between pilot nodes associated with different control areas should be
low to avoid interactions among control systems.
It is worthwhile noting that the determination of voltage pilot nodes involves

finding a solution to a changing problem, in which system topology and operating
conditions are always changing. This is an issue that must be addressed from a
dynamic perspective.
5.2.3.2 Selection of Control Plants

Generating units and flexible AC transmission system (FACTS) devices participat-
ing in wide-area control schemes are often selected on heuristic criteria requiring
Cluster j
Inter-cluster
distances
Cluster i
α12 α23
α24
1 α13 mi 3
Cluster k
4
α34
α25
α15 α45
α35
Cluster i with ni = 5 nodes and a centroid mi
Figure 5.1 Illustration of clusters associated with weighted, connected graphs. Each node
within cluster i has several connections (dense connection) with other members of the
cluster and fewer links (sparse connections) with nodes outside the cluster.
that they belong to the control area under analysis, the size of their reactive power
capability, and the highest electrical coupling with the pilot nodes (see Eq. (5.3)).
These methods, however, are commonly characterized by several limitations:
•• The methods do not fully recognize the dynamic nature of system behavior;
In many applications, not only frequently pilot nodes need to be reselected fol-
lowing topology changes but also the control areas and the associated reactive
power sources need to be updated;
• Communication and bandwidth structure and requirements cannot be

determined.
This has motivated the development of alternative, measurement-based

techniques.
5.2.4 Other Approaches

Recently, approaches to address the challenge of determining reactive zones have
been proposed. In Ref. [19], spectral clustering was proposed to discover clusters in
5.3 Data-driven Approaches 153
power networks from the orthogonal structure of spectral embedding. Other

approaches to identify clusters include Refs. [20, 21]. In the former approach,
the authors discussed the problem of determining voltage–VAR control areas
using spectral graph clustering applied to the system graph representation
obtained from the power system flow equations. By examining the spectral proper-
ties of a graph-based representation, it was possible to identify the location and
number of VAR–voltage control areas. In most of these applications, however,
use is made of spatial information which precludes their application in a real-time
setting.
In Sun et al. [22], an adaptive zone-division-based automatic voltage control sys-
tem based on the concept of VAR, control space was introduced and applied to the
Chinese power grid. A key feature of this approach is that the control zones are no
longer fixed but are reconfigured online and updated following variations in the
grid structure. Yet another approach is the use of online voltage stability assess-
ment techniques to simultaneously identify voltage control areas and reactive
power reserves [23]. Such identification techniques are invaluable when dealing
with frequent changes in system behavior or system structure.
5.3 Data-driven Approaches
Real-time voltage and reactive power measurements from time-synchronized

wide-area monitoring systems (WAMSs) provide the opportunity to analyze and
cluster motion trajectories, thus offering a useful complementary approach to con-
ventional static analysis. One potential drawback of these data is that they may
involve various timescales and differ in magnitude and nature. In practical appli-
cations, determining reactive control areas directly from time-domain simulations
may be difficult since no sensitivity information is readily available, and therefore
measures of similarity (coherency) are needed.
Further, with a wide array of multimodal, multichannel PMU data being rou-
tinely acquired for system monitoring, there is a pressing need for quantitative
tools to combine these varied channels of information [21]. The goal of these mea-
suring and analysis techniques is to combine complementary measurements such
as voltage and reactive power measurements while improving on the predictive
ability of any individual modality.
In the following subsections, wide-area measurement-based structures and con-
trol strategies to decompose the system into an arbitrary number of voltage control
zones are introduced that may complement static approaches. Then, data-driven
techniques are proposed and tested to identify and characterize voltage con-
trol zones.
5.3.1 Wide-Area Voltage and Reactive Power Regulation

Figure 5.2 shows a component diagram of a wide-area voltage monitoring system
adopted in the analysis that incorporates ideas from previous work [24]. The volt-
age control structure is hierarchical and consists of two major levels or measure-
ment hierarchies: primary (local) and secondary (regional).
Inputs to the WAMS include time series PMU data such as bus voltage magni-
tudes and angles, and reactive power of wind and PV generators, major generating
plants, and SVCs. The outputs of the system may be used to select and update PMU
or voltage control areas, select sensor locations, or be used as inputs to a voltage
monitoring and stability assessment or prediction modules with the ability to
determine reduced-order models, and trajectory classification and visualization.
The ability to select loosely or non-interacting voltage control areas for system
monitoring from sensor measurements such as supervisory control and data acqui-
sition (SCADA) system and phasor voltages is important for at least three reasons.
First, detecting voltage and reactive power deviations in near real-time can help
track reactive reserves and make decisions about reactive power management.
Second, the detection of voltage and reactive power deviations can produce
Voltage control
Area 1
∼ ∼ Q charts
Qg Direct ...
1 clustering
Wide-area Voltage control
Dimensionality areas,
V1 ∠θ1 voltage reduction Pilot buses,
∼ Vp1 ∠θp1 monitoring Trajectory
Q-V correlation,
classification
Qgn Map visualization
...
∼ ∼
∼ Reactive control area

Pilot bus
Voltage control Remote Sensor location
Area k renewable generation
Figure 5.2 Overview of the steps in computing voltage control areas. Here, V p j represents
the set point of pilot buses, Vj and θj are the bus voltage magnitude and phases, and Qg j
represent reactive power output from major generators and network reactive power
compensation devices.
knowledge about the strength of interactions between multimodal data, i.e. volt-
age and reactive power. Further, measurement techniques give some insight into
the nature of control structure and communications required for efficient imple-
mentation of practical WAMSs, especially in power systems with RESs [25].
Three main activities are of interest here:
1) Direct trajectory clustering;

2) Dimensionality reduction and spectral clustering; and
3) Trajectory classification and visualization.
The issues of dimensionality reduction and spectral clustering are addressed in

Sections 5.4.1 and 5.4.2. The issue of direct clustering is deferred to Section 5.4.3.
5.3.2 PMU-Based Voltage Monitoring

Recent attempts to develop WAMS using synchrophasor technology allow for effi-
cient network partitioning and the identification of critical voltage areas. In the
bulk of these applications, each data type, i.e. voltage or reactive power measure-
ments is analyzed independently and then combined to determine its relationship
with other data modalities. Cross-information between data types, however, is lost
or the sequential combination of the independent model may result in inaccurate
system characterization.
Processing of large volumes of voltage and reactive power observations or simu-
lations, on the other hand, raises significant challenges as:
• Data represent different physical units. Simple data normalization techniques

may affect results;
• Bus voltage magnitudes from different physical regions may show similar
behavior thus obscuring physical interpretation. Phase information may be
needed to improve the extraction of coherent structures;
• Practical voltage control systems should only consider selected voltage system
behavior (model reduction) from which global system behavior can be
reconstructed;
• Correlation techniques are also needed to identify common features between

different datasets.
5.4 Theoretical Framework
In this section, an approach based on spectral analysis techniques is introduced to

examine dynamic trends and phase relationships between key system signals from
measured data. Drawing on graph-based techniques and spectral analysis
methods, a technique based on the notion of diffusion maps (DMs) is used to

express an ensemble of measured data as a nonlinear combination of modal coor-
dinates from which voltage control zones can be determined.
First, the notion of dynamic trajectories in the context of WAMS is introduced.
5.4.1 Dynamic Trajectories

Measured dynamic data have a useful interpretation in terms of spatiotemporal
motion trajectories and associated graphs that evolve with time, as shown sche-
matically in Figure 5.3. Referring to Figure 5.3, assume that xk(tj) denotes a
sequence of observations of a simulated or measured transient process at locations
xk, k = 1,…,m, and time tj, j = 1,…,N, where the m locations represent sensors [26].
At each time instance tj, the instantaneous (pointwise) distance between two
motion trajectories (time sequences) xi(tj) and xj(tj), is given by:
dij x i , x j, t j = x i t j − x j t j 57
Distance
function
where xi(tj) and xj(tj) represent the instantaneous values of the time sequences at
time instance tj (a time slice in the spatiotemporal representation in Figure 5.1). In
practice, distances can be obtained for a time window of interest and a full distance
t1 tk tN
d11 d12 ... d16 d17

d21 d22 ... d26 d27
...
...
...
...
...
d61 d62 ... d66 d67

d71 d72 ... d76 d77
Similarity
(distance) Normalization
matrix
Vertex i v Coarse-graining
i
weight Transition • Dimensionality
wij reduction
vj probability
• Anomaly detection
matrix
• Visualization
Undirected graph
Figure 5.3 Illustration of distance (similarity) matrices and the associated undirected
graphs.
matrix can then be defined as D = [dij]. In this case, however, the distance defi-
nition does not account for the time-ordering of the signals.
This allows the application of graph-clustering algorithms or coarse-grained
methods to distance (similarity) matrices.
5.4.2 Spectral Graph Theory

Similarity or distance matrices based on (5.7) have an interesting interpretation in
terms of weighted, undirected graphs, where dij = dji [27]. Referring back to
Figure 5.3, a weighted graph G(V, w) with m vertexes V = v1 v2 … vm and
nonnegative weights wij = w(vi, vj) can be constructed by computing a pairwise
similarity distance between trajectories. Each measurement (or sensor) location
corresponds to a node or vertex, while the edges (weights) represent the interaction
strength or affinity between trajectories [24, 27, 28].
Using the distance between motions or trajectories, the weight on the graph
edges is given by:
wij = x i t − x j t
with
m
wii = wij
j=1
An m × m distance matrix, W = [wij] can now be defined whose nodes corre-

spond to the dynamic trajectories [29–32]. Figure 5.3, which is excerpted from
Ref. [33], illustrates the definition of vertices and weights in the concepts of the
intended application. Similar interpretations can be found in the context of spatial
filtering or the modeling of spatial processes [28].
As implied in Figure 5.3, the calculation of weights, wij, requires consideration of
all pathways on the graph. In practical applications, such as those associated with
dynamic trajectories, however, transition probabilities between different nodes
may be small and a threshold can be used to obtain a sparse representation of
the graph.
Coupled with efficient visualization techniques, graphs can also be used to dis-
play efficiently coherent groups.
5.4.3 Kernel Methods

Graph-based kernel methods have been recently explored to extract dominant
behavior from measured data [29]. Given a measurement matrix X in (5.1), a pos-
itive-definite matrix kernel (a Gaussian kernel) K can be obtained whose (i, j)th
element is given by:
xi t − x j t
K = k ij = exp − , 1 ≤ i, j ≤ m 58
εi ε j
where the kernel bandwidth εi, εj controls the hop size of the random walk and can
be automatically tuned based on the distribution of matrix A. The use of other ker-
nels is discussed elsewhere.
Physically, matrix K can be interpreted as the adjacency matrix of a graph, each
of whose nodes represents one sensor or trajectory. The kernel bandwidth scale, εi,
εj modulates the notion of distance in (5.7) and can be tuned to cluster system
behavior. If εiεj maxij xi(t) − xj(t) , then for all edges {i, j}, kij ≈ 1; for low kernel
bandwidth values, εiεj 0, the pairwise distances kij become increasingly similar
and large fluctuations in the density of points in the high-dimensional space are
smoothed out.
From the matrix kernel in (5.8), one now can define a diagonal matrix D = [dij]
whose entries are the row sums of K, namely:
m
j = 1 K 1j 0 … 0
m
D = dij = 0 j = 1 K 2j … 0
m
0 0 … j = 1 K mj
m
where dii = j = 1 K ij is the degree of node xi.
The distance matrix has an interesting interpretation in terms of an undirected
probabilistic graph. In this concept, the transition probability pij from i to j can now
m
be obtained as pij = k ij k = 1 k ik. The right-stochastic Markov matrix or probabil-
ity matrix, M, can then be defined as:
k 11 k 12 k 1m
m m … m
j = 1 K 1j j = 1 K 1j j = 1 K 1j
k 21 k 22 k 2m
m m … m
M = mij = D −1
K= j = 1 K 2j j = 1 K 2j j = 1 K 2j
k m1 k m2 k mm
m m … m
j = 1 K mj j = 1 K mj j = 1 K mj
59
in which the matrix element mij = pij can be interpreted as the probability pij of
hopping from point i to point j in t steps of a discrete random walk [32].
Physically, the transition probability matrix, M, defines the random walk of a
particle on the graph. Formally, suppose that the initial probability of the particle
being at a vertex vj is poj j = 1, …, n . It follows that the probability of the trajectory
vj taking the edge wij is mij poj. The extension to the multivariate case follows along
the same lines.
Paramount to the automated extraction of low-dimensional representations is
the adaptive computation of the Gaussian kernel widths, ε. In general, the scale
parameter is related to the statistics and the geometry of the data points. Following
Ref. [29], let Xj denotes a cloud of points around xi. The variance of the distance
between the point xi to all the points xj Xj is given by Ref. [30]:
2
xi − x j − X i
εi =
xi X j
Xi
where
Xi = xi X
xi − x j Xi
It is easy to prove that matrix M is nonnegative, unsymmetrical, and invariant to

the observation modality and is resilient to measurement noise. It is a stochastic
transition (probability) matrix with the following important properties:
mij ≥ 0
j m ij = 1 for j = 1, …, n row stochastic
Diagonalization of M produces an ordered set of eigenvectors and eigenvalues

{(ϕi, λi), i = 1, …, d} with λ1 = 1, and λ1 ≥ λ2 ≥ … ≥ λd. Once a distance matrix is
obtained, spectral techniques can be used to obtain reduced-order representations
as well as to identify and eliminate nonrelevant clusters. The main goal of cluster
analysis is data reduction, by subdividing a set of objects into a hierarchical
arrangement of homogeneous subgroups. A significant outcome is reduced com-
plexity with a minimal loss of information, which fosters a better understanding of
the analyzed data.
Examples of this class of algorithms are nonlinear reduction techniques such as
DMs [29], the Mei–Sheila algorithm [33], and the Markov clustering (MCL) algo-
rithm described in Ref. [34]. Two approaches discussed here, spectral clustering
and the MCL algorithm, show promise for automated identification of coherent
patterns and model reduction.
5.4.3.1 Markov Matrices

A key step of spectral analysis algorithms is the projection into a reduced-order
subspace. Let matrix M have a set of eigenvalues λj with associated eigenvectors,
ψ j. From linear system theory, the eigenvectors of matrix M in (5.9) satisfy
M − λ j I ψ j = 0, j = 1, …, n 5 10
For purposes of analysis, it is convenient to perform spectral analysis on a sim-

ilar matrix, M s = msij = D − 1 2 KD1 2 (a symmetric matrix) that results in a sym-
metric eigenvalue problem. From (5.10), it follows readily that:
M = D − 1K = D − 1 2
D − 1 2 KD1 2
D−1 2
= D − 1 2MsD − 1 2
5 11
and thus
M s = D − 1 2 KD1 2
= D − 1 2 DMD − 1 2
= D1 2 MD − 1 2
is a normalized affinity matrix (a normalized kernel), where use has been made of
the identity K = DM.
Since matrices M and Ms are related by a similarity transformation, they share
the same eigenvalues. To prove this, let ψ denote the eigenvalues of M and Φ
denote those of Ms. Observing that Ms = D−1/2KD−1/2, it can be inferred that
Ms is a symmetric matrix, which allows the use of special techniques for calculat-
ing the associated singular values. Griffith [31] gives an interesting interpretation
of these matrices in the context of spatial analysis and filtering.
Collecting all eigenvalues yields
MΨ = λ j Ψ
MsΦ = λ jΦ
with Ψ = ψ o ψ 1 ψ m − 1 and Φ = ϕo ϕ1 ϕm − 1 .
From these relationships, the eigenvalues λj of Ms satisfy
M s − λ j I ϕ j = 0 = D1 2 MD − 1 2 − λ j I ϕ j = M − λ j I D − 1 2 ϕ j = 0
ψj
from which it follows that ϕj = D1/2ψ j is an eigenvector of Ms with eigenvalue λj,

and matrices M and Ms share the same eigenvalues.
Moreover, because of the symmetry conditions, the eigenvalues are real and
satisfy λo = 1, and λo > λj > λj + 1, j = 1, 2, 3,…, n. Noting further that Ms is a
row-stochastic matrix j msij = 1 , the solution of the system Msϕo = λoϕo,
for λo = 1 gives the constant vector, ϕo = 1 1… 1 T , which is often associated
with the overall trend or average behavior.
The mapping from the original space into the new DM space is now defined at
time t, as the map:
T
Ψε x i xi λt1 ψ 1 i , λt2 ψ 2 i , …, λtN − 1 ψ N − 1 i ᑬN × 1 5 12
where ψ m(i) denotes the ith element of ψ m.
In practice, a d-dimensional DM can be defined as:
T
Ψdε x i X λ1 ψ 1 i , λ2 ψ 2 i , …, λd ψ d i ᑬd 5 13
where d << N is the number of relevant coordinates or subset of dominant eigen-

vectors (the intrinsic dimensionality). Typically, two to three coordinates suffice to
capture relevant system behavior.
By truncating the spatial patterns, dimensionality reduction can be performed.
Figure 5.4 gives a schematic illustration of this model, showing three main activ-
ities or features: (i) nonlinear mapping of the system dynamics, (ii) the calculation
of a low-dimensional embedding, and (iii) inverse mapping and model
reconstruction.
Associated with the spectral model, the diffusion distance can now be defined in
terms of the forward probabilities M as:
m
mir − mjr
Dij = ;
j=1
Ψ xr
m 5 14
M jm
j=1
Ψ xr = m m
k=1 j = 1 M jk
Observational
High-dimensional data
(input) space t1 tk tN
Reconstruction
Clusters
Linear of data Inverse
(nonlinear)
mapping
mapping
Ψ εd (x1) λ1ψ1 (1) λ2ψ2 (1) λdψd (1)

..
.
Low-dimensional λ1ψ1 (2) λ2ψ2 (2) λdψd (2)

.. .. . . ..
..
.
embedding . . . .
Ψ εd (xm) λ1ψ1 (m) λ2ψ2 (m) λdψd (m)
..
.
Ensemble
Figure 5.4 Mapping from the original space to the DM space.

Physically, the diffusion distance is small if there are many high-probability

paths of length t between two points.
Several remarks regarding time-coupled DMs are in order [33]:
• The low-dimensional approximation to the original model in physical space

(reconstruction) can be expressed as:
m d m
X = x1 x2 … xd = a j ψ Tj = a j t ψ Tj + ak t ψ Tk
j=o j=o k=d+1
Relevant system behavior non − essential coordinates
• The eigenvectors associated with the largest singular values correspond to slow
modes governing the long-time system evolution, i.e. ψ j, j = 1, …, d;
• Numerical experience with large spatiotemporal models shows that

ψ o ψ To > ψ 1 ψ T1 > … > ψ d ψ Td , and ψ i ψ Tj ≈ 0, i, j = 1, …, d, i j ; In practice, a
spectral gap can be observed at λd, such that:
λ1 ≥ λ2 ≥ … ≥ λ d λd + 1 ≥ … λm
Slow motion
Now, attention is turned to other graph-based clustering algorithms based on the

Markov matrix.
5.4.3.2 The Markov Clustering Algorithm

The MCL algorithm is an unsupervised graph-based algorithm for clustering
graphs based on simulation of stochastic flow in the graph [34]. Given a stochastic
transition matrix, M, the MCL method iteratively applies two operators called
expansion and inflation until convergence is obtained. Both expansion and infla-
tion are operators that map the space of column stochastic matrices onto itself.
Additionally, a pruning step is performed at the end of each inflation step to save
memory. By combining this approach with spectral analysis, a reduced-order
model can be obtained.
The inflation value parameter of the MCL algorithm is used to control the gran-
ularity of the tightness of these clusters. As noted by Refs. [35, 36], the inflation is
equivalent to taking the Hadamard power of a matrix followed by a scaling step to
ensure the matrix is stochastic again.
The method is based on the analysis on a transition network which is obtained
by (i) mapping the dynamic trajectories onto a discrete set of microstates and
(ii) building a transition network in which the nodes are the microstates and a link
is placed between them if two microstates are visited one after the other along the
trajectory.
To formalize the adopted model, let X ᑬm × N be the matrix of measurements

corresponding to a given operating scenario and M ᑬm × N be the transition or
Markov matrix. The MCL algorithm is summarized in Table 5.1.
In practical applications, the resulting models, however, may be difficult to
interpret and visualize, especially in the case of several dynamic trajectories. To
facilitate understanding of the resulting graphs in the developed algorithms,
the eigenvalues and eigenvectors of the steady-state matrix are calculated as
(MMCL(r) − λMCLI)ϕMCL = 0. This allows direct comparison with spectral analysis
techniques in Subsection 5.4.2.
As discussed in Ref. [37], the parameter r determines the granularity or tightness
of the clustering. This approach can be used to identify states sharing common,
dynamically meaningful characteristics such as coherency or similar timescales.
Experience shows that a low value of r may be enough to capture the dynamics
of interest (r = 4 has been found to give good results in this research). Several var-
iations of this method have been discussed in the literature.
Several properties are inherent to this model such as the following two cases:
Table 5.1 The Markov clustering algorithm.
Given a trajectory matrix X ᑬm × N ,

1) Obtain the undirected graph using the procedures in Sections 5.4.1 and 5.4.2;
2) Build a squared transition matrix, M = P = pji ᑬm × m , using the procedure in
Section 5.4.3, in which each element pji represents the transition probability from node j
to node i;
3) Normalize the matrix to ensure that the matrix is stochastic again;
4) Expand by taking the ith power of the matrix (a Hadamard product):
M i=M×M×…
i times
and normalize each column to one;

5) Inflate each of the columns of matrix M with power coefficient r as:
r
M pq
Γr M pq = m r
q=1 M pq
where Γr is the inflation operator;
6) Repeat steps 3 through 5 until MCL converges to a steady-state matrix MMCL(r);
7) Compute the first few right eigenvalues and eigenvectors of the resulting matrix
MMCL(r), using spectral analysis of matrix MMCL(r); this results in a set of real eigenvalues
and eigenvectors ui.
1) Like the case of the spectral representation in DMs, the first eigenvector has a
unit value and can be associated with average behavior;
2) The second and third eigenvector capture dominant system behavior.
5.4.4 Spatiotemporal Clustering

Spatiotemporal clustering methods with the ability to extract and monitor clusters
are well suited for automated voltage monitoring. They can be considered as direct
clustering techniques, since they can be applied directly to measured data. Among
several clustering techniques, C-means clustering offers a powerful means to clus-
ter dynamic trajectories directly from measurements or simulations [33].
Fuzzy clustering can be posed as the solution of the energy function [38, 39]:
K n
2
f U, v = ki x i − C k
um 5 15
k=1i=1
where U is the fuzzy c-partition of the data, v is the vector of centers,

v = C1 C2 Cc , the Ck are the cluster centers, the um ki represent associated
membership likelihoods (the degree of membership), n is the number of clusters,
m is a weighting exponent, 0 ≤ uki ≤ 1, of the trajectory xi being associated with the
cluster center Ck [40].
Using this view, the similarity that a point (node) shares with a given cluster can
be represented by a function whose values vary between 0 and 1, subject to the
constraints:
K
u
k = 1 ki
= 1, i = 1, …, n
uki ≥ 0, k = 1, …, K, i = 1, …, n
More formally, given n dynamic trajectories, xi, i = 1, …, n, the problem can be

stated as follows:
K n
2
f U, v = ki x i − C k ,
um
k=1i=1
K
5 16
subject to u
k = 1 ki
= 1, i = 1, …, n
uki ≥ 0, k = 1, …, K, i = 1, …, n
Outputs of this algorithm are the clusters and the time evolution of the centroids.
Combined with statistical quality control charts, the method can be used to detect
anomalous operation and islanding conditions.
5.5 Case Study 165
5.5 Case Study
To illustrate the utility of data-driven techniques in aiding voltage monitoring

and system partitioning, a fully represented model of a large-scale 5449 bus
network was developed for the study. Figure 5.5 shows a simplified schematic
of the study region showing interconnections with the boundary systems. The
study region of interest consists of 174 buses, 104 generators, three major SVCs,
and several wind farms. The test system has transmission levels 115, 230, and
400 kV; the total area load is 3263 MW.
The study simulates the entire test system, but results are limited to the study
region.
5.5.1 Sensitivity Studies

Sensitivity studies are first conducted to assess the influence of system structure
and control action on the nature and distribution of critical modes. In this analysis,
the output of a large-scale power flow analysis program is used to determine bus
voltages and reactive power sources having a strong influence on system behavior.
Hydro generator
To Areas 2 and 3 Thermal/nuclear
Bus
To Areas 4 and 5 105 SVC

156 25
153 32 Wind farm
108
87 113 33 Regional boundary
7 29 61 151
83 36
34 173 21
162
94 22 24 To Area 7
65 20 88 4 39
1 160
18 8
167 125 149 26
136 70 53
170 165 SVC-2 100
17 13
81
84 148 44 73 69 66
SVC-1 150 2
10 40 67 122 SVC-3
101 90
117 15 132
57 35
124
Wind farms
To Area 4
Figure 5.5 Single-line diagram of the study region showing the 230-/400-kV transmission
system and its interconnection with other regional systems.
To permit comparison with well-established procedures, two main analysis tech-

niques have been considered: (i) conventional short-circuit studies and
(ii) sensitivity studies based on Section 5.2.
In the first stage, short-circuit studies were conducted to determine the main
buses and generators involved in the identification of voltage control zones for
both the pre-contingency and post-contingency cases [2]. Then, sensitivity coeffi-
cients are computed from the Jacobian equations.
For the sake of comparison, Table 5.2 lists the maximum short-circuit levels for
major 400-kV transmission buses within the study region extracted from the base
case power flow solution ranked in decreasing order of magnitude. As discussed in
Ref. [2], pilot nodes are usually chosen based on robustness and coupling charac-
teristics. Candidate pilot nodes are expected to be strong and must exhibit low cou-
pling with other candidate nodes leading to dynamic interactions between reactive
power control loops.
The strongest buses identified include buses in the 400-/230-kV network in the
northeastern region of the study area (153, 33, 25, 32, 61, 94, and 108) along with
a pocket of buses at the center and southeastern portions of the study system
(136, 44, 73).
Table 5.2 Short-circuit levels (MVA) for the

400-kV transmission network.
Bus number Short-circuit level (MVA)
153 19 495
33 19 063
25 18 859
32 18 859
108 18 843
73 16 809
94 14 372
65 13 982
126 13 549
167 13 409
136 13 074
61 10 405
44 9157
5.5 Case Study 167
Table 5.3 The largest attenuation values αij computed using (5.4). Base operating case.
Node Node (attenuation value αij)
153 25 (0.069), 33(0.073), 154 (0.029), 32(0.069), 108(1.449), 156 (4.297)

73 81 (1.755), 122 (1.372), 2 (2.484), 44 (2.17), 18 (2.375), 70 (2.357), 136 (2.82)
136 13 (1.067), 26 (0.355), 18 (1.133), 44 (1.834), 81 (1.805), 126 (1.894), 22 (3.10)
94 65 (0.3601), 88 (1.006), 126 (1.649), 136 (2.135)
108 113 (2.254), 153 (1.449), 25 (1.518), 33 (1.523), 32 (1.518), 156 (2.848)
61 108 (4.51), 162 (4.346), 151 (4.623), 152 (4.623), 24 (4.96), 22 (5.24), 153 (1.449),
32 (1.518)
167 183 (1.253), 186 (2.717), 167 (1.426), 170 (1.974), 168 (1.949), 126 (3.026), 94 (4.335)
126 94 (1.649), 65 (1.98), 22 (1.206), 136 (1.894), 26 (1.889)
Table 5.3, in turn, shows the electrical distances αij computed using the sensitiv-
ity relation:
ΔV i ∂V i ∂V j
αij = =
ΔV j ∂Q j ∂Q j
Results are found to be in good agreement with short-circuit simulations in

Table 5.2.
Sensitivity studies were conducted to calculate the attenuation values for the
network under consideration. A sample of these results is shown in Figure 5.6 indi-
cating that the attenuation matrix is numerically sparse. Carefully analyzed and
12
Attenuation value [αij]
8
6
4
108
2
32, 25, 33
153
0
0 20 40 60 80 100 120 140 160 180

Bus
Figure 5.6 Sensitivity coefficients associated with bus 153 (unnormalized values).
Table 5.4 Candidate pilot buses selected using the information from electrical distances.
Zone Bus (Atot i )
1 153 (51 413), 32 (51 438), 25 (51 438), 33 (51 440)

2 73 (48 905), 81 (48 168), 122 (49 746), 2 (50 421)
3 136 (47 464), 26 (47 724), 18 (47 837), 13 (47 865), 94 (47 956), 126 (48 041), 65 (48
193), 88 (48 294), 44 (48 917), 22 (49 287)
4 167 (49 666)
5 61 (51 574)
6 108 (50 733)
7 156 (51 472)
sorted, electrical distances can be used to divide the system into several voltage
control zones. Candidate pilot buses selected from this analysis include buses
153, 73, 136, 167, 61, 108, and 156 (refer to Table 5.4).
Based on the results of this exploratory analysis, seven major voltage control areas
or electrical zones were initially identified for the regional voltage control scheme:
Zone 1 consists of major transmission and generation buses in the northeastern

part of the system. These include buses 32, 25, 153, and 33. From the analytical
results in Table 5.3, bus 153 is selected as the pilot bus for monitoring voltage
within this zone.
Zone 2 groups buses in the southernmost part of the study regions and includes
buses 2, 73, and 122. Bus 73 is selected as the pilot bus.
Zone 3 consists of buses 44, 26, 18, 70, 13, 36, 94, 65, 22, and 126. Within this zone, it
is possible to identify several groups or subclusters. The analysis of electrical dis-
tances in Table 5.3 identifies bus 136 as a candidate pilot bus.
Zone 4 located in the northern portion of the study area comprises bus 167 and a
pocket of buses in areas 4 and 5.
Zone 5 includes bus 61 and other lower-voltage buses.
Zone 6 comprises bus 108 and neighboring buses.
Zone 7 comprises bus 156 and neighboring buses.
Figure 5.7 shows the spatial distributions and principal transmission resources
associated with these voltage control zones. In practice, clusters can be combined if
they share certain properties or lack of generation sources.
Further, Table 5.5 lists the bus pilot nodes selected for analysis taking into
account generation resources, suggesting that bigger areas could be obtained by
relaxing the initial design criteria.
5.5 Case Study 169
105
Zone 6
156 25
Zone 1
153 32
108
87 113 33
Zone 5
7 29 61 151
36
83 173 21
34
162
Zone 4 65 22 24
94 39 Zone 3
20 88 4 160
1 8
18
167 125 149 136 53
26 70
170 165 SVC-2 100
17 13
81 69
84 148 44 73 66
SVC-1 150 2 Zone 2
10 40 122
101 90
117 15 132
35 67
57
124
Figure 5.7 Estimated spatial distribution of voltage control areas based on sensitivity
analysis.
Table 5.5 Bus pilot nodes selected from the identification of

reactive power sources for each voltage control area.
Zone Pilot node
1 153
2 73
3 136
4 167
5 61
In the studies that follow, complementary studies using data-driven techniques

are conducted. Primary goals are to examine the inclusion of renewable generation
to existing reactive control schemes.
5.5.2 Data-Driven Analysis

Selected system measurements were used to develop measurement-based
approaches to voltage–VAR partition. The data used in this analysis consist of
Table 5.6 Selected system measurements.
Data modality Features
Bus voltage magnitudes (XV) 173 signals
Bus voltage magnitude phases (Xθ) 173 signals
Generator reactive power output 105 generator reactive power output signals
signals (XQg)
SVC reactive power output signals Three reactive power output signals (SVC-1,
(XSVC) SVC-2, and SVC-3)
bus voltage magnitudes and phases, and generator and SVC reactive output power.
Table 5.6 summarizes the main characteristics of signals selected for analysis.
For each of these scenarios, a set of six contingencies of interest was examined
for system characterization (see Table 5.7).
Based on these results, the snapshot (measurement) data is defined as:
T
X V = V 1 V 2 … V 173 T
= X busvolt X buvoltwfs
X θ = θ1 θ2 … θ173 T 5 17
T
X Q = Q1 Q2 … Q105 T
= X Qgen X Qsvc
Table 5.7 Contingency scenarios selected for analysis.
Contingency
scenario Description Remarks
CE01 Generation No-fault, generation outage, in Area 7

outage
CE02 Generation No-fault, generation outage in Area 3
outage
CE03 Single-line No-fault, line tripping. Circuit 1 of tie-line between
outage Area 6 and Area 7 (study area)
CE04 Single-line No-fault, line tripping. Tie-line between Area 6 and
outage Area 3
CE05 Load 2% load shedding at bus 100
shedding
CE06 Double-line No-fault, double-line tripping
outage
5.5 Case Study 171
in which wfs denotes wind farms, and V j = V j t 0 V j t 1 … V j t N ,

V j = V j t 0 V j t 1 … V j t N , where Q j = V j t 0 V j t 1 … V j t N ,
and N = 3603. In this case, the data need to be concatenated.
Figures 5.8–5.11 show selected simulations for CE06 in Table 5.8. These simula-
tions are representative of other system responses and are selected to stimulate lin-
ear system response. N = 3600 snapshots corresponding to 30 sec are used for all
contingency scenarios in Table 5.7.
5.5.3 Measurement-Based Reactive Control Areas

5.5.3.1 Diffusion Maps
Earlier studies using DMs showed that projection of measured data onto a low-
dimensional manifold can capture system motion with only a limited number
of interacting degrees of freedom (eigenvectors) [29]. The goal of this section is
to examine the application of nonlinear reduction methods to determine reactive
power control areas.
Following the general approach outlined in Section 5.4.2, a two-step hierarchical
procedure for identifying voltage control areas and the associated reactive power
sources was explored. The first step identifies a reduced-order model using the rel-
evant DM coordinates a k-means clustering technique. In the second step,
4
Q [MVAr]
2 G. 25 U1
G. 25 U2
0 G. 25 U3
–2
6 7 8 9 10 11 12 13 14 15
1
Q [MVAr]
G. 32 U1
0 G. 32 U2
–1 G. 32 U3
–2
6 7 8 9 10 11 12 13 14 15
34
Q [MVAr]
32 G. 35 U1
G. 35 U2
30 G. 35 U3
28 G. 35 U4
G. 35 U5
6 7 8 9 10 11 12 13 G. 35 U6 15
Q [MVAr]
39
38
G. 156
37
6 7 8 9 10 11 12 13 14 15
Time [sec]
Figure 5.8 Reactive power deviations following a double-line outage for contingency
scenario CE06.
Bus 2
Q [MVAr]
–13.5
–14
–14.5
–15
6 7 8 9 10 Bus 73 11 12 13 14 15
Q [MVAr]
–8
–9
6 7 8 9 10 Bus 81 11 12 13 14 15
Q [MVAr]
15
14
6 7 8 9 10 Bus 100 11 12 13 14 15
Q [MVAr]
0.5
0
–0.5
6 7 8 9 10 11 12 13 14 15
Time [sec]
Figure 5.9 Reactive power output for generators in zone 2 for the contingency
scenario CE06.
10
–10
–20
SVC 1
–30 SVC 2
SVC 3
–40
–50
–60
–70
6 7 8 9 10 11 12 13 14 15
Time [sec]
Figure 5.10 SVC reactive output power for contingency scenario CE06.
generators, renewable generation, and SVCs having a large strength of interaction

with the selected voltage control areas are determined.
The whole process can be summarized as follows:
1) Construct the Markov matrix M from the observational data, X;

2) Perform spectral analysis of the Markov probability transition matrix. Deter-
mine the intrinsic dimensionality and modal properties;
5.5 Case Study 173
1.05
1.04
1.03
1.02
Voltage [pu]
1.01
0.99
0.98
0.97
6 7 8 9 10 11 12 13 14 15
Time [sec]
Figure 5.11 Bus voltage magnitudes for contingency scenario CE06.
Table 5.8 Clustering of bus voltage signals using c-means.
Cluster Buses
1 2, 122, 132, 124
2 40, 67
3 8, 53, 69, 160
4 13, 18, 26, 73, 44, 81, 100, 136
5 15, 90, 150
6 1, 4, 24, 39, 151, 162,
7 17, 84, 101, 117, 110,170
8 13, 18, 26, 70, 73, 81, 100, 136, 148,
9 25, 32, 33, 153
10 34, 88, 94, 125, 149, 167
m = 10 and contingency scenario CE06.
3) Construct the mapping from the original space to the DM space (refer to
Figure 5.4):
Ψdε = ψ 1 ψ 2… ψd ,
4) Cluster the DM embedding using k-means.

105
156
25
153 32
108
87 113 33
7 29 61 151
36
83 173 21
34
162
22 24
65 94 39
20 88 4 160
1 8
18
167 125 149 136 53
26 70
170 165 SVC-2 100
17 SVC-1 13
81 69
84 148 44 73 66
150 2
10 40
101 90 122
15 132
117
35 67
57 SVC-3 124
Wind farms
Figure 5.12 Approximate boundaries and geographical locations for clusters extracted
using DMs. Modified case with an SVC at bus 40 and a large cluster of WFs connected at
bus 67.
Figure 5.12 shows a schematic representation of the study system showing

the approximate location of 10 voltage control zones for contingency scenario
CE06 above determined using this procedure. As discussed below, similar
results are obtained for other contingency scenarios which are not dis-
cussed here.
For the ease of visualization, 60 out of the original 173 buses in the database
are selected for display. Also, the base case scenario was modified to include an
additional SVC (SVC 3) as well as the representation of a cluster of wind farms in
the vicinity of bus 67. While not discussed in detail, the identified voltage control
zones largely coincide with the critical areas determined for other contingency
scenarios.
By further adjusting the diffusion kernels, ϵi, ϵj voltage control areas can be fur-
ther subdivided into two or more control areas or merged into a bigger area.
Observe that two control areas in the neighborhood of major SVCs are identified
associated with the location of large wind farms. Control strategies associated with
this voltage control zone are discussed in Chapter 7.
5.5 Case Study 175
5.5.4 Direct Clustering

The problem of direct clustering of measured bus voltage deviations is addressed
by comparing the performance of c-means fuzzy analysis with the results of diffu-
sion-based clustering in Figure 5.12. Table 5.8 gives the clusters determined using
c-means fuzzy clustering. For this comparison, the number of clusters, m, was cho-
sen based on the previous application of diffusion methods in Figure 5.12. Results
are found to be in good agreement.
To verify the accuracy and suitability of the models and the appropriateness of
the selected operating scenarios, comparisons are provided with other analysis
methods. Dynamic mode decomposition is first used to determine the buses hav-
ing the largest participation in the oscillations. These methods use the voltage-
based mode shapes of the dominant oscillatory modes extracted from selected
bus voltage magnitude recordings.
Table 5.9 shows the top 10 entries determined using the dynamic mode decom-
position (DMD) method in Ref. [41]. It is noted that DMD results are highly con-
sistent regardless of the location or severity of the contingencies analyzed. Also,
Figure 5.13 gives a comparison between the candidate pilot bus locations deter-
mined conventional sensitivity analyses in Table 5.4 with those identified using
DMD in Table 5.9.
For comparison, Figure 5.14 shows the leading eigenvector of the DMs for con-
tingency scenarios CE04 and CE05; it is observed that while the modes excited
depend on the location of the disturbance and the structure of the network,
reduced-order models can still accurately capture relevant system behavior.
Table 5.9 Top 10 (unnormalized) entries of the voltage-based DMD eigenvector

(dominant mode).
Contingency
scenario Bus (magnitude of mode shape)
CE01 153 (0.259), 32 (0.220), 31 (0.175), 24 (0.156), 152 (0.156), 108 (0.146),
156 (0.141), 113 (0.130), 33 (0.128), 25 (0.111), 36 (0.102), 106 (0.126),
95 (0.083), 87 (0.086)
CE02 32 (0.229), 153 (0.229), 156 (0.195), 111 (0.172), 31 (0.166), 108 (0.158),
24 (0.137), 110 (0.153), 157 (0.152), 152 (0.139)
CE03 32 (0.227), 153 (0.226), 111 (0.168), 31 (0.162), 107 (0.154), 24 (0.139),
110 (0.152), 156 (0.149), 25 (0.132), 109 (0.135)
CE04 32 (0.212), 153 (0.212), 111 (0.158), 155 (0.173), 31 (0.161), 107 (0.147),
157 (0.141), 110 (0.144), 152 (0.135), 24 (0.135)
CE05 159 (0.132), 65 (0.123), 54 (0.108), 7 (0.106), 68 (0.103), 25 (0.093),
33 (0.094), 8 (0.088), 99 (0.094), 160 (0.096)
105
Sensor placement (DMD)
156 25
108 153 32 Pilot node
87 113 33
7 29 61 151
36
83 173 21
34
162 24
65 20 94 22
88 4 39
1 8 160
18
167 126 149 136 70 53
26 100
170 165 SVC-2 13 SLC
17
81 69
84 148 44 73 66
SVC-1 150 2
10 40
101 122
LCP 90 132
117 15
35 67
57
124
Figure 5.13 The candidate pilot node locations. Empty dashed circles show pilot node
locations from Table 5.5. Filled gray squares show the candidate node locations obtained
from DM information in Table 5.9.
Similar results are obtained using the MCL algorithm in Figure 5.15. In this case,
the leading eigenvector is obtained solving the standard eigenvector equation:
M MCL r − λMCL2 I ϕMCL2 = 0
where the power coefficient r is set to r = 4, and ϕMCL2 denotes the second eigen-
vector in the spectral decomposition.
5.5.5 Correlation Analysis

The performed studies have shown that spectral techniques can be used to identify
voltage coherent areas and the associated dominant buses. This section focuses on
how to extend data-driven techniques to account for interactions between bus volt-
age deviations and reactive power sources.
The simplest approach to determine correlation measures between two data
matrices X and Y can be obtained from the correlation matrix C = X T Y N. Reac-
tive power and voltage data, however, are intrinsically dissimilar in nature and
some sort of scaling or normalization may be required. Also, data fusion may
be needed to allow for the analysis of true interactions between the data
modalities.
5.5 Case Study 177
(a)
×10–3
4
0
Real part of mode
–2
–4 Bus 156
Bus 33
–6
Bus 108
–8 Bus 32 Bus 153
Bus 25
–10
Bus
(b)
×10–3
2 Bus 69 Bus 100 Bus 160
Bus 2 Bus 53
1 Buses
20,73
–1
Magnitude
Bus 167
Bus 113
–2
–3
Bus 33
–4
Bus 156
Bus 108 Bus 153
–5 Bus 25 Bus 32
–6
Bus
Figure 5.14 Leading eigenvector of the Markov transition matrix: (a) contingency scenario
CE04 and (b) contingency scenario CE05 (unnormalized values).
In this section, basic correlations measures are extended to allow simultaneous

analysis of multiple datasets within the framework of data fusion and diffusions
maps. Three approaches to the analysis of correlation measures are discussed here:
1) Direct analysis of concatenated data;

2) The application of multiview DMs;
3) Consensus analysis of multivariate data.
0.1
–0.1
Real part of mode
Bus 108
–0.2 Bus 156
–0.3
–0.4
–0.5
Bus 25 Bus 33 Bus 153
–0.6
Bus
Figure 5.15 Dominant eigenvector extracted using the MCL for contingency
scenario CE04.
5.5.5.1 Direct Analysis of Concatenated Data

A simple approach to enable a comparison of multitype data of complementary
nature is the use of detrending techniques. Along the same vein as Ref. [33],
the row-wise concatenated data are defined as:
X = XV X Qgen X Qsvc 5 18
Because of the non-homogeneity of the data, data fusion techniques need to be

utilized to reveal important relationships between pilot nodes and reactive power
outputs. Alternatively, conventional normalization techniques can be combined
with detrending techniques to allow a joint analysis of multimodal data. This is
the approach adopted here.
In the first attempt to determine correlation measures, normalization techni-
ques were applied to (5.18). Assume to this end, that the individual data types
in (5.18) are independently normalized. In the second step, the overall data matrix
X is detrended using linear or nonlinear detrending techniques.
The following three-step procedure is used to determine sensitivity relations
between bus voltage deviations and reactive power outputs:
1) Normalize the individual data matrices in (5.18);

2) Detrend the data using wavelet shrinkage or a nonlinear detrending technique;
3) Determine a reduced-order model using data fusion. Compute the distance
matrix M and determine the distance coefficients dij between generation buses,
SVC buses, and wind farm buses, and bus voltage magnitudes using (5.13).
5.5 Case Study 179
Table 5.10 The five largest interaction coefficients mij

computed using (5.5) for the base operating case.
Bus Bus (diffusion coefficients αij)
153 25, 32, 33, 35, 156, 113

73 2, 811, 122, 100
136 26, 148, 81, 73, 2
167 17, 61, 87, 26, 136
61 61, 62, 94, 22, 125
Table 5.10 shows the five largest diffusion coefficients obtained using this
approach. Comparisons with reactive power sensitivities show that DMs can pro-
vide an accurate characterization of the strength of interactions between bus volt-
age magnitudes and reactive power reserves and also have a strong physical
interpretation. It is noted that the adopted approach identifies both SVCs and gen-
erators showing the largest strength of interaction with nearby buses.
The practical application of these concepts to complex power systems with
increased wind penetration is deferred to Chapter 8.
5.5.5.2 Two-Way Correlation Analysis

With reference to Eq. (5.18), let X = XV and Y = X Qwfs X Qgen X Qsvc . Motivated
by the notion of multiview DMs [42], the concatenated data are defined as:
O1 XTY
X= 5 19
YTX O2
where O1 and O2 are matrices of zeros of appropriate dimensions. Observe that

other formulations such as row-concatenated data can be used for the case of more
than two data types.
5.5.5.3 Partial Least Squares Correlation

Given two data matrices X ᑬN × p and Y ᑬN × m , this technique analyzes the
correlation between matrices X and Y. Following the formulation of Krishnan
et al. [43] and Abdi and Williams [44], the cross-block correlation matrix, R, is
defined by:
T
R= Y X 5 20
Singular value decomposition (SVD) of R yields
R = UΣV T 5 21
Here, U and V are orthonormal matrices called saliences containing the left
and right singular vectors of R, respectively, and Σ is a diagonal matrix of the
nonzero singular values. The latent variables, Lx and Ly, which are linear combi-
nations of the original values, are obtained by projecting the data matrices onto
their respective saliences as:
Lx = XV
and
Ly = YU
Several observations are of interest here:
•• Matrices Lx and Ly give the datasets spatial patterns or shapes;

The correlation matrix R gives the correlation between a given output signal, Qj,
and a given input signal, Vk. Accordingly, the correlation coefficient Rij between
ith bus’s voltage magnitude with respect to the ith reactive power output is
obtained from the rows of the correlation matrix.
When coupled with real-time information, these techniques have also the poten-
tial to correlate reactive power reserves with bus voltage deviation and provide
operational warnings.
–1
Real part of mode
–2
Bus 148
–3 units 1–2
–4
Bus 73
–5 units 1,2,4,5
Bus 2 Bus 100
units 1–5 Bus 81 units 1–6
–6 units 1–6
–7
0 10 20 30 40 50 60 70 80 90 100
Generator
Figure 5.16 Partial least squares regression of datasets X and Y.

References 181
Figure 5.16 shows partial least squares regression results for data matrices let
X = XV and Y = X Qwfs X Qgen X Qsvc . Results correlate well with previous results
in Figures 5.14 and 5.15.
This example demonstrates that correlation analyzes may identify key relation-
ships between observed bus voltage deviations and output reactive power from
wind farms and synchronous generators, such as sensitivity relationships, or levels
of strength associated with reactive power control areas. It is found that both sim-
ple correlation analysis and partial least squares regression perform well for the
simulated records.
5.6 Summary
Over the last few decades, various forms of voltage monitoring and reactive power
management systems have been developed. These architectures offer the possibil-
ity to calculate in near-real-time several aspects of voltage control and VAR man-
agement such as distances to voltage collapse and network reactive control,
distances to instability, and sensitivity calculations. The (geographical) dispersion
of dynamic recorders and the changing network structure and operating condi-
tions, however, create wide-area voltage monitoring issues that must be addressed
using advanced data-based analysis approaches and correlation techniques.
In this chapter, a new use of spectral analysis tools for dimensionality reduction
and clustering of high-dimensional datasets, based on the normalized graph Lapla-
cian has been introduced. The proposed technique combines the inherent abilities
of graph-theoretical techniques with spectral clustering and visualization methods
to identify loosely interconnected voltage control areas and reconstruct system
behavior using selected measurements. Attention has been focused on three main
aspects, namely the identification of critical system zones showing a coherent
behavior and the associated reactive power sources, the use of spectral analysis
to identify the critical buses, and the computation of reduced-order models.
Issues such as reserve capacity control and the efficient utilization and coordi-
nation of reactive power sources need to be addressed to achieve fast voltage mon-
itoring and control and keep the reactive power output margin against system
contingencies.
References
1. Coordinated Voltage Control in Transmission Networks. (2007). CIGRE Task
Force C4.602 (February 2007).
2. Corsi, S., Possi, M., Sabelli, C., and Serrani, A. (2004). The coordinated automatic
voltage control of the Italian transmission grid - Part I: reasons of the choice and
overview of the consolidated hierarchical system. IEEE Transactions on Power
Systems 19 (4): 1723–1732.
3. Cañizares, C.A., Cavallo, C., Pozzi, M., and Corsi, S. (2005). Comparing secondary
voltage regulation and shunt compensation for improving voltage stability and transfer
capability in the Italian power system. Electric Power Systems Research 73: 67–76.
4. Lagonotte, P., Sabonnadiere, J.C., Léost, J.Y., and Paul, J.P. (1989). Structural
analysis of the electrical system: application to secondary voltage control in France.
5. Taylor, C.W. (1994). Power System Voltage Stability. McGraw-Hill.
6. Taylor, C.W. (1993). Survey of effective and practical solutions for longer-term
voltage stability. Electrical Power and Energy Systems 15 (4): 217–220.
7. Taylor, C.W. (2000). The Future in on-line security assessment and wide-area
stability control. Proceedings of the 2000 IEEE Power Engineering Society Winter
Meeting.
8. AEMO. (2013). Australian energy market operator, wind turbines plant capabilities
report. Wind Integration Studies. https://www.aemo.com.au/-/media/Files/PDF/
Wind_Turbine_Plant_Capabilities_Report.pdf/.
9. Miller, N.W., Guru, D., and Clark, K. (2008). Wind generation applications for the
cement industry. Proceedings of the IEEE Cement Industry Technical Conference
Record.
10. Miller, N., MacDowell, J., Chmiel, G. et al. (2012). Coordinated voltage control for
multiple wind plants in eastern wyoming: analysis and field experience.
11. Ullah, N.R. and Bhattacharya, K. (2008). Wind farms as reactive power ancillary
service providers—technical and economic issues. IEEE Transactions on Energy
Conversion 24 (3): 661–672.
12. Taranto, G.N., Martins, N., Falcao, D.M. et al. (2000). Benefits of applying
secondary voltage control schemes to the Brazilian system. Proceedings of the 2000
Power Engineering Society Summer Meeting (July 2000).
13. Sancha, J.L., Fernandez, J.L., Cortez, A., and Abarca, J.T. (1996). Secondary voltage
control: analysis, solutions and simulation results for the Spanish transmission
system. IEEE Transactions on Power Systems 11 (2): 630–638.
14. Paul, J.P., Leost, J.Y., and Tesseron, J.M. (1987). Survey of the secondary voltage
control in France: present realization and investigations. IEEE Transactions on
Power Systems PWRS-2 (2): 505–511.
15. Kundur, P. (1994). Power System Stability and Control, The EPRI Power System
Engineering Series. New York, NY: McGraw-Hill.
16. Carpentier, J.L. (1987). “CRIC”, a new active-reactive decoupling process in load
flows, optimal power flows and system control. IFAC Proceedings 20 (6): 59–64.
References 183
17. Zhong, J., Nobile, E., and Bose, A. (2004). Localized reactive power markets using
the concept of voltage control areas. IEEE Transactions on Power Systems 19 (3):
1555–1561.
18. Cotilla-Sanchez, E., Hines, P., and Barrows, C. (2013). Multi-attribute partitioning
of power networks based on electrical distance. IEEE Transactions on Power
Systems 28 (4): 4979–4987.
19. Tyuryukanov, I., Popov, M., van der Meijden, M.A.M.M., and Terzija, V. (2018).
Discovering clusters in power networks from orthogonal structure of spectral
embedding. IEEE Transactions on Power Systems 33 (6): 6441–6451.
20. Jiang, T., Bai, L., Ji, L., and Li, F. (2007). Spectral clustering-based partitioning of
volt/VAR control areas in bulk power systems. IET Generation, Transmission &
Distribution 11 (5): 1126–1133.
21. Saugata, S., Biswas, T., and Srivastava, A.K. (2014). Performance analysis of a new
synchrophasor based real time voltage stability monitoring (RT-VSM) tool.
Proceedings of the 2014 North American Power Symposium (NAPS).
22. Sun, H., Guo, Q., Zhang, B. et al. (May 2013). An adaptive zone division-based
automatic voltage control system with applications in China. IEEE Transactions on
Power Systems 28 (2): 1816–1828.
23. Morison, K., Wang, X., Moshref, A., and Edris, A. (2008). Identification of voltage
control areas and reactive power reserve; an advancement in on-line voltage
security assessment. Proceedings of the 2008 IEEE Power and Energy Society
General Meeting – Conversion and Delivery of Electrical Energy in the
21st Century.
24. Roman-Messina, A. (2015). Wide-area Monitoring of Interconnected Power Systems,
IET, Power and Energy Series 77, Stevenage. UK: IET, The Institute of Engineering
and Technology.
25. Li, H., Li, F., Xu, Y., Y. et al. (2010). Adaptive voltage control with
distributed energy resources: algorithm, theoretical analysis, simulation,
and field test verification. IEEE Transactions on Power Systems 25 (3):
1638–1647.
26. Messina, A.R. and Vittal, V. (2007). Extraction of dynamic patterns from wide-area
measurements using empirical orthogonal functions. IEEE Transactions on Power
Systems 2 (2): 682–692.
27. Newman, M. (2010). Networks, 2e. Oxford, UK: Oxford University Press.
28. Dray, S., Legendre, P., and Peres-Neto, P.R. (2006). Spatial modelling: a
comprehensive framework for principal coordinate analysis of neighbour matrices
(PCNM). Ecological Modelling 196: 483–493.
29. Arvizu, C.M. and Messina, A.R. (2016). Dimensionality reduction in transient
simulations: a diffusion maps approach. IEEE Transactions on Power Delivery
31 (5): 2379–2389.
30. David, G. and Averbuch, A. (2012). Hierarchical data organization, clustering and
denoising via localized diffusion folders. Applied and Computational Harmonic
Analysis 33 (1): 1–23.
31. Griffith, D. (2000). Eigenfunction properties and approximations of selected
incidence matrices employed in spatial analyzes. Linear Algebra and Its
Applications 321 (1-3): 95–112.
32. Moghadas, S.M. and Jaberi-Douraki, M. (2019). Mathematical Modelling. Hoboken,
NJ: Wiley.
33. Román-Messina, A. (2020). Data Fusion and Data Mining for Power System
Monitoring. Boca Raton, FL: CRC Press.
34. S. Van Dongen, Graph Clustering by Flow Simulation. PhD thesis, University of
Utrecht, May 2000.
35. Enright, A.J., Van Dongen, S., and Ouzounis, C.A. (2002). An efficient algorithm
for large-scale detection of protein families. Nucleic Acids Research 30 (7):
1575–1584.
36. Malmstrom, R.D., Lee, C.T., Van Wart, A.T., and Amro, R.E. (2014). Application of
molecular-dynamics based Markov state models to functional proteins. Journal of
Chemical Theory and Computation 10: 2648–2657.
37. Cazade, P.A., Zheng, W., Prada-Garcia, D. et al. (2015). A comparative analysis of
clustering algorithms: O2 migration in truncated hemoglobin I from transition
networks. The Journal of Chemical Physics 142: 025103-1-15.
38. Bezdek, J.C. (1981). Pattern Recognition with Fuzzy Objective Function Algorithms.
New York, NY: Plenum Press.
39. Bezdek, J.C., Ehrlich, R., and Full, W. (1984). FCM: The Fuzzy c-means clustering
algorithm. Computers and Geosciences 10 (2–3): 191–203.
40. Zadeh, L.A. (1971). Similarity relations and fuzzy orderings. Information Sciences
3: 177–200.
41. Barocio, E., Pal, B.C., Thornhill, N.F., and Roman-Messina, A. (2015). A dynamic
mode decomposition framework for global power system oscillation analysis. IEEE
42. Lindenbaum, O., Yeredor, A., Salhov, M., and Averbuch, A. (2020). Multi-view
diffusion maps. Information Fusion 57: 127–149.
43. Krishnan, A., Williams, L.J., b., McIntosh, A.R., and Abdi, H. (2011). Partial least
squares (PLS) methods for neuroimaging: a tutorial and review. Neuroimage
5: 455–475.
44. Abdi, H. and Williams, L.J. (2013). Partial least squares methods: partial least
squares correlation and partial least square regression. In: Computational
Toxicology, Methods in Molecular Biology, 930, vol. II (eds. B. Reisfeld and A.N.
Mayeno), 543–579. New York, NY: Human Press-Springer.
185
Advanced Control Synthesis
Modern power grids, characterized by high penetration of inverter-interfaced gen-

eration sources, suffer from a lack of rotational inertia and governor control, which
may cause faster frequency dynamics and larger frequency deviations and even
instability problem. Over the last few years, several inertia-emulation-based con-
trol strategies have been developed and applied to support system frequency in
power systems with significant renewable generation. Conceptually, those control
strategies manipulate the converter power injections in response to local frequency
deviations to support system stability.
In Chapter 4, the problems of inertia emulation and placement were discussed,
based on the notion of virtual synchronous generator (VSG) control. In this
chapter, advanced inertia-emulation-based control strategies to enable high
penetration levels of microgrid (MG) are discussed. These approaches combine
an adaptive energy storage system (ESS) dispatch strategy with an MG-controlled
islanding scheme to provide frequency support and enhance small-signal stability
and voltage regulation.
6.1 Introduction
Power grids worldwide are experiencing a significant transformation, arising from

the increased penetration of distributed generations (DGs). As discussed in previ-
ous chapters, reduced rotational inertia in the grid may adversely affect frequency
response and system control and lead to degrade performance of traditional
control schemes. This, in turn, may result in large frequency and voltage devia-
tions and, potentially, load shedding and instability [1–5]. Advanced control of
grid-connected MGs, however, has the potential to offset the intermittent nature
of distributed energy resources and provide stability support to the host utility
during emergency conditions, which is emphasized in this chapter.
186 6 Advanced Control Synthesis
6.2 Frequency Dynamics Enhancement
6.2.1 Background: The Concept of Flexible Inertia

In Chapter 2, the concept of center of gravity (COG) was introduced to study the
long-term power–frequency transients following large perturbations. Using the
COG concept, the original power system can be represented by the simplified
equivalent model of Figure 6.1 [6].
To develop an advanced ancillary control scheme for frequency support, con-
sider an equilibrium mechanical system with equilibrium point δ0. The equilib-
rium condition requires that the resultant torque in the COG vanishes, namely [7],
F 1 cos δL1 − F 2 cos δL2 ; F k = mk ghk , k = 1, 2 61
where F, L, and δ denote the applied force, the lever arm length, and the angle
between the beam and the horizon, respectively. Figure 6.2 shows a schematic rep-
resentation of this model.
60 Hz Area 2
Intertial
response
Nadir
Local COI
dynamics

tie
PCOI 1,COG tie
PCOI 2,COG
δCOI1, MCOI1
COG [∆PL – ∑ PCOI
tie
– ∆PiMMG]
Area n
tie
PCOI δCOG , MCOG MCOG j
i,COG
j,COG Mi (∆Pmech –∆Pelec)
tie ΔfCOIi = ΔfCOG
PCOI M 2πD
n,COG 1 + COG Δf
Mi (∆Pmech –∆Pelec) COG
Area i
δCOIk, MCOIk
[∆PL – ∑ PCOI
tie
,COG – ∆Pi
MMG]
i
Mi j
Mj (∆Pj –Pijtie – ηj∆PiMMG)
Δfj = ΔfCOI
1 + Mi 2πD i
Δf conv. ESS+
Mj (∆Pj – Pijtie– ηj∆PiMMG) COIi δCOIi, MCOIi + M MMGs
Figure 6.1 Conceptual overview of the adopted control scheme showing the grid-parallel
configuration of interconnected MGs.
L1
L2
F1cosδ δ
h1 F1
h2
COG
F2
Figure 6.2 Mechanical system explaining the adopted control scheme.
Equation (6.1) establishes that any change in L should be compensated by F and

vice versa to keep the equilibrium point δ0.
The same reasoning used in (6.1) and Figure 6.2 may be applied to the equivalent
model of Figure 6.1. In analogy to a mechanical system, the inertia M can be inter-
preted as the mass, the fictitious reactance X is the length of the lever arm, and the
inertia-weighted disturbance ΔPi/Mi is a force.
The size of the inertia-weighted perturbation will cause a frequency variation,
given by the swing equation of the form [8]
df t Tm t − Te t
= 62
dt 2M
where Tm(t) and Te(t) are the time-varying mechanical input torque and the elec-
trical output torque, respectively. Incorporating the tie-line flows, (ΣPtie), as well as
the load dynamics (D), into (6.2) results in [3, 9]
1
f = ΔP − 2πDΔf − Ptie 63
2πM
where Ptie in the equivalent system of Figure 6.1 has the form [10]
V COI i V COG
Ptie
COI i ,COG = sin δCOI i − δCOG 64
X tie
COI i ,COG
Furthermore, the fictitious reactance Xtie can be calculated using (2.28), i.e.
min S
s = 1 ps ξs− − ξs+
X tie
COI , COG
i
st
n n
1 V COI i V COG
sin δCOI i − δCOG + ξs− − ξs+ = ΔPMMG
ω
i = 1 COG X tie
COI i ,COG i=1
i,s
−
ξs , ξs+ ≥ 0
0 < X tie
COI i ,COG ≤1
65
According to (6.3) and (6.5), the Eq. (6.1) for the equivalent system of Figure 6.1
can be reorganized as
ΔP1 ΔP2 ΔP3 ΔPn
X1 + X2 + X3 + … + Xn = 0 66
M1 M2 M3 Mn
The equilibrium condition (6.6), as the basic principle of the inertia-based
control scheme for frequency support, relies on the following simple yet effective
premise [11]: “Any variation in the equivalent reactance, X in Figure 6.1, following
a disturbance, will be compensated by a corresponding change in ΔPi/Mi so that
the equilibrium condition is satisfied.” Area i in Figure 6.1 schematically describes
the variation of F, through manipulation of M, in the equivalent system. Referring
to Figure 6.1 and taking the contribution of multi-MGs (MMGs) into account, the
swing Eq. (6.3), relative to the COG, can be expressed in the form of (6.7) [11].
1
f COI i = ΔPi − 2πDi Δ f COI i − Ptie
COI i ,COG − ΔP i
MMG
67
2πM COI i
Equation (6.7) establishes that frequency deviations exceeding a predetermined

threshold can be limited by suitable modifying the center of inertia (COI)
inertia, MCOIi. This, in turn, results in a change in PtieCOIi,COG according to (6.3),
which further affects frequency dynamics (see Eq. (2.4) in Chapter 2). Moreover,
(6.7) suggests that the proper transition of MMGs from the grid-connected
mode to the islanded mode can be used to aid frequency support. Generally stated,
for area i, the total inertia can be explicitly decomposed into three constituent
components, as:
M i = M Conv
COI i + M
ESS
+ M MMGs 68
ESS
where M Conv
COI i , M , and MMMGs represent, respectively, the conventional synchro-
nous inertia and the inertia provided by the ESS and MMGs.
Key parameters of interest to be controlled via manipulation of (6.8) include the
frequency nadir, the rate of change of frequency (RoCoF), and the frequency devi-
ation during a given time interval of interest. Using this framework, the ESS output
is manipulated to reduce the RoCoF and the frequency nadir through the emulated
virtual inertia. Disconnecting the MMGs from the network (islanding) in a
time horizon greater than the response time of ESS, on the other hand, increases
the frequency evolution to satisfy the rolling window criterion. The effects of the
MMGs islanding strategy on the frequency response are examined based on (6.8)
for two different operating conditions of the MGs, i.e. power import from the grid
and power export to the grid. These cases are discussed separately as follows [11]:
1) Power import operating mode: In the inertial response period where Tm < Te,
the MG islanding causes the torque Te to decrease, which in turn decreases the
RoCoF and nadir. On the other hand, for the time interval beyond the nadir,
where Tm > Te, islanding leads to a greater acceleration torque and thus a faster
frequency recovery.
2) Power export operating mode: In this case, the MG islanding, reinterpreted as
loss of inertia according to (6.8), renders frequency dynamics faster. The iner-
tia-based control strategy imposes the islanding of the MMGs beyond the fre-
quency arrest period.
6.2.2 Frequency Dynamics Propagation

Inertia manipulation via coordinated control action of ESS and MMGs relies on
the propagation of frequency dynamics in the reduced equivalent system, repre-
sented by Figure 6.1 and (6.5) [see Equations (2.15) and (2.17) in Chapter 2]. Using
this notion, it is assumed that any disturbance may be aggregated and applied to
the COG. As discussed in Chapter 2 and [12], any change in the instantaneous sys-
tem frequency following a disturbance can be presented in the COG formulation,
given by (6.9).
df COG t
2M COG = ΔR t − ΔPL 69
dt
Note that (6.9) incorporates the additional power delivered through frequency
response, i.e. R [MW], into the formulation [13]. By integrating (6.9) and assuming
that the frequency response is delivered by linearly increasing the active power
with a fixed slope (R/TD) during inertial response [14], one could write
1R 2
t − ΔPL t
2T
ΔfCOG t = D 6 10
2M
where ΔfCOG, ΔPL, and TD are the COG frequency deviation, imbalance power,
and the delivery time of primary frequency response, respectively. The frequency
nadir for the COG can be calculated by setting the derivative of (6.10) with respect
to t equal to zero, namely
∂ Δ f COG t
=0 6 11
∂t
which leads to
R
t − ΔPL
TD
=0 6 12
2M
Equation (6.12) is satisfied when

ΔPL
t= 6 13
R
TD
By substituting (6.13) in (6.10), one obtains:
ΔP2L T D
Δ f COG
nadir = f nadir − f 0 =
COG
6 14
4MR
Further substitution of (6.2) into (6.14), and noting that
dΔf t
= RoCoF 6 15
dt
results in
2
dΔf
4M 2 TD
dt M RoCof 2 T D
Δ f COG
nadir = = 6 16
4MR R
All the independent parameters in (6.16) are system-dependent except for R/TD.
As a common assumption, the dependent variable Δfnadir is set to the maximum
allowable value to calculate the worst-case R/TD. Accordingly, the COG frequency
dynamics may be specified by (6.16). The problem of interest, however, is to cal-
culate the local dynamics of specified buses exhibiting undesired behavior. Alter-
natively, local behavior in (6.16) can be obtained using sensitivity relations in the
reduced equivalent system of the form (6.5). Dividing (6.7) by the classical form of
(6.9) gives
M COG ΔPi − 2πDΔ f COI i − PCOI i ,COG − ΔPi

tie MMG
df COI i
= 6 17
df COG M COI i ΔPmech − ΔPelec
where ΔPMMG defines the difference between the total generation and the local
MMG load. By knowing the COG dynamics of interest in (6.9) and (6.16), and
the frequency sensitivity relation in (6.17), one can write
M COG ΔPi − 2πDΔ f COI i − PCOI i ,COG − ΔPi

tie MMG
or equivalently,
ΔPL − COI i ,COG − ΔP i

Ptie MMG
M COG j
M ΔPmech − ΔPelec
Δ f COI i = COI i Δ f COG 6 19
M COG 2πD
1+ Δ f COG
As is apparent in (6.19), the MMGs capacity ΔPMMG affects the COI dynamics
through its effect on the power imbalance and the associated sensitivity factor.
Let now the swing Eq. (6.3) be rewritten as
1
fj= ΔP j − Ptie
ij − 2πD j Δ f j − η j ΔP i
MMG
6 20
2πM j
for bus j, where
η j = 1; j i; i 6 21
j
and ηj is defined as the ratio of the total MMGs capacity at bus j to the total MMGs
capacity in the associated area i. Following the same procedure as that in (6.19), it
can be shown that

Ptie MMG
M COI i j
Mj ΔP j − Ptie
ij − η j ΔP i
MMG
Δf j = Δ f COI i 6 22
M COI i 2πD
1+ Δ f COI i
M j ΔP j − Ptie − η ΔPMMG
ij j i
where Mj and MCOG have the same interpretation as in (6.8).
6.2.3 Inertia-Based Control Scheme

For each generator bus, the frequency nadir, the RoCoF, and the frequency evolu-
tion, calculated by (6.19) and (6.22), may be compared with those of the standard
values. For buses that frequency trends away from the acceptable values, dispatch-
ing of the ESSs and tripping of the MMGs may be used to change Mi( j) and ηj to
mitigate undesired frequency variations. Figure 6.3 shows a flowchart representa-
tion of the advanced inertia-based control scheme.
To build some intuition about the flowchart of Figure 6.3, assume that the
decline of frequency response of area i COI (generator j) deviates from the accept-
able value. Referring back to Figure 6.2, and by properly dispatching the ESSs and/
or the MMG islanding, the position of the COI equivalent generator relative to the
COG may be changed to oppose variations in frequency response. An illustration
of the inertia-based control strategy is shown in Figure 6.4. It should be empha-
sized that the transfer function of the phase-locked loop (PLL), Kf (s), used in
the voltage source converter control strategies, takes the SOC and the frequency
as inputs and represents the internal inverter control strategy behavior. Using
the approach, the virtual inertia loop would be activated upon the RoCoF and
the frequency nadir exceeds the normal values, following the inception of a fault.
Start
COG-reduced equivalent system of

Fig. 6.1.
COG dynamics calculation using

(6.10), (6.14) and (6.16)
Computation of local buses dynamics

using (6.19) & (6.22)
nadir –
fj >f?
Yes
or
nadir –
dfj df
> ?
dt dt ESS dispatching
NO
– Yes
Δfj 15–sec > Δf ?
NO MMG islanding
End
Figure 6.3 Flowchart representation of the proposed control strategy; f represents an

upper acceptable frequency value.
6.2.4 Flexible Inertia: Practical Considerations

It is mandatory to reconnect the islanded MMGs to the host grid in steady state
after the fault. Some considerations should be made regarding the synchronization
of the MMGs and the host grid to ensure a smooth voltage and current transition
Virtual Inertia Loop

SOC
MESSs Kf (s)
ΔPmech – 1 1 Δfj
+
– (MConv. + ηj MMMG)s 2π
Figure 6.4 Frequency response model with virtual inertia.
and safe reconnection of the two grids [15, 16]. The early static switch closing
causes low-frequency beat voltage across the switch. This beat voltage can cause
a significant beat current, which leads to adverse effects such as possible resonance
with mechanical structure and additional power loss. Figure 6.5 shows the beat
voltage and the current flow from the switch for the early closing of the switch.
Figure 6.5 Beat voltage for early closing of the switch; upper plot: beat voltage across the
switch, second plot: current of phase A, third plot: current of phase B, fourth plot: current of
phase C.
2
V [pu]
–2
10 16 22 28 34 40 46 52
Time [s]
Figure 6.6 Voltages on either side of the static switch [15].
Monitoring the voltage across the static switch reveals that the MMGs and the
utility should be synchronized when the phase difference between voltages of
the two grids is approximately zero. Figure 6.6 shows voltages on either side of the
static switch. From these results, the interval time between 33 and 42 seconds is an
appropriate synchronization time to close the switch [17].
6.2.5 Results and Discussions

Exploratory studies to assess the impact of the ESSs dispatching and the MMGs
islanding on the system inertia and frequency dynamics control are conducted
on two test systems: (a) a simple two-area, four-machine test system and (b) a
16-machine, five-area 68-bus test model of the New York New England (NYNE)
test system. Without loss of generality, uniform distribution for the uncertain
parameter ΔPMMGs is considered in (6.5).
1) Two-area system:
The single-line diagram of the system is shown in Figure 6.7. The disturbance
considered is the trip of 1400 MW generation, i.e. ΔPL = 14 pu, in Area 2. The
Area 1 Area 2
1 10 20 3 101 13 120 11
Gen3
Gen1
110
Gen2 Gen4
2 4 12
14

reduced equivalent model of the system is derived based on (6.5) to gain insight
into the nature of the frequency behavior. For this purpose, 10 random scenar-
ios of equal probability, i.e. Ps = 0.1 in (6.5), are generated. Minimization of
(6.5) reveals that the reduced equivalent model can be characterized by
X tie
COI 1 ,COG = 0 2134; X tie
COI 2 ,COG = 0 3172 6 23
Bus 1 frequency behavior in the conventional system, obtained by setting η = 0

in (6.22), can be calculated by substituting (6.23) in (6.19) and (6.22) as
1 67
Δf1 = Δ f COG 6 24
1 + 0 97Δ f COI 1
Similarly, for bus 12, in Area 2, one can write:
1 16
Δ f 12 = Δ f COG 6 25
1 + 0 08Δ f COI 2
Equations (6.24) and (6.25) suggest that the frequency nadir, for generator buses
in Area 2, limits the ability of the host grid to accommodate high levels of MGs
generation. A related problem of interest is that of determining the number of ESSs
as well as the generation capacity of the MMGs that can be integrated into the grid
to efficiently support system frequency. From (6.2) and for the base system, one
can write
df 1 14pu
= 6 26
dt M1
and, for the system with high penetration of MGs generation and installed ESSs,
we have
df 2 14pu
= 6 27
dt M 2 + M ESS
where M2 = M1 + MMMG. Dividing (6.27) by (6.26) gives
df 2 M1
= 6 28
df 1 M 2 + M ESS
Multiplying both sides of (6.28) in the term Δf1, on the other hand, gives
df 2 M1
× Δf1 = × Δf1 6 29
df 1 M 2 + M ESS
Equation (6.29) can be rewritten in the compact form as
M1
Δf2 = Δf1 6 30
M 2 + M ESS
in which MESS and MMMG (M2) are unknown parameters to be determined.

A similar expression for the frequency evolution during a time interval of interest
may be derived to complete the set of two equations with two unknown para-
meters. As the frequency dynamics beyond the inertial response horizon is pre-
dominantly affected by the MMGs islanding, a 15-second window starting from
the frequency nadir time is used to characterize frequency dynamics as [18]
1 1
R T n + 15 − ΔPL R T n − ΔPL
f T n + 15 − f T n = 2 MMG
− 2 6 31
M1 + M M 1 + M MMG + M ESS
where Tn is the time at which the frequency nadir occurs.
Substituting the actual values (Δf1 = 1.1, M1 = 2 × 6.5) and the marginal quan-
tities [17] (Δf2 = 0.8, Δf15-sec = 0.96) in (6.30) and (6.31) reveals that the ESSs should
provide 0.9% of the base system inertia and that the emulated inertia is
MMMG = 1.4.
Experimental results for the UOK-MG show that
PGenset 5
M MMG = 1 89 MMG
= 1 89 = 0 43 s 6 32
P 22
where PGenset and PMMG define the generated power of the synchronous-based DG
and the overall MG generation, respectively (details on how to calculate (6.32) are
given in Chapter 2), and the inertia constant in (6.32) is on a 15-kVA base. Accord-
ingly, MMMG=1.4 would be realized in the base of the system, i.e. 100 MW, by
penetration of
1 4 × 100
PMMG = × 0 022 = 477 52 MW 6 33
0 43 × 0 015
which allows a 17.5% penetration level of MMGs. On the other hand, providing
0.9% of the base system inertia by ESSs results in [11]
0 5J VI ω2 − VAhESS VAhESS
M ESS = 0 009 × M 2 = 0 009 × 6 5 =
VArated 900
VAhESS = 52 65
6 34
It then follows that the number of ESS units, which could provide such an
energy, is
52 65 = N × V ESS × AhESS = N × 12 × 480 N = 9140 6 35

This means that 9140 ESS units, rated 12 V and 480 Ah, are required to maximize
the MG penetration level. Following the same procedure as that of (6.25) for the
MG-penetrated system including 9140 ESS units gives
0
Δf2 [Hz]
–0.1
–0.2 Without Controller

With Controller
–0.3
0 2 4 6 8 10 12 14 16 18 20
Time [s]
Figure 6.8 ESS effect on frequency response.
1 015
Δ f 12 = Δ f COG 6 36
1 + 0 01Δ f COI 2
Comparing (6.25) with (6.36) suggests that the inertia-based control scheme can
effectively reduce the frequency nadir. Figure 6.8 exhibits the appropriateness of
the adopted control strategy, assuming that, in the online control system center,
the frequency event is detected after receiving five samples with increasing or
decreasing rate [17]; in our simulations, the ESSs are triggered 0.1 second follow-
ing the inception of the fault.
2) NYNE system:
NYNE test system of Figure 6.9 is used to further illustrate the efficiency of the
inertia-based control scheme. Five different contingency scenarios, including
the trip of major generating units and load shedding at selected buses, are con-
sidered for the studies. Results in Table 6.1, comparing the capability of the host
grid to accommodate MGs with and without the advanced control scheme,
show the effectiveness of the inertia-based control scheme.
Table 6.1 reveals that the penetration level of MMGs into the system is limited by
the frequency nadir at bus 60, in Area 1, for contingency 3. This contingency
causes the COG and COI to experience the frequency nadirs of 49.78 and
49.89 Hz, respectively. Using (6.19) and (6.22), the frequency dynamics of bus
60 can be expressed in terms of the COG frequency behavior by the sensitivity
relation
1 346
f nadir = f 6 37
60
1 + 0 363 f COI 1 COG
then
1 346
f nadir
60 = 49 78 = 49 19 6 38
1 + 0 363 49 89 50
14 Area 3 Area 1
66 8
41 40 48 47 60
1
25 26 29
53
28
10 1 2 61
27
62 9
30
31 3
11
63 18 17
16 24
46 38 32
33 15 22
Area 5 34
9
4
21
19 58
49 14
35 6
42 36 5 20
8 6 12 56 23
67 64
12
57 4
15 45
51 5
59
50 44 37 11 13 7
7 54
2
65 10
52 43 13
39
55
68
3
Area 4 16 Area 2
Figure 6.9 Single-line diagram of the 68-bus system showing coherent areas and their
interconnections.
Table 6.1 Comparison of maximum penetration level for NYNE test system with and
without the controller.
MP without MP with the proposed ESS/

Scenario Fault controller (%) controller (%) load (%)
1 G12 14.23 22.37 2.83

2 L14 11.34 21.78 3.01
3 G16 10.87 28.11 2.98
4 L37 10.56 26.06 3.54
5 L42 12.01 29.73 3.16
MP: maximum penetration level.
Inspection of (6.38) shows that when the COG frequency nadir reaches 49.78 Hz,
the frequency at bus 60 drops to 49.19 Hz, thus limiting the penetration of MG gen-
erations into the system. Installing 11 MW of ESSs at bus 60 increases the first
swing frequency amplitude to 49.42 Hz, and the penetration level of the system
increases by 17%. Table 6.2 reports the required ESSs to increase the penetration
level of MMGs. It should be noted that as generators G14–G16 represent the aggre-
gated behavior of areas 3–5 connected to the New York power system, the
Table 6.2 Required ESS capacity to enable high penetration of MG.
Scenario Area 1 (%) Area 2 (%) ESS/load (%)
1 5.71 3.89 2.99

2 3.89 2.03 1.82
3 5.91 4.21 3.16
4 6.33 1.70 2.39
5 3.51 4.53 2.60
specification of the ESSs capacity for such equivalent areas is far from reality.
While the reported results in columns 2 and 3 of Table 6.2 are expressed in per-
centage based on each area generation capacity, the last column is expressed in
percentage based on the total system load.
It should be emphasized that using this approach, each area in the system is
characterized by two equations of the form (6.30) and (6.31). Given a system with
n areas, solving the set of 2 × n equations allows to determine the number of ESSs
and MMGs ratings in the overall system.
Figure 6.10 compares the penetration levels as well as the frequency nadirs for
the system with and without the inertia-based controller. For completeness, the
30
25
Penetration [%]
20
15
10
5
0
1 2 3 4 5 6 7 8
49.4
49.3
49.2
fmin[Hz]
49.1
49
48.9
48.8
1 2 3 4 5 6 7 8
Scenario
Without Controller With Controller [19] With inertia-based controller
Figure 6.10 Penetration level as a function of frequency nadir.

1
With ESS
Without ESS
f [pu]
0.96
0.92
0 1 2 3 4 5 6 7 8 9 10
Time [s]
Figure 6.11 ESS effect on frequency response.
effectiveness of the inertia-based control scheme is compared with those strategies

introduced in [19] where synchronous condensers in addition to the inertial
response from wind farms are employed to enhance frequency dynamics.
Furthermore, Figure 6.11 shows the time-domain simulation results for
scenario 3. It indicates that the RoCoF and the frequency nadir are much less
pronounced in the case with the ESS and that without the ESS. It should be noted
that frequency nadir occurs before the fourth second of simulation, as the delivery
time of primary frequency in the base system is 3.5 seconds, according to Ref. [20].
6.3 Small Signal Stability Enhancement
6.3.1 Key Concept

In this section, an advanced control scheme based on the flexible inertia concept
is proposed to simultaneously improve voltage regulation and small-signal
stability performance. For this purpose, phase difference versus voltage deviation
(Δδ − ΔV) graphs are introduced as a powerful tool for the realization of a flexible
inertia-based control scheme. In conventional systems, automatic voltage regula-
tor (AVR) and power system stabilizer (PSS) are assumed to act on the voltage and
the rotor angle to improve voltage regulation/transient stability and small-signal
stability performance. The AVRs and PSSs produce torques in phase with the rotor
angle and the speed variations, respectively. However, both the AVR and PSS
employ the field voltage to produce these torques, which are not in phase. More
generally stated, a single control signal is employed to satisfy two conflict control
actions [21]. Accordingly, it seems that using the Δδ − ΔV graphs helps to make a
trade-off between the voltage regulation and small-signal stability performance in
the control synthesis process [22, 23]. In what follows, the possibility of dispatch-
ing the ESSs in some generator buses in addition to disconnecting the MMGs in
some others based on the (Δδ − ΔV ) indices is investigated.
6.3.2 Control Scheme Design

Following a system perturbation, a graph for each generator bus is plotted in the
plane of phase difference versus voltage deviation (Δδ − ΔV ). In this plane, the
variables are defined as:
Δδi t = δi t − δ0i 6 39
and
ΔV i t = V i t − V 0i 6 40
where δ0i and V0i are the initial values of rotor angle and terminal voltage related to
ith generator, respectively. To establish the effective criteria for triggering of the
inertia-based control scheme, parameters normalization is suggested. For this pur-
pose, consider the maximum variable variations as:
Δδmax t = max Δδi t 6 41
and
ΔV max t = max ΔV i t 6 42
Accordingly, one could rewrite (6.39) and (6.40) in the normalized form of
Δδi t
Δδi t = 6 43
Δδmax t
and
ΔV i t
ΔV i t = 6 44
ΔV max t
Economic reasons in addition to environmental constraints cause transmission
lines to operate close to their limits. Therefore, the desired operation of a power
system after being subjected to a disturbance could be theoretically achieved when
the system returns to the planned operating point, characterized by voltage profile,
nominal frequency, and transmitted power. Considering the power flow between
buses i and j, following a system perturbation, in the form of one could define the
following conditions to guarantee the desired performance:
V 0i + ΔV i t V 0j + ΔV j t
Pij = sin δ0i + Δδi t − δ0j + Δδ j t
X ij
6 45
1) Voltage deviations ΔVi for each generator converge to zero.

2) The difference between the angle deviations, i.e. Δδi − Δδj, converges to zero.
Ideally, all the connected generators satisfy the aforementioned conditions, and
hence, they are characterized by (1, 1) in the normalized (Δδ − ΔV) plane. How-
ever, in a real power grid, the system returns to a new operating point different
from the initial one, and hence, the voltage and the difference between the rotor
angle deviations of the generators differ from each other. Therefore, control
actions should be taken to force the terminal voltage deviations and the difference
between the rotor angle deviations into the desired value, i.e. zero. In other words,
for the stable, secure, and reliable system operation, all the generators’ operating
points must be located in the minimum distance from (1, 0) in the normalized
(Δδ − ΔV ) plane. This could be mathematically represented by the minimization
problem of
min Δδi + ΔV i − 1, 0 6 46
which could be rewritten as
2 2
min Δδi + ΔV −1 6 47
in the Cartesian system.

The minimum of (6.47) can be calculated by setting the derivative of the equa-
tion with respect to ΔV or Δδ (both give the same results) to zero, namely
2Δδi + 2ΔV i dd ΔV
Δδi
i
=0 6 48
2 2
2 Δδi + ΔV i
Hence,
d ΔV i
2Δδi + 2ΔV i =0 6 49
d Δδi
from which it follows that
Δδi = ΔV i = 0 707 6 50
As stated, the inertia-based control scheme tries to conduct all the committed
generators to the prefault condition, i.e. (1,1) in the normalized plane. Therefore,
one could define the secure operation region in the plane as:
Δδi ≥ 0 707 6 51
and
ΔV i ≥ 0 707 6 52
Moreover, parameters normalization imposes

Δδi ≤ 1 6 53
and
ΔV i ≤ 1 6 54
The sign of the acceleration torque in (6.2) specifies the sign of phase and voltage
variations in (6.51) through (6.54). According to (6.51)–(6.54), the stable region of
the system in the normalized plane of phase difference versus voltage deviation
can be represented in Figure 6.12.
For the generators located in the CDEFC area, the conventional controllers are
well-tuned on which the synchronizing and damping torques are positive. How-
ever, for the generators located out of the desired region, the conventional control-
ler parameters, including AVR and PSS, should be retuned to return the generators
to the CDEFC area. However, the main aim is to perform the same job as AVR–PSS
via the realization of flexible inertia.
Assume that some generating units, for a given time of interest, are located in the
desired region CDEFC while others are out of the region. The aim is to conduct all
the generating units to the desired region CDEFC. The logic of the inertia-based
control scheme is:
While for the generating units located in the desired region CDEFC, dis-
patching of the ESSs and consequentially increasing the effective grid
Figure 6.12 Stable region of power Δδ

system
1.41
G 0.707 A ΔV
45
E 0.707
F
D C B
inertia can decrease the parameter variations for the machines out of the
desired region, disconnecting the MMGs decreases effective grid inertia
which in turn increases the parameter variations.
Accordingly, the inertia-based control scheme changes Δδi, ΔVi and the maxi-
mum variable variations in (6.43) and (6.44) and hence, causes the generators to
move toward the desired region. Of note that the control scheme would be realized
in several steps in such a way that in each step, and for generators in the region of
interest, only 10% of the installed ESSs would be dispatched. On the other hand, only
10% of the synchronized MMGs are disconnected in each step of the algorithm.

The effectiveness of the inertia-based control scheme is investigated on the NYNE
test system of Figure 6.7. In each generator bus of the system, the ESSs are placed
according to the optimal results of Figure 6.13. Details on how to calculate
Figure 6.13 are given in Chapter 4 using Figure 4.21.
Figure 6.14 demonstrates the capability of the inertia-based control scheme to
enhance the system stability in compliance with the voltage regulation as well
as small-signal stability performance. Figure 6.14a shows the position of the
0.35
0.3
0.25
Virtual Inertia
0.2
0.15
0.1
0.05
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Generator Number

(a)
–0.2
–0.4
ΔδN
–0.6
–0.8
–1
–1 –0.9 –0.8 –0.7 –0.6 –0.5
ΔVN [deg]
(b)
–0.975
–0.98
–0.985
ΔδN
–0.99
–0.995
–1
–1 –0.9 –0.8 –0.7
ΔVN [deg]
Figure 6.14 Trace of generating units in the Δδ – ΔV plane in response to the visualization
of flexible inertia after the outage of generator 2.
committed generators in the normalized Δδ – ΔV plane just 1 second after the out-
age of generator 2. It can be seen that except for six generators, others are out of the
desired region. As discussed earlier, dispatching the ESSs in the buses associated
with the generators in the desired region and disconnecting the MMGs in other
buses improve the system performance. Figure 6.14b suggests that the flexible-
inertia-based control scheme can return all the generating units to the desired
region.
Further, the effectiveness of the flexible inertia concept to improve the voltage
regulation as well as the small-signal stability performance is assessed considering
(a)
–0.2
–0.4
ΔδN
–0.6
–0.8
–1
–1 –0.9 –0.8 –0.7 –0.6 –0.5
ΔVN [deg]
(b)
–0.99
–0.992
–0.994
ΔδN
–0.996
–0.998
–1
–1 –0.95 –0.9 –0.85 –0.8 –0.75

ΔVN [deg]
Figure 6.15 Trace of generating units in the Δδ – ΔV plane in response to visualization of

flexible inertia after outage of generator 16.
the outage of generator 16. Figure 6.15 reveals that all the generating units are
located in the desired region in about 20 seconds after the inception of the fault.
The time-domain simulation result for the applied fault further demonstrates the
high capability of the proposed control scheme (Figure 6.16).
Regarding the time horizon of voltage instability (>2 seconds) and transient
instability (about 2 seconds), the sampling time interval to dispatch the ESSs
and disconnect the MMGs can be set to 2 seconds. However, the proposed control
strategy employs the first sample at a smaller time (less than 2 seconds) to improve
its reliability.
References 207
60
40
δ [deg]
20
0
0 5 10 15 20 25 30 35 40
1.1
V [p.u.]
0.9
0 5 10 15 20 25 30 35 40
Time [s]
Figure 6.16 Effects of flexible inertia on voltage regulation and small-signal stability.
6.4 Summary
This chapter deals with the designing of advanced control schemes to improve
power system stability. The discussed advanced control schemes rely on the inertia
manipulation in the system to mitigate undesired frequency, voltage, and rotor
angle dynamics. The developed control schemes depend on the stochastic equiv-
alent model of the power system to enable high penetration levels of MGs. The
approaches combine the adaptive ESS dispatch strategy with the MG controlled
islanding scheme to provide a stability support for the host grid.
References
1. Hatziargyriou, N. et al. (2020). Stability definitions and characterization of dynamic
behavior in systems with high penetration of power electronic interfaced
technologies. IEEE, Tech. Rep.
2. Hatziargyriou, N., Asano, H., Iravani, R., and Marnay, C. (2007). Microgrids. IEEE
Power and Energy Magazine 5 (4): 78–94.
on power system stability and operation. arXiv preprint arXiv:1312.6435.
4. Hatziargyriou, N. et al. (2017). Contribution to bulk system control and stability by
distributed energy resources connected at distribution network, IEEE, PES-TR22.
5. Lasseter, R. et al. (2002). Integration of Distributed Energy Resources. The CERTS
Microgrid Concept. Berkeley, CA: Lawrence Berkeley National Labaratory (LBNL).

33 (1): 1026–1035.
7. Meriam, J.L. and Kraige, L.G. (2012). Engineering Mechanics: Dynamics. Wiley.
8. Bevrani, H. (2009). Robust Power System Frequency Control. Springer.
9. Golpîra, H. (2019). Bulk power system frequency stability assessment in presence of
microgrids. Electric Power Systems Research 174: 105863.
10. Golpira, H. and Bevrani, H. (2011). Application of GA optimization for automatic
generation control design in an interconnected power system. Energy Conversion
and Management 52 (5): 2247–2255.
12. Golpira, H., Atarodi, A., Amini, S. et al. (2020). Optimal energy storage system-
Transactions on Power Systems.
13. Teng, F. and Strbac, G. (2017). Full stochastic scheduling for low-carbon electricity
systems. IEEE Transactions on Automation Science and Engineering 14 (2):
461–470.
14. Teng, F., Trovato, V., and Strbac, G. (2016). Stochastic scheduling with inertia-
dependent fast frequency response requirements. IEEE Transactions on Power
Systems 31 (2): 1557–1566.
15. Lasseter, R.H. (2006). Control and design of microgrid components. PSERC
Publication 06–03.
16. Chen, C.-L., Wang, Y., Lai, J.-S. et al. (2009). Design of parallel inverters for smooth
mode transfer microgrid applications. IEEE Transactions on Power Electronics
25 (1): 6–15.
17. Standard BAL-003-1 (2013). Frequency Response and Frequency Bias
Setting. NERC.
20. GE (2010). Frequency Response: Basics. USA: GE Energy.
21. Golpîra, H., Bevrani, H., and Naghshbandy, A.H. (2010). A survey on coordinated
design of automatic voltage regulator and power system stabilizer. International
Review of Automatic Control 3 (2): 172–182.
22. Golpira, H., Bevrani, H., and Naghshbandy, A.H. (2012). An approach for
coordinated automatic voltage regulator–power system stabiliser design in large-
References 209
scale interconnected power systems considering wind power penetration. IET

Generation, Transmission & Distribution 6 (1): 39–49.
23. Batmani, Y. and Golpîra, H. (2019). Automatic voltage regulator design using a
modified adaptive optimal approach. International Journal of Electrical Power &
Energy Systems 104: 349–357.
211
Small-Signal and Transient Stability Assessment Using

Data-Driven Approaches
In the past decades, a variety of approaches have been introduced to assess small
and large systems performance, from linear, model-based techniques to data-
driven approaches. This includes efforts to extract spatial shapes, determine modal
properties, and assess the energy exchange between interconnected systems,
among other issues. A common approach is to simultaneously analyze the data
sequences using techniques such as the multichannel Prony method or Koopman
mode decomposition. Despite the importance of these methods, more general,
global multiscale methods that rely on the joint analysis of large data sets are
needed to cope with the ever-increasing utilization of distributed generators
(DGs) and the quantity and complexity of the recorded system behavior.
As the number and distribution of inverter-based generation resources grow, it
becomes increasingly challenging to extract modal properties from heterogeneous
and complex, measured responses following system perturbations. Knowledge of
the fundamental characteristics of global oscillation modes provides valuable
information about the stability of oscillatory phenomena and may help to identify
machines and wind and solar photovoltaic (PV) farms and their control systems
involved in the exchange of oscillating energy and the design or modification of
controllers.
This chapter identifies basic issues important in understanding the oscillatory
performance of wind and solar PV penetrated power systems. A fundamental
study of the characterization of the dynamic behavior of power systems with
increased renewable generation is presented. The study is motivated by the need
to further clarify the participation of wind and PV farms in interarea oscillations.
A second goal is to assess the performance of modern data-driven modal tools and
analysis techniques.
The use of modal characterization techniques is illustrated on two test systems: a
6-machine, 10-bus test system, and a 5449-bus, 635-generator test system. By using
a simple example, the effect of renewable generation on system dynamic

212 7 Small-Signal and Transient Stability Assessment Using Data-Driven Approaches
performance is first investigated. The challenges that need to be confronted for

such approaches to be more widely applied by the power system community
are then discussed in a large-scale power system. Analytical criteria to describe
the energy relationships in the observed oscillations are derived, and a physical
interpretation of the system modes is suggested. Future research topics are sum-
marized and described.
7.1 Background and Motivation
The rapid growth of inverter-based distributed renewable generation and the

increased availability of measured data have made the application of data-driven
techniques to the analysis of system dynamic behavior imperative. Conceptually,
the study of large systems characterized by a complex mix of generating resources,
often involving multiple contingency scenarios and more sophisticated (and non-
linear) control strategies, poses multiple challenges. Critical aspects include the
identification of wind and solar PV locations that participate significantly in crit-
ical oscillatory behavior [1–3], the extraction of modal information (spatial and
temporal modes) of the system [4–6], and the analysis of control interactions [7].
Power system oscillatory behavior can be analyzed in terms of modes, expressed
as exponentially modulated sinusoids, exhibited in signals measured on the system
[8]. These signals are driven by the behavior of a large, nonlinear, time-variant sys-
tem and may contain responses involving disparate timescales. Typically, system
oscillatory behavior is commonly analyzed using small-signal (linear) analysis and
time-domain simulations. The modal properties (i.e. modal damping, frequency,
and shape) of these modes can be affected by various factors including flow pat-
terns, control actions, market conditions, and the increased penetration of renew-
able energy sources (RESs) [9]. Moreover, measured data from distributed energy
resources may exhibit nonlinear trends, transient ramps, or random fluctuations
that may obscure or make difficult physical interpretation. In addition, the use of
converter interfaces precludes a direct interpretation of the effect of renewable
generation on modal behavior [10].
More advanced algorithms are required to extract the spatiotemporal structures
concealed in the data that represent the relevant system dynamics as well as to
assess the impact of operational and planning strategies on system performance.
Using small-signal analysis to assess the impact of DGs/RESs on system dynamic
performance can be a difficult and challenging task. First, power system models
are becoming more complex and nonlinear. Further, current renewable genera-
tion models are designed to capture, essentially, linear, balanced, fundamental fre-
quency behavior and may not be suited for analyzing control interactions or other
phenomena. Second, the analysis and interpretation of oscillatory behavior in

large power systems with significant penetration of distributed generation may
be difficult or provide partial information on various aspects of system
performance.
This motivates the need to develop simplified models that significantly reduce
the number of degrees of freedom by neglecting less relevant details of the involved
physics to remain computationally tractable. Algorithms for multiple unknown
signal extraction from selected system measurements are well suited to analyze
complex dynamic behavior and can provide complementary information to the
small-signal analysis. Using data-driven approaches, dynamic trends and phase
relationships between key system signals from measured data can be obtained effi-
ciently [11, 12]. In addition, energy relationships of interest in the study and clas-
sification of global modes can be readily determined.
Modern data-driven methods rely increasingly on advanced algorithms and
analysis methods that help to overcome some of the size-related limitations of
measured data sets. This chapter investigates the application of modal extraction
techniques to examine oscillatory behavior in the presence of high wind and solar
PV penetration.
Techniques to extract modal parameters directly from observational data are
critically examined and compared with well-established modal analysis techni-
ques. Factors that affect the performance of the method are also discussed, includ-
ing the effects of nonlinear trends, data quality, and sampling design. Connections
with other modal identification methods are also investigated.
7.2 Modal Characterization Using Data-Driven

Approaches
Measured data sets are often relatively large and may exhibit nonlinearities and
other artifacts. In this section, several data-driven approaches are examined to var-
ious aspects associated with the integration of RESs. First, a conventional mass–
spring system is used to introduce the nature of oscillatory phenomena. The
notation is also introduced.
7.2.1 Modal Decomposition

A great deal of physical insight about the nature of oscillatory phenomena can be
gained by viewing a multimachine system as a lossless mass–spring system. With-
out loss of generality, consider an n-degree-of-freedom lossless mechanical system,
Mq + Kq = 0, where q ᑬn × 1 is the coordinate vector, and the mass and spring
constant matrices M ᑬn × n and K ᑬn × n are symmetric matrices [13, 14], and

damping effects are neglected.
The equation of motion can also be written as a pair of first-order equations
as [14]
M 0 q 0 −M q 0
+ =
0 M p K 0 p 0
−1/2
Use of the transformation q = M x in the second-order model yields the sec-
ond-order system x + Ax = 0, with A = M−1/2KM−1/2. A similar interpretation
can be given for the first-order set of linear differential equations.
The free natural response is given by
n
x t = A j sin ω j t + θ j v j 71
j=1
aj t
where n is the number of modes, the v j s are the natural modes, and the terms
aj(t) = Aj sin(ωjt + θj) represent the time modulation of the natural modes, in
which Aj, ωj, and θj are the modal amplitude, frequency, and phase, associated
with the jth mode respectively.
Let now a t = a1 t a2 t … an t T ᑬn × 1 be the vector containing the
time evolution of the modes, aj(t) = Aj sin(ωjt + θj), j = 1, …, n, and
V = v 1 v 2 … vn ᑬnxn be the modal matrix of modal coordinates. It follows
that:
x t = Va t 72
By evaluating (7.2) for a given time interval, t1, t2, …, tN, the system response can
be written in terms of the modal matrix and the matrix of modal coordinates as
x1 t 1 x 1 t2 x1 t N
x2 t 1 x 2 t2 x2 t N
X t = x t1 x t2 … x tN = = VA t 73
xn t 1 xn t 2 xn t N
where, x k t j = x k t j xk t j … xk t j , j = 1, …, N, and
a1 t 1 a1 t 2 a1 t N
a2 t 1 a2 t 2 a2 t N
A=
an t 1 an t 2 an t N
is the matrix of time-dependent coefficients.
Alternatively, one can rewrite (7.3) as

n
X t = a j t v j = Va t 74
j=1
where the aj(t) are time-dependent coefficients.

Equation (7.4) shows that the time evolution of the system states can be
expressed as a linear combination of modal components and that this representa-
tion can be used to analyze mode–state relationships. In the literature, several
measures of energy and coupling between modes and states have been proposed
in both analytical models [15, 16] and data-driven techniques [12].
To introduce the more general ideas that follow, let the matrix A have a set of
n distinct eigenvalues (λ1, λ2, …, λn) and a corresponding set of right and left eigen-
vectors be U = (u1, u2, …, un) and V = (v1, v2, …, vn). Defined in this way, the modal
components, U, form a set of orthogonal basis vectors [16]. In Refs. [15, 16], it is
shown that participation factors (PFs), Pki = ukivki can be considered as the partic-
ipation of the kth state when only the ith mode is excited. Alternatively, the PFs
can be shown to be modal energies of the unforced system x = Ax or be associated
with more general mode–state relationships as discussed in later sections of this
chapter.
In what follows, the applicability of time-domain methods to characterize sys-
tem dynamic behavior in power systems with large penetration of distributed
renewable generation is investigated. These methods provide a basis for the modal
decomposition of an ensemble of measured data obtained from measurements or
experiments that can be related to aforementioned linear analyses.
Among the various multichannel modal extraction methods, attention is
focused on three methods: Koopman mode decomposition analysis, singular value
decomposition (SVD)-based Prony analysis, and dynamic mode decomposi-
tion (DMD).
7.2.2 Multisignal Prony Analysis

The Prony methods have been described in detail in several papers [17–20]. For
completeness, a concise review of these methods is presented as follows.
7.2.2.1 Standard Prony Analysis

Following the nomenclature of Hauer [17], consider a linear time-invariant system
represented by the model x = Ax, y = Cx, with x t o = x o . Let now the noisy
measured signal y t , t = 0, …, N − 1 be expressed as a superposition of the true
signal, y(t), and noise ε(t), which for the sake of convenience is assumed to be
white, as [21]
y t = y t + ε t , t = 0, …, N − 1 75
where N is the number of samples.
Assume further that the ringdown waveform y(t) can be written as a superposi-
tion of Q exponentials as [18, 22]
Q
yt = Ai eλi t cos 2π f i t + ϕi 76
i=1
Defining
Bi = Ai e jϕi
zi = eλi Δt = e σi + jωi Δt
Equation (7.6) can be rewritten in the discrete domain (t = tk = kΔt), in the form
n
yk = Bi zki 77
i=1
where y(k) = y(tk), k = 0, 1, …, N − 1, the Bi C are the signal residues, λi λj

for i j, zi = eλi tk , Δt is the sampling space, and the parameters {Bi, zi}, i = 1,
…, n are unknown complex parameters to be determined; n denotes the order
of the model.
Neglecting noise in (7.5), and assuming that N = 2n, it is possible to fit y t to a
model of the form (7.7) [23]. Formally, for 0 ≤ k ≤ N − 1, it follows from this
expression that
z10 z20 zn0 B1 y0

z11 z21 zn1 B2 y1
= , k = 0, …,N −1 78
z1N − 1 z2N − 1 znN − 1 Bn y N −1

B Cn×1 y ᑬn × 1
Z Cn× n
or, in compact form, ZB = y, where Z is a Vandermonde matrix [24].

Associated with matrix Z is the characteristic equation (the Prony polynomial),
n n n−1
pz = z − zi = am z n − m = z n − am z m , ao = 1 79
i=1 m=0 m=0
and the companion or Frobenius matrix [24]

0 0 0 ao
1 0 0 a1
Cn = 0 1 0 a2 Cn × n
0 0 1 an − 1
where the ajs are unknown coefficients to be determined and det(zIn − Cn(a)) =
p(z), in which In is the n × n identity matrix. It is noted that the eigenvalues of
Cn coincide with the zeros of the Prony polynomial in (7.9).
It can be verified after some algebra that (7.9) satisfies a forward linear predictor
model (LPM) of the form [25, 26]
n
am y k − m = 0
m=0
Using the known data values, y(k), this equation can be rewritten in matrix
form as
yn y n−1 y1 y n+1
y n+1 yn y2 a1 y n+2
a2 y n+3
y n+2 y n+1 y3 = − , n+1≤k≤N
an
y N −1 y N −2 y N −n a
yN
Y y
7 10a
or
y n−1 y n−2 y0 yn
y n−1 a1
yn y1 y n+1
a2 y n+2
y n+1 yn y2 = − , n ≤ k ≤ N −1
an
y N −1 y N −2 y N −n−1 a
y N −1
Y y
7 10b
in which it is assumed that ao = 1.
Equations (7.10a) and (7.10b) can be rewritten in the form Ya = y, where Y is a
matrix of dimension (N − n) × n. This system of equations can be solved directly
for the ajs if N = 2n, as a = Y−1y [23].
As noted in Ref. [21], the Prony method is essentially a procedure to determine

the zis in (7.7) without resorting to nonlinear optimization. This can be achieved
using the following three-step approach:
1) Determine vector a by solving (7.10),

2) Compute the roots of the polynomial p(z) = 0 by solving (7.9),
3) Solve (7.8) for the Bi as B = Z−1y.
This is often referred to as the standard or classical Prony method.

Having determined the complex amplitudes and phases, the modal parameters
in (7.6) can be computed as
zi
σ i = log
T
ωi = tg − 1 Im zi Re zi T
and
Ai = zi
φi = tg − 1 Im zi Re zi
in which T = Δt is the sampling interval.
The reader is referred to Refs. [17, 19, 20, 27] for further details about the prac-
tical implementation of this method and its potential applications.
7.2.2.2 Modified Least-Squares Algorithm

In practical applications, there are more data points than parameters N > 2n. This
results in an overdetermined equation of the form (7.10b)
y n−1 y n−2 y0 yn
yn y n−1 y1 a1 y n+1
a2 y n+2
y n+1 yn y2 = − , N > 2n
an
y N −1 y N −3 y N −n−1 a
y N −1
Y b
7 11
which has a noniterative, least-squares solution of the form
−1
a = Y †b = − Y T Y Y Tb
where Y† denotes the Moore–Penrose pseudoinverse of Y.

Similarly, one can write
z01 z02 z0n B1 y0

z11 z12 z12 B2 y1
= 7 12
zN1 − 1 zN2 − 1 zN2 − 1 Bn y N −1
The aforementioned results extend readily to an arbitrary number of measured

data as discussed further.
7.2.2.3 Multichannel Prony Analysis

Extension of the aforementioned procedures to the multichannel case has been
considered by Trudnowski and coworkers [19]. Let ym, m = 1, …, M denote a
set of measurements (channels) with N data points each, k = 0, 1, …, N − 1. In this
case, the LPM in (7.10b) can be rewritten as
y1 n − 1 y1 n − 2 y1 0 y1 n
y1 n y1 n − 1 y1 1
y1 n + 1
Block 1
y1 N − 1 y1 N − 2 y1 N − n − 1 a1
a2 y1 N − 1
= −
yM n − 1 yM n − 2 yM 0 yM n
an
yM n + 1
yM n yM n − 1 yM 1 am
Block M
yM N − 1 yM N − 2 … yM N − n − 1 yM N − 1
ym
Ym
7 13
where the a j s are the modified signals’ amplitudes, and each signal is defined by a
vector, y j = y j 0 y j 1 y j N − 1 T , j = 1, …, M.
Given this result, the following three-step procedure is used to compute the
Prony parameters
1) Build M data blocks and construct (7.13) as a column-wise concatenation of

data blocks as
Y1
Y2
Ym =
YM
with
yj n−1 yj n−2 yj 0
yj n yj n−1 yj 1
Yj= , j = 1, …, M
yj N −1 yj N −2 yj N −n−1
2) Solve (7.13) for the unknown vector am ,

3) Obtain the roots of the Prony polynomial (7.9),
4) Calculate the residues Bi, from the linear system,
y1 0 y2 0 yM 0
y1 1 y2 1 yM 1
y1 2 y2 2 yM 2 = V vand B
y1 N − 1 y2 N − 1 yM N − 1
with
1 1 1
z1 z2 zn
V vand = z11 z12 z1n
zN1 − 1 zN2 − 1 zNn − 1

1 2 M
B1 B1 B1
1 2 M
B2 B2 B2
1 M
B= B3 2
B3 B3
1 2 M
Bn Bn Bn
Robust, more efficient interpretations of this model are discussed in subsequent

sections.
7.2.2.4 Hankel-SVD Methods

These methods take advantage of the Hankel structure of the data matrix and can
be used to efficiently exploit the application of advanced analytical methods. Other
related techniques based on SVD analysis of a Hankel matrix are the MUSIC
(multiple signal classification) methods [28] described in Chapter 6 and the
approximate Prony method.
Given a data matrix ym, m = 1, …, M, one can define a Hankel matrix, Y H, of the
form [29, 30],
y2 y3 y4 yL
y3 y4 y5 y L+1
Y H = Y ij = y4 y5 y6 y L+2
yL y L+1 y L+2 yN
ᑬL × L , a Hankel matrix
where L = N − M + 1, M = N/2, and the elements Y ij depend only on the sum of the
indices, i.e. Y i + 1,j + 1 = y i + j , i + j = 0, 1, 2, …, L.
An algorithm is now suggested for extracting Prony modes from multichannel
data based on Hankel SVD analysis. As outlined in previous sections, the
minimum norm solution to the linear prediction equation Y a = b is given by
†
a = Y b. A more efficient solution for noisy measurements can be obtained from
SVD analysis of matrix Y .
Suppose the SVD of the Hankel matrix Y H is given by
Y H = UΛV T 7 14
where Λ is a diagonal matrix with positive, real values in the upper left and zeroes
elsewhere, and U,V are the left and right singular vector matrices, respectively.
†
In Refs. [31, 32], it is shown that the pseudo inverse Y H of Y H is related to the
SVD of Y H by the relation
†
YH = V Σ† U T 7 15
in which Σ†is obtained from Σ by replacing each positive diagonal entry by its
reciprocal. Related approaches are introduced in Refs. [22, 32], where connections
are made with the Matrix Pencil method and the eigensystem realization algorithm.
From the SVD of the Hankel matrix in Eq. (7.15), it is then possible to construct a
least-squares estimate of the Hankel matrix for the ideal, noiseless signal. Once the
model (7.15) has been computed and the LPM is solved, the roots, z, of the char-
acteristic polynomial, are obtained; the associated Ritz eigenvalues [12], and their
modal amplitudes are computed from
λk = conj − log zk Δt
znk = eλk Δt 7 16
j−1
Z jk = znk
An overview of the developed Kumaresan–Tufts (KT) procedure used in the sec-

tions that follow to extract modal parameters is given in Table 7.1. Some comments
on the details of this algorithm are given as follows.
The extension to the multichannel case is immediate; in this case, m row-
concatenated Hankel matrices Y H = Y H 1 Y H 2 Y H M are needed for effi-
cient computation of the model. It should be observed that the method analyzes
all data blocks or channels simultaneously; as a result, the correlation between
them can be fully exploited, thereby improving the signal-to-noise ratio (SNR)
and robustness.
As the number of measurements grows, as in the case of distributed generation,
and measurements are increasingly nonlinear and sparse, it becomes increasingly
challenging to extract modal properties from large data sets. The following subsec-
tions explore the application of other dynamic mode representations.
7.2.3 Koopman and Dynamic Mode Decomposition Representations

An alternative method to the modal analysis of nonlinear complex systems is based
on the notion of the Koopman operator [33–36]. A concise review of this method is
given in this section.
Table 7.1 Hankel-based Prony analysis algorithm.
Given a record of data sequences ym, m = 1, …, N:
1) Arrange the data into a Hankel matrix Y H mi + 1,j + 1 = y i + j , i + j = 0, 1, 2, …, L.

2) Compute the SVD of the Hankel matrix
UΣV = svd Y H ; Y H = V ΣT U
and arrange the singular values in order of decreasing magnitude σ 1 ≥ σ 2 ≥ … ≥ σ n,
3) Solve the LPM of the system model and determine a least-squares estimate of the
measurement data,
4) Calculate the complex amplitudes zi = eλi t , Z s, r = zi s − 1
5) Determine the residues of the signals ym.
7.2.3.1 The Koopman Operator

Consider a discrete-time system evolving on an M-dimensional manifold
xk + 1 = f xk , x ᑬM 7 17
k = 0, 1, 2, …, N − 1; where k is an integer index.
Let now g z M − ᑬ be any scalar-valued function (a measurement of the state
or observable). The Koopman operator, U, is a linear operator that maps g into a
new function:
Ug x = g z
The key idea behind Koopman analysis is to study the system dynamics in equa-
tion (7.17), from measured data using the eigenspectrum of U. Assume to this end
that φj and λj denote the eigenfunctions and eigenvalues (Koopman modes) of the
Koopman operator, respectively, given by
Uφ j x = λ j φ j x , j = 1, 2, … 7 18
where for a sufficiently long N, the Koopman eigenfunctions form an orthonormal

expansion basis.
In practical applications, one is interested in functions g x =
g1 x g2 x … gp x M p, with p < N. If each of the components of
g lies within the span of the eigenfunctions, the time evolution of the functions
can be expanded as
∞
g x = φj x vj 7 19
j=1
From (7.18) and (7.19) one has that

∞ ∞
xk + 1 = g xk = U k φ j xo v j = λkj φ j x o v j 7 20
j=1 j=1
where use has been made of (7.18). Physically, (7.20) indicates that the observable
g(xk) is decomposed into vector coefficients, vj, called Koopman modes whose tem-
poral behavior is given by the associated eigenvalues λj.
As discussed further, the phase of the eigenvalues determines its frequency,
while its modulus determines the growth rate. The magnitude φj(xo)vj is used
as a measure of the relative participation of a mode to the modal decomposition.
Refer to [12, 34] for the detailed implementation of the method. Applications to
other power system dynamic phenomena are discussed in [37].
7.2.4 Dynamic Mode Decomposition

Recently, the DMD method and its variants have been developed to approximate a
few functions φj, using two sets of time-ordered sequences of data snapshots.
To introduce this method, assume that xj(tk) denotes an element of observation,

where xj is the jth grid or measurement point (a sensor), and tk, k = 1, 2, …, N
is the time at which the observations are made.
Define the data matrix X as
x 1 t1 x 1 t2 x 1 tN
x2 t1 x 2 t2 x 2 tN
X = X N1 = x 1 x2 xN = ᑬm × N
x m t1 x 2 t2 x m tN
x1 xN
7 21
T
where x j = x 1 t j x2 t j … xm t j , j = 1, …, N, and the superscript, N,
denotes end time; m represents the number of sensors or measurement locations.
The method assumes that the data sequences or snapshots, xj, in (7.21) are gen-
erated by a discrete-time linear dynamical system whose evolution is governed by
the linear mapping. Formally, these methods postprocess the sequence of snap-
shots (7.21) and relate two consecutive data fields through a linear mapping of
the form
x k + 1 = Ax k + ηk , k = 0, 1, …, N − 1 7 22
where A is an unknown operator matrix of dimension m×m for a time step Δ t, and
matrix A is real, asymmetrical, and high-dimensional; ηk is some noise process.
This equation defines a Krylov sequence [38]
x 1 = Ax o
x 2 = Ax 1 = A2 x o
x 3 = Ax 2 = A3 x o
7 23
x N = Ax N − 1 = AN x o
or, in a compact form
X N2 ᑬm × N = x 2 x3 xN = A x1 x2 x N − 1 = AX N1 − 1
−1
XN
2 XN
1
7 24
with
X N2 ᑬm × N = x 2 x3 xN
X N1 − 1 ᑬm × N = x 1 x2 xN − 1
It immediately follows that the linear operator A can be obtained as follows
†
A = X N2 X N1 − 1 7 25
The process of computing A directly from (7.25) may not be feasible, especially
when the size of the data set increases or when the model is ill-conditioned. Several
formulations to estimate the eigenvalues of matrix A have been developed and
tested. For completeness, a brief review of these methods is presented further.
Connections to other approaches are reviewed.
7.2.4.1 SVD-Based Methods

The DMD algorithm used in this section is adapted from [12, 38, 39], which is
briefly summarized further.
Following the derivation in [38], consider the data sequence (7.23). The
Krylov method relies on the fact that the vectors of the sequence
X N1 − 1 = x 1 x 2 x N − 1 sooner or later become linearly dependent. When
this condition is satisfied, one has that,
x N = c1 x 1 + c2 x 2 + + cN − 1 x N − 1 7 26
where the cj s are the unknown expansion coefficients.

Equation (7.26) can be rewritten in compact form as
x N ≈ c1 x 1 + c2 x 2 + + cN − 1 x N − 1 = X N1 − 1 c 7 27
where c = c1 c2 … cN − 1 T is a vector of unknown coefficients.

Now, recalling that X N2 = x 2 x 3 x N and making use of (7.27)
results in
X N2 = x 2 x3 xN = x2 x3 X N1 − 1 c 7 28
and hence
X N2 = AX N1 − 1 = X N1 − 1 S 7 29
where S is the Frobenius or companion matrix

0 0 0 co
1 0 0 c1
S= 0 1 0 c2 ᑬN − 1 × N − 1 7 30
0 0 1 cN − 1
As discussed earlier, the eigenvalues of S are a subset of the eigenvalues of

matrix A and are obtained from the characteristic polynomial
pS λ = λN − cN − 1 λN − 1 − cN − 2 λN − 2 − …c1 = 0
In practical application, linear dependence in (7.27) occurs gradually. As a
result, a residual is obtained from (7.29) of the form [38]
r = X N2 − X N1 − 1 S
AX N1 − 1 − X N1 − 1 S = reTp
where
0 0 ρ1
reTp =
0 0 ρn
The problem is now how to find the vector c, which determines the residues, r,
and therefore matrix S. From the aforementioned results, the eigenvalues of A
approximate those of S when r 2 0.
Following the earlier discussion, it can be seen that matrix S can be
determined by
argmin X N2 − X N1 − 1 S 2
7 31
S
It is straightforward to show that a solution to the aforementioned optimization

†
problem is given by S = X N1 − 1 X N2 . In practical applications, however, an alter-
nate algorithm to compute matrix S and its associated eigenvalues and eigenvec-
tors, based on SVD of the companion matrix, is favored when the matrix is rank
deficient (M > N) [40].
Let to this end the SVD of matrix X N1 − 1 be expressed as
X N1 − 1 = UΣW T 7 32
where U is an orthogonal matrix of right singular values, Σ is the diagonal matrix

of singular values, and W is the matrix of left singular values. Noting now that
X N2 = AX N1 − 1 , and substituting into (7.32), yields
X N2 = AUΣm W T 7 33
−1
XN
1
Inserting (7.29) in this expression, one has that
X N2 = X N1 − 1 S = UΣW T S = AUΣm W T 7 34
Multiplying (7.34) on the left by UTresults in
ΣW T S = U T A UΣm W T
from which it follows that
S = U T AU 7 35
Then, solving (7.33) for A and inserting in (7.33) yields
S = U T AU = U T X N2 W T Σm− 1 7 36
where Λ = diag λ1 λ2 λp is the diagonal matrix of eigenvalues, and Φ =

T
ϕ1 ϕ2 ϕr is the matrix of eigenvectors, with ϕ j = ϕ1j ϕ2j … ϕmj
Suppose now that S is diagonalizable with an eigenvalue decomposition
S = Y ΛY − 1 7 37
where Λ = diag λ1 λ2 … λm Cm × m is a diagonal matrix consisting of empir-

ical Ritz eigenvalues, λj, and Y = y1 y2 … ym C m × m is the matrix of right
eigenvectors, respectively.
Use of (7.36) into (7.37), yields
S = Y ΛY − 1 = U T X N2 W T Σm− 1 7 38
and hence,
X N2 = USΣm W m = U Y ΛY − 1 Σm WTm 7 39
or
X N2 = USΣm W m = UY ΛY − 1 Σm W m = ΦΛΓm t 7 40
spatial Temporal
sructure structure
where Γm(t) = Y−1ΣmWm, and Φ = UY. A physical interpretation of this model is

provided in Barocio et al. [12].
The computation of the modal decomposition can be summarized in the follow-
ing five steps:
1) Given a data set X = x 1 x 2 xN ᑬm × N , build matrices X N1 and

X N1 − 1 ,
2) Decompose the data matrix X using SVD,
3) Compute
A = U ∗X T W Σ − 1,
4) Compute the eigenvalues and eigenvectors of A from
A − Λ W = 0,
5) Calculate the DMD modes as
Ψ = X T V Σ − 1W
7.2.4.2 The Companion Matrix Approach

As discussed earlier, the eigenvectors of S form a basis for the span of A. Now,
referring back to Eq. (7.23), one can write
x 1 = Ax o
x 2 = Ax 1
x 3 = Ax 2 = A2 x 1
x 4 = Ax 3 = A3 x 1
x N = Ax N − 1 = AN − 1 x 1
or x i = Ai − 1 x 1 , i = 2, 3, …, N
From linear system theory, each data vector can be expanded in the form (refer
to Eq. (7.1)). Following [40] let
m
x1 t = a jϕ j 7 41
j=1
and
m
xi t = Ai − 1 a j ϕ j , i = 1, 2, …
j=1
It is left to the reader to prove that the time evolution of the ith state can be
expressed as
m
x i t, x = λij− 1 a j ϕ j , i = 1, 2, … 7 42
j=1
where the aj s are the time-dependent amplitudes, the φjs are the eigenvectors of
S, and λj = σ j + iωj is the complex frequency of the associated mode.
Equation (7.42) can be conveniently rewritten in the vector-matrix form
X t = ΦV vand c 7 43
where Φ = ϕ1 ϕ2 ϕm , is the matrix of spatial structures (mode

shapes), and
1 λ1i λki λki
1 λki λki λki
V vand = 7 44
1 λki λki λki

is the Vandermonde matrix.
Several approaches to compute the vector of modal amplitudes, c, have been
suggested in the literature. References [41, 42] are cited to this end.
Table 7.2 gives a simplified procedure for this approach.
Table 7.2 Krylov-based DMD algorithm.
Dynamic mode decomposition algorithm

Given a data set X = x 1 x2 xN ᑬm × N :
1) Build matrices X N2 and X N1 − 1 ,

2) Compute the SVD of X oN − 1 , U, Σ, V = svd X No − 1 ,
3) Compute the rank, r, of Σ and compute a truncated approximation of U, S, V,
4) Compute matrix X as
X = U T X N1 V Σ − 1 ,
5) Calculate the eigenvalues and eigenvectors, Λ, U of X, with Λ = diag λ1 λ2 … λr ,

6) Compute the Vandermode matrix using (7.44),
7) Determine the Ritz eigenvalues and their associated frequency and damping factors
σ and ω.
8) Compute mode shapes and mode amplitudes in (7.40),
9) If desired, reconstruct the original measurements using (7.43) as
X = ΦcV vand ,
7.2.4.3 Energy Criteria

In pursuing the analogy with linear modal analysis in Section 7.2.1, assume that
the time evolution of a measured signal, x(t), can be represented by a decomposi-
tion of the form (7.21). In discrete time
r−1
t Δt
x t = λj a jϕ j 7 45
j=0
where k = t/Δt.
In what follows, analytical criteria to describe the energy relationships in the
observed oscillations are derived and a physical interpretation of the system modes
is suggested. For simplicity and clarity of exposition, assume that the eigenvectors
φj are normalized to unity, so that
ϕj
ϕj =
ϕj 2
It follows from basic considerations that the total averaged energy, E(T), over a
time interval [0, T] can be expressed as
T T 2
r−1
1 2 1 t Δt
E T = x t dt = λ j c jϕ j dt
T T j=0
0 0
or
T 2
r−1
1 t Δt
T
t Δt
E T = λj c jϕ j λj c jϕ j dt 7 46
T j=0
0
Performing the product on the right-hand side of (7.46) yields

T
r −1
1 T t Δt
E T = ϕ j ϕ jλ j c j dt
T j=0
0
Hence, the averaged energy for the jth mode is

T
1 T t Δt
Ej T = ϕ j ϕ jλ j c j dt 7 47
T
0
Noting, finally, that elogx = x, and integrating (7.47) with respect to t gives
1 2
T
2σ t 2 e2σ j T − 1
Ej = ϕj λ j j dt = ϕ j 7 48
T 0 2σ j T
which is used in Ref. [43] as a metric for ranking system modes.

As shown in (7.48), modal energy is proportional to the square of the mode shape
amplitude, the selected time interval, T, and the damping decay, σ j. This makes
this criterion useful to identify and isolate specific dynamic behavior associated
with a given mode and timescale.
The next sections describe the application of the aforementioned techniques in
the context of the oscillatory system response in more detail.
7.3 Studies of a Small-Scale Power System Model
As a first motivating example, a six-machine test system is used in this section to

investigate the effect of inverter-based generation on system dynamic performance
[44]. The test system consists of 10 buses, five generators, and a synchronous com-
pensator; the total system load is 430 MW.
Figure 7.1 depicts a single-line line diagram of the test system showing genera-
tion and transmission resources. In this representation, Bus 1 is an infinite bus and
is modeled by a classical representation with a large inertia value. The rest of the
generators (buses 1, 3, 4, 6, 11, and 12) are represented by a fourth-order model
equipped with an IEEE type 1 excitation system. In this model, the machine at
bus 4 is a synchronous condenser (SC).
7.3.1 System Data and Operating Scenarios

Relevant load flow and transmission parameters for this system are given in
Tables 7.3–7.5. Full details of the model are provided in Refs. [44, 45], which
are briefly summarized here.
11 10 4 2 1
∼
∼
3
∼ ∼
12
SC
7 5 6
∼
∼
Tap location
Figure 7.1 Ten-bus, six-machine test system.

Table 7.3 Load-flow data.
Load Generation
Bus Voltage (pu) Angle ( ) MW MVAr MW MVAr
1 1.0000 0.0000 0.00 0.00 52.69 −2.04
2 1.0388 −8.2649 100.00 25.00 0.00 0.00
3 1.0500 −1.9973 0.00 0.00 100.00 16.28
4 1.0500 13.7510 100.00 25.00 0.00 26.85
6 1.0683 1.4698 0.00 0.00 0.00 0.00
7 1.0504 −7.7070 30.00 15.00 0.00 0.00
10 1.0360 −13.8039 200.00 50.00 0.00 0.00
12 1.0500 −7.5449 0.00 0.00 100.00 18.65
Table 7.4 Transmission line data.
Number of circuits Line charging in MVAR

From bus To bus in parallel R in pu X in pu at 1 pu voltage
1 2 2 0.08 0.6 2.5
2 4 2 0.08 0.4 10.0
4 7 2 0.08 0.4 10.0
4 10 4 0.08 0.4 10.0
5 7 2 0.08 0.4 10.0
Table 7.5 Transformer data.
From bus To bus R in pu X in pua Transformer tap ratiob
2 3 0.02 0.12 1.02
5 6 0.01 0.06 1.02
10 11 0.02 0.12 1.02
10 12 0.02 0.12 1.02
a Impedances are in pu on a 100 MVA base.

b The tapped side corresponds to the bus listed on the left.
Vref ,...
Plant level
fref ,... control Vter Pgen Qgen
Pgen Qref Iqcmd Iq

Torque Electrical Generator
Prefo Ipcdm Ip
control control converter
Pref
ω
Pgen Drive
ωref
train Pord
Pm
Pitch angle θ Aerodynamic

control model
Figure 7.2 Generic wind farm/solar photovoltaic model.
In the studies described further, wind turbines of the wind farms (WFs) are mod-
eled to represent typical (Type 3) doubly fed induction generator (DFIG) config-
urations, similar to the GE WFs’ models [46, 47]. Three main control loops are
considered in the analysis (see Figure 7.2 adapted from [46, 47]):
1) The electric control system model

2) The mechanical control system model
3) The generator model (DFIG or voltage source converter model)
Chapter 8 discusses the particular implementations of these models.

Four different operating scenarios were used in the simulations described as
follows:
Case A. A base case with an infinite bus representation at bus 1,

Case B. A modified case where the infinite bus at bus 1 is replaced by a wind
generator,
Case C. A modified case in which the infinite bus at bus 6 is replaced by an equiv-
alent wind generator and bus 1 is an infinie bus,
Case D. A modified case in which the infinite bus 1 and the thermal generator at
bus 6 are replaced by equivalent wind turbine generators.
Exploratory studies assessing the impact of renewable generation on system

dynamic performance are described in the following subsections.
7.3.2 Exploratory Small-Signal Analysis

The base-case system (Case A) has five modes of oscillation at approximately 0.5,
0.89, 1.47, 1.72, and 2.24 Hz associated with synchronizing power flows among the
five generators and the synchronous compensator [48]. Table 7.6 lists the eigenva-
lues, frequencies, damping ratios of the electromechanical modes of interest. Col-
umn 5 in this table summarizes the oscillation patterns.
For reference, the speed-based mode shapes are given in Figure 7.3 for each
mode in Table 7.6, computed used a small-signal analysis software. Values are nor-
malized with respect to the largest entry. Note that, because of the infinite bus rep-
resentation, the magnitude of the speed (angle) entry for all modes is zero. Readers
can find a more detailed discussion on the nature of these modes in Ref. [49].
Eigenvalue analysis of Case A identifies two low-damped low-frequency electro-
mechanical modes at about 0.50 and 0.89 Hz and three higher frequency modes at
about 1.47, 1.75, and 2.34 Hz associated with more localized behavior. From a
physical perspective, the right eigenvector entries have large magnitudes at
machines having strong participation in the oscillations.
Tables 7.7–7.9 give eigenvalue results for Cases B–D.
Table 7.6 Electromechanical modes of the system.
Frequency Damping
Mode Eigenvalue (Hz) (%) Oscillation pattern
1 −0.0055 ± 0.501 0.180 Buses 6, 12, 11, 4, 3 swing coherently

j3.1503 against bus 1
2 −0.1105 ± 0.894 1.97 Buses 12, 3, 11, 4 vs. bus 6
j5.6202
3 −0.2461 ± 1.473 2.66 Buses 3, 4, 11 vs. 6 and 12
j9.2564
4 −0.0918 ± 1.752 0.83 Buses 4, 11 vs. 6, 12, 3
j11.0113
5 −0.4009 ± 2.326 2.74 Bus 11 vs. 6, 12, 4, 3
j14.6170
Case A with an infinite bus at bus 1.

0.501 Hz 0.894 Hz
1 1
Magnitude
Magnitude
0.5
0.5
0
0
6 12 11 4 3 1 12 11 4 3 1 6
Bus Bus
1.433 Hz 1.752 Hz
1 1
Magnitude
0.5 0.5
0
0
–0.5
3 4 11 1 6 12 4 11 1 6 12 3
Bus Bus
2.326 Hz
1
Magnitude
0.5
0
11 1 6 3 12 4
Bus
Figure 7.3 Normalized speed-based mode shapes of the electromechanical eigenvalues in

Table 7.6 (base operating case).
Table 7.7 Electromechanical modes of the system (Case B with a wind farm at bus 1).
Mode Frequency (Hz) Damping (%) Oscillation pattern
1 0.827 2.12 Buses 12, 3, 11, 4 vs. bus 6

2 1.335 3.81 Buses 3, 4, 11 vs. 6 and 12
3 1.752 0.63 Buses 4, 11 vs. 6, 12, 3
4 2.324 2.74 Bus 11 vs. 6, 3, 12, 4
Table 7.8 Electromechanical modes of the system (Case C with a WF farm at bus 6).
1 0.728 3.49 Buses 12,11,4,3 vs. 1

2 1.463 2.81 Buses 3, 4, 11 vs. 12
3 1.738 1.23 Buses 4, 11 vs. 12, 3
4 2.341 2.75 Bus 11 vs. 3, 12, 4
In all analyzed cases, damping improves with high wind generation, but as dis-
cussed earlier the effect of wind generation on system dynamic performance
depends on several interacting factors.
In the following subsections, data-driven approaches that supplement informa-
tion on conventional small-signal performance are investigated.
Table 7.9 Electromechanical modes of the system (Case D with wind farms at buses
1 and 6).
1 1.318 4.41 Buses 3, 4, 11 vs. 12

2 1.737 1.14 Buses 4, 11 vs. 12, 3
3 2.322 2.87 Bus 11 vs. 3, 12, 4
7.3.3 Large System Performance

Detailed system simulations have been carried out to characterize transient behav-
ior as well as to examine the accuracy of the base case model. Emphasis is placed
on the analysis of Cases B through D, which are assumed to represent distributed
wind generation.
7.3.3.1 Cases B–C

Selected system simulations for cases B and C are shown in Figures 7.4–7.7. These
solutions are thought to be representative of other operating scenarios described
further. Based on these results, time-domain identification methods described in
Section 7.2.2 are used to extract the modal characteristics of the system.
(a)
64.3
Wind farm at bus 1
Frequency (Hz)
64.2
64.1
64
63.9
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
(b)
60.1 Bus 1
Frequency [Hz]
Bus 2
60.05 Bus 3
Bus 4
60 Bus 5
59.95
59.9
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
Figure 7.4 Speed deviations response following a three-phase fault at bus 7 (Case B);
(a) bus 1, and (b) buses 1–5.
50
40
30
Angle [deg]
20
Bus 1
10 Bus 2
Bus 3
Bus 4
0 Bus 5
–10
–20
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
Figure 7.5 System rotor angle deviations following a three-phase fault at bus 7 (Case C).
60.04
60.03
60.02
60.01
Frequency [Hz]
60
59.99
Bus 1
59.98 Bus 2
Bus 3
Bus 4
59.97 Bus 5
Bus 6
59.96 Bus 7
Bus 8
59.95 Bus 9
Bus 10
59.94
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
Figure 7.6 System bus frequency deviations following a three-phase fault at bus 7 (Case B).
Due to the small number of measurements, multisignal Prony analysis was used
to identify the modes observed in the transient stability simulations [19].
A discussion of other modal extraction techniques is postponed until Section 7.4.
Table 7.10 shows Prony results for the generator speed deviations and genera-
tors’ active output power (Case B) following a three-phase fault at bus 7. The good
agreement between modal and data-driven approaches in Tables 7.5 and 7.8
1.08
Bus 1
Bus 2
Bus 3
1.07 Bus 4
Bus voltage magnitude [pu]
Bus 5
Bus 6
1.06 Bus 7
Bus 8
Bus 9
Bus 10
1.05
1.04
1.03
1.02
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
Figure 7.7 System bus voltage deviations following a three-phase fault at bus 7 (Case B).
Table 7.10 Multisignal Prony analysis for speed deviations and generator active power at
buses 1, 2, 4, 6, 11, and 12 (Case B).
Generator speeds Generator active power
Mode Frequency (Hz) Damping (%) Frequency (Hz) Damping (%)
1 0.863 1.798 0.862 1.740

2 1.381 2.311 1.381 2.271
3 1.741 0.632 1.741 0.628
4 1.879 9.191 1.879 9.191
ensures the correctness of the model. Results are found to be consistent, although
generator active power is seen to provide a better characterization of the damping
characteristics of mode 1 in Table 7.6, especially for the damping of mode 1.
A comparison of analytical results in Tables 7.6 with data-based results allows us
to validate the small-signal analysis results.
7.3.3.2 Case D
In the discussion on the effect of renewable generation on the system dynamic per-
formance, and to highlight the impact of inertia reduction, a two-WF case is now
considered. In this analysis, generators at buses 1 and 6 were replaced by
(a)
64.5
Bus 1
Frequency [Hz]
Bus 6
64
63.5
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
(b)
60.1
Frequency [Hz]
60
Bus 11
Bus 12
59.9 Bus 3
Bus 4
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
Figure 7.8 Speed deviations of system machines and wind farms following a three-phase
fault at bus 7 (Case D); (a) buses 1 and 6, and (b) buses 3, 4, 11, and 12.
equivalent type 3 WFs. For simplicity and clarity of exposition, the same set of
parameters for the control systems was utilized.
Figure 7.8 shows the system response to a three-phase fault at bus 7. Of interest
in this plot are correlated motions of WFs at buses 1 and 6, which suggest the
importance of identifying dynamic patterns and utilizing dynamic equivalencing
techniques.
Further insight into the participation of WFs into system oscillatory behavior
can be obtained from the analysis of PFs for the 1.318 Hz mode in Table 7.8. Sim-
ilar results are obtained for other modes and are not discussed here.
As shown in Figure 7.9, the participation of the WFs to the modes of interest can
be easily identified. The phase of PFs, however, does not have a direct interpreta-
tion, and other measures may be needed to supplement information on energy
exchange, as well as the interaction paths.
Based on this analysis, data-driven approaches are used to supplement informa-
tion on the effect of renewable generation on system dynamic behavior. To dem-
onstrate the application of the aforementioned formulation, consider the time
evolution of speed deviations following a three-phase stub fault at bus 7 in
Figure 7.8. The time window selected for analysis is 20 seconds.
Figure 7.10 shows the time evolution of the extracted modes using Koopman
analysis. Similar plots and results are obtained using DMD and other approaches.
From Figure 7.10, system dynamic behavior is essentially dominated by two well-
damped modes: a 1.21 Hz mode, and a 1.35 Hz mode.
Gen 3
Angle, speed
1
0.9 Gen 12
Angle, speed
0.8
0.7
Magnitude
0.6
0.5
0.4
WF 1
Gen 4
0.3 WF 1 Converter
Angle, speed
Converter Gen 3
Gen 12 WF 6
0.2 Ψd
Ψq Converter
0.1
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
State
Figure 7.9 Participation factors for mode 1 in Table 7.5.
Table 7.11 shows the captured dominant modes. Multichannel Prony analysis
indicates two modes with frequencies near 1.17 and 1.35 Hz that capture over
99% of the observed system response. Mode selection (column 3 in Table 7.11)
was based on the transient energy ratio described in Refs. [43, 49].
0.15
Mode 2
0.1 Mode 1
0.05
Real part of mode
–0.05
–0.1
–0.15
4 6 8 10 12 14 16 18 20
Time [sec]
d
Figure 7.10 Dominant DMD modes from the decomposition j = 0a j t ψ Tj following a
three-phase fault at bus 7.
Table 7.11 DMD analysis of generator bus frequency deviations.
Mode Frequency (Hz) Damping (%) Participation (%)
1 1.717 0.504 66.88

2 1.355 2.401 22.76
Three-phase fault at bus 7.
The importance of the modal approximation in Eq. (7.20) or Eq. (7.28) is now
obvious. Simulation results suggest that the time evolution of the system states
can be approximated by the reduced-order model X t = ψ 1 t aT1 t + a2 t ψ T2 ,
where, as discussed earlier, the temporal coefficients ajs provide the temporal
(amplitude) evolution of the modes, and the vector ψ give the approximate mode
shape.
7.3.4 Mode Shape Identification

In the context of modal characterization, a key issue in the integration of wind and
solar PV farms is the extraction of spatial relationships. Mode shapes play increas-
ingly central roles in assessing energy transfer paths, assessing system damage,
and complementing time-domain simulations [5]. DMD analysis is employed here
to extract mode shapes from time-domain simulations.
Information on the nature of these modes can be gleaned from the mode shapes.
To illustrate this issue, the shape of mode 1 for Case D in Table 7.12 was deter-
mined using conventional Prony analysis. In this case, the phase for mode 1
was determined by applying Prony analysis to each speed deviation in Figure 7.7.
In this analysis, each speed deviation was decomposed in the form
Table 7.12 Voltage-based mode shape of the interarea mode at 1.35 Hz extracted using
Prony analysis for bus voltage magnitudes in Table 7.7 (Case D).
Machine/wind farm Frequency, fi (Hz) Phase, φi ( ) Amplitude, Ai
WF at bus 1 1.355 −358.18 0.0211

Gen at bus 11 1.356 −176.55 0.0076
Gen at bus 12 1.356 −346.75 0.0233
Gen at bus 3 1.356 −171.77 0.0521
Gen at bus 4 1.356 −163.38 0.0205
WF at bus 6 1.355 −355.15 0.0086
Q
Δω j t = Ai eλi t cos 2π f i t + ϕi , j = 1, …, 5 7 49
i=1
Table 7.12 shows Prony results (PRS) for the speed-deviation signals.
Several conclusions can be drawn for this analysis.
• The WFs at buses 1 and 12 and the WF at bus 6 swing nearly in phase and in
opposition to generators at buses 11, 3, and 4.
• Bus 3 has the largest participation in the mode, followed by the generators at
buses 12, 4, and the WF at bus 1.
Depending on the chosen algorithm, spatial structures can be real or complex.

Insight into the applicability of these techniques can be gleaned from Figure 7.11
that shows the extracted voltage-based mode shape from simultaneous DMD anal-
ysis of the voltage measurements at generation buses 1, 3, 4, 6, 11, and 12.
Comparison of Prony results in Table 7.10 with DMD analysis in Figure 7.11
shows the correctness of the formulations.
7.3.5 Temporal Clustering

Clustering and dimensionality reduction play a critical role in the study of inter-
system oscillations. Paramount to this approach is the ability to compress the com-
plicated variability of the original data set into the smallest possible number of
basic modes and their associated centroids.
0.2
0.1
Real part of mode
–0.1
–0.2
–0.3
–0.4
1 11 12 3 4 6
Bus
Figure 7.11 Speed-based mode shape for mode 1 in Table 7.3 computed using DMD
analysis (Case D).
In Chapter 5 [49], an exploratory clustering technique based on fuzzy c-means

that identifies groups of trajectories with similar compositions was implemented
to divide the data into clusters and study power system coherent behavior.
Fuzzy clustering can be posed as the solution of the energy function:
K n
2
f U, v = ki x i − mk
um 7 50
k=1i=1
where mk are the cluster centers, the um

ki represent associated membership likeli-
hoods (the degree of membership), 0 ≤ uki ≤ 1, of the trajectory xi being associated
with the cluster center mk.
The optimization problem is subject to the constraints
K
uki = 1, i = 1, …, n
k=1
7 51
uki ≥ 0 , k = 1, …, K, i = 1, …, n
Using this approach (see Table 7.13), the system is divided into four areas and a
centroid is associated with each area. In practice [49], it has been noted that, in the
case of systems with high renewable generation, the centroids determined using
the aforementioned techniques are a natural selection to monitor aggregated
behavior.
In turn, the analysis of speed deviations in Figure 7.8 results in four clusters
(m = 4 in [7.50]):
Cluster 1, which includes buses 11 and 4,

Cluster 2, which includes buses 12 and 3,
Cluster 3, which consists of bus 1, and
Cluster 4, which consists of bus 6.
Figure 7.12 shows the time evolution of the centroids for clusters 1 and 2 along
with the buses included in the clusters.
Table 7.13 Clusters determined from c-means

analysis of generator bus frequency data.
Cluster Buses
1 12
2 4, 6, 11
3 3
4 1
Three-phase fault at bus 7.

(a)
60.1
Frequency [Hz]
60
59.9 Bus 11
Centroid 1
Bus 4
59.8
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
(b)
60.1
Frequency [Hz]
60.05
60
59.95 Bus 12
Centroid 2
59.9 Bus 3
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
Figure 7.12 Speed signals following a three-phase fault at bus 7; (a) Cluster 1, and
(b) Cluster 2.
Note that, by applying this technique, the center of angle (frequency) can be
determined without simplifying assumptions.
7.4 Large-Scale System Study
In the system studied thus far, attention has been confined to basic active and reac-
tive control strategies. The purpose now is to introduce some specific control alter-
natives in the study of a practical test system.
7.4.1 Case Study Description

The test system is based on a seven-area, 5449-bus representation of a large inter-
connected system. It includes several large-scale wind and solar PV farms, and
SVCs. A single-line diagram of this system showing main buses and transmission
lines and the location of selected WFs and solar PV plants is shown in Figure 7.13.
The system model consists of 635 fully represented generators, 5449 buses, 5292
transmission lines, and the detailed representation of 26 WFs and 3 equivalent
solar PV farms; the total interconnected system load is 35 000 MW.
61 Area 2
85 Area 3
107
69 1
89
2
62 11
Area 1 194
143
124 137 153 Wind farms
21–26
192 184
3 31 115
105
47
26 Area 7
Area 4 Area 6
8 117 83 15
42 18 109
146 161
Area 5 92
47 95
78 7
144
Wind farms
11–20
Wind farms
1–10
Figure 7.13 Schematic illustration of the seven-area test system showing major buses and
transmission lines.
7.4.2 Renewable Generator Modeling

In the system model, all WFs are represented using detailed user models similar to
those described in Section 7.2. Table 7.14 summarizes the main characteristics of
these WFs. The block diagrams of the control systems, which are adapted from [46,
47], are given in Figure 7.14, while Figure 7.15 shows the two-mass model used in
the simulation studies.
For this study, the following assumptions are made:
•• All WFs are modeled by type 3 (DFIG) generic wind turbine models.
Wind plant reactive power control is enabled.
• Voltage-dependent load characteristics are represented in detail.
7.4.3 Effect of Inverter-Based DGs on Oscillatory Stability

Small-signal stability analyses were conducted to examine the impact of wind gen-
eration on modal stability. Table 7.15 synthesizes the main characteristics of the
slowest interarea modes in the system.
Table 7.14 Characteristics of DFIGs used in the study.
Operating Voltage Active power

Wind farm Type voltage (kV) control control
1–2, 5–10, 14–15, Type 3 0.69 Yes Yes

17–19 DFIG
3, 11–13 Type 3 12 Yes Yes

DFIG
4, 16 Type 3 13.2 Yes Yes

DFIG
7.4.4 Large System Performance

Early transient stability studies have indicated that distributed nonsynchronous
generation in Areas 3, 4, and 6 have a noticeable impact on small-signal stability.
To excite the slowest electromechanical modes and investigate the impact of
renewable generation on the oscillatory stability, several contingency scenarios
were developed. These include large-loss-of-generation events, load shedding,
double line outages, and short circuits at major buses.
Selected contingency scenarios are listed in Table 7.16.
The following measurements are selected for study:
•• 935 bus frequency signals,

200 bus voltage magnitudes,
•• 138 generator speed deviations, and

26 WF speed deviations.
Figures 7.16 and 7.17 show plots of frequency deviations of WFs 1–26 for three
critical contingency scenarios in Table 7.16. In all cases, a nonlinear trend is
observed, especially in the first few seconds of the simulations.
Studies conducted to extract dominant system behavior are discussed further.
Table 7.17 shows the dominant system modes extracted using DMD applied to
the generator speed deviation signals in Figure 7.16.
7.4.5 Model Validation

Extensive studies have been conducted to validate the system model using multi-
signal modal extraction techniques. From the original simultaneous results, two
data sets were initially selected to compare system results
(a)
Vrfq
KIV Qmax
Vc s +
1 1 1
1 + sTR Σ Σ
fN 1 + sTFV
KPV +
1 + sTV Qmin
Pfarep
tan
Pele –1 1 Qele Vterm vltflg

1 XIqmax
π
1 + sTP Qmax
–
– >0 Eqcmd
KQI KQV
Σ Σ
Qref 0 + s + s
0
Qmin
varflg XIqmin
(b)
MWbase Pelec wg
MWrat
Vterm
Pmax
Speed RPmax
+ IPmax
1 KIP 1 ÷
π
1 + sTPWR – Σ Σ
KPP +
s sTPP
RPmin
Pmin
MWcap
MWrat
Ipcmd
Figure 7.14 Active and reactive control characteristics: (a) reactive and (b) active control
characteristics. Adapted from [46, 47].
1) Data set 1 including 40-speed deviations at major selected generators and

26 WFs. This scenario allows for Prony analysis of the data,
2) Data set 2 set including 983 bus frequency signals,
3) Data set 3 with 164-speed deviation signals.
Figure 7.18 shows the speed deviations for data set 1.

Tmech – + ωo
Pmech Δωt
1 + ωt
÷ ∑ 2Mts ∑
+
Turbine
–
+
Δωtg Δωtg 1
Dshaft ∑ Kshaft
s
– + ωo
Pe + Δωg
1 ωg
÷ ∑ ∑
2Mgs
Te – + Generator
Figure 7.15 Two-mass wind mechanical model.
Table 7.18 compares the amplitude and frequency of the extracted modes using
Prony analysis and DMD analysis for the 66-speed deviation signals (data set 1)
with the eigensolution of the linear model x = Ax. Results between Prony analysis
and DMD analysis are found to be in good agreement although some discrepancies
are found.
Table 7.15 Slowest interarea modes of the system.
Frequency Damping
Mode Eigenvalue (Hz) (%) Swing pattern
1 −0.0330 ± j2.481 0.394 1.33 Areas 1–3 vs. Areas 6 and 7

2 −0.2415 ± j3.430 0.547 7.02 Area 1 vs. Areas 3, 6, and 7
3 −0.3549 ± j4.103 0.653 8.62 Areas 1 vs. Areas 2, 3, and 6
3 −0.3280 ± j4.376 0.696 7.47 Area 1 vs. Areas 2, 6, and 7
5 −0.4944 ± j5.1061 0.812 9.64 Area 5 vs. Area 6
Base case condition.

Table 7.16 Contingency scenarios selected for the study.
Contingency
scenario Description
CS01 Loss of the largest unit in the system (generators 1 and 2, Table 8.1)
CS02 Three-phase fault at bus 153 in Area 3
CS03 Three-phase fault at the POISa of WFs 1–19 in Area 6
CS04 Single-line outage, 8024–8052 in Area 6
CS05 The simultaneous outage of 400 kV lines 8024–8520 and 7415–7426

in Area 6
CS06 The outage of generator 1, unit 1 (750 MW) in Area 6
a Point of interconnection with the system.
(a)
44.6
Frequency [Hz]
44.4
44.2
44
43.8
0 5 10 15 20 25 30
Time [sec]
(b)
72.3
Frequency [Hz]
72.2
72.1
72
71.9
0 5 10 15 20 25 30
Time [sec]
Figure 7.16 Frequency deviations for contingency scenario CS04; (a) WFs in Area 6, and (b)
WFs in Areas 2 and 3.
Eigenvalues obtained using a small signal stability program exhibit good agree-
ment with the time-domain studies. Study experience shows that conventional
Prony analysis becomes prohibitive for high-dimensional data sets, and some
expertise is needed to determine both the sampling interval and the study time
window.
(a)
44.3
Frequency [Hz]
44.2
44.1
44
43.9
0 5 10 15 20 25 30
Time [sec]
(b)
72.2
Frequency [Hz]
72.1
72
71.9
5 10 15 20 25 30
Time [sec]
Figure 7.17 Frequency deviations for contingency scenario CS02: (a) WFs in Area 6 and (b)
WFs in Areas 2 and 3.
Table 7.17 Frequencies and damping extracted from generator

speed deviations using DMD for contingency scenario CS04.
Mode Frequency Damping
1 0.399 0.389
2 0.547 7.000
3 0.666 8.29
3 0.696 7.473
5 0.798 9.527
7.4.5.1 Reconstructed Flow Fields

To further verify the accuracy of the models, Koopman and DMD analyses were
used to extract the dominant modes of data set 2 above, where X f ᑬ2804 × 983 =
Δf 1 Δf 2 … Δf 983 . Application of Koopman mode decomposition (KMD) and
DMD in Table 7.19 to the data set identifies three dominant modes with frequen-
cies of nearly 0.395, 0.555, and 0.697 Hz. Figure 7.19a shows the original signals,
Δfj while Figure 7.19b shows the reconstructed signals Δf j using the Koopman
procedure in Section 7.2.3.
3
Generators in Areas Generators in Areas
2 2 and 3 5 and 6
1
Frequency [Hz]
–1
–2 WFs 20–26 WFs 1–19
–3
–4
–5
0 5 10 15 20 25 30
Time [sec]
Figure 7.18 System response to a three-phase fault at the POIS (contingency scenario
CS03) Signals detrended for clarity of presentation.
Table 7.18 Modal extraction analysis for simulation results in Figure 7.18.
Multisignal Prony analysis DMD analysis Small-signal analysis
Frequency Damping Frequency Damping Frequency Damping

Mode (Hz) (%) (Hz) (%) (Hz) (%)
1 0.395 1.287 0.395 1.291 0.395 1.331

2 0.659 8.587 0.657 8.601 0.653 8.620
3 0.548 7.182 0.552 7.128 0.545 7.000
Contingency scenario CS04. Time window 20–30 seconds.
Similar results are obtained using DMD analysis. The overall reconstruction
error, ε = 1.277× 10−3, is computed as
N
2
xk
k=1
ε xk , xk = N
7 52
2
xk
k=1
where x k , k = 1, …, N is the DMD approximation, and xk is the simulated data.

The second problem of key importance is that of simultaneous analysis of mea-
surements involving different physical units. To illustrate these ideas, assume that
several data sets (such as speed deviations, frequency, bus voltages, etc.) are
Table 7.19 Comparison of Koopman and DMD results for data set 2.
Koopman analysis DMD analysis
Mode Frequency (Hz) Damping (%) Frequency (Hz) Damping (%)
1 0.397 1.288 0.395 1.291

2 0.677 8.345 0.657 8.615
3 0.550 7.230 0.552 7.012
Time window 20–30 seconds.
(a)
0.08
0.06
Bus frequency deviations [p.u.]
0.04
0.02
–0.02
–0.04
–0.06
5 10 15 20 25 30
Time [s]
(b)
0.08
0.06
Bus frequency deviation [p.u.]
0.04
0.02
–0.02
–0.04
–0.06
5 10 15 20 25 30
Time [s]
Figure 7.19 Bus frequency signals following a three-phase fault at the POIS (contingency
scenario CS03): (a) original, simulated signals, and (b) reconstructed signals using DMD.
7.5 Analysis Results and Discussion 253
obtained from the same disturbance. They could represent measurements from
multisignal records or represent historic information.
Let each data set be of the form X set 1 ᑬN 1 × m1 , X set 2 ᑬN 2 × m2 , …, X set L
ᑬN L × mL . If each data matrix has the same length (N1 = N2 = … = NL), one can
express the concatenated matrix as
X set 1
X set 2
X=
X set L
This motivates the need for a joint analysis of multiple data sets much as in the
case of feature-based fusion of data. Three main issues are of interest here: (i) to
study the complementary nature of data types, (ii) to assess the effect of a given
data type of modal behavior and which combination of data modalities can be used
for specific analysis or control purpose, and (iii) to find common structures and
relationships between data types.
7.4.6 Identification of Mode Shapes Using DMD

As discussed earlier, identification of mode shapes in power systems with highly
distributed energy resources is a critical objective in assessing the applicability of
data-driven methods. Because of their inherent characteristics, mode decomposi-
tion methods (global multiscale methods) are well suited to extract mode shapes
and other global characteristics.
Figure 7.20 shows the speed-based mode shape extracted from the speed devia-
tions in Figure 7.16 (data set 1) using DMD analysis. To avoid cluttering, 40 gen-
erators and 26 WFs are selected for display. Figure 7.20 indicates that WFs in Area
6 swing (nearly) in phase with generators in Areas 6 and 7 and in opposition to
generators and WFs in the south systems. It is also apparent in these plots that
WFs 10, 17, and 19 in Area 6, and WFs 20 and 21 in Area 5 show the largest con-
tribution to the slowest swing mode.
7.5 Analysis Results and Discussion
Multichannel modal extraction methods are rapidly becoming a powerful and

ubiquitous tool for modal analysis of large sets of simulations or measurements.
Effective modeling and characterization of global oscillatory phenomena
0.15
0.1
Real part of mode
0.05
Wind generators
Synchronous generators
(Areas 1–3)
0
Synchronous generators Area 6
(Areas 6, 7)
–0.05
–0.1
0 10 20 30 40 50 60
Generator/wind farm
Figure 7.20 Speed-based mode shape of 0.395 Hz interarea mode 1.
involving large-scale penetration of RESs require many observational data often

involving disparate spatial or temporal time scales. Extensive numerical experi-
ence shows that the effectivity of these methods depends on several interacting
factors such as the presence of nonlinearities, the presence of noise, and the size
of the observation window, to mention a few factors.
In this chapter, several multichannel modal extraction techniques have been
benchmarked by comparison with multichannel Prony analysis and small-signal
studies. These tools include techniques for extracting modal properties, determin-
ing spatiotemporal patterns, analyzing the system dynamics, and evaluating the
system response. It is demonstrated that these methods can be efficiently applied
to analyze large, complex data from selected system measurements and shed light
on the impact of distributed generation on system oscillatory behavior. A crucial
part of these applications concerns understanding the role of distributed energy
resources in the observed oscillations. Methods for interpreting the structure
and distribution of spatiotemporal patterns are given and techniques to visualize
these patterns are pointed out.
Simulation experiences show that while WFs’ speed signals can be used to
extract spatial patterns, often other signals such as bus terminal frequencies offer
a useful alternative to modal characterization. Other studies reveal that trends
exhibited by WF signals may affect the interpretation of modal characteristics.
Further analysis is needed to assess the performance of these methods when
applied to measured, heterogeneous data.
References 255
References
1. Vittal, E., O’Malley, M., and Keane, A. (2012). Rotor angle stability with high
penetrations of wind generation. IEEE Transactions on Power Systems 27 (1):
353–362.
2. Vittal, V. and Ayyanar, R. (2013). Grid Integration and Dynamic Impact of Wind
Energy. New York, NY: Springer.
3. Gautam, D., Vittal, V., and Harbour, T. (2009). Impact of increased penetration of
DFIG based wind turbine generators on transient and Small-signal stability of
power systems. IEEE Transactions on Power Systems 24 (3): 1426–1434.
4. Power System Dynamic Performance Committee, Task Force on Identification of
Electromechanical Modes, Sanchez-Gasca, J.J. Chair. (2012). Identification of
electromechanical modes in power systems. IEEE/PES, Special Publication TP462
(June 2012).
5. Wilson, D., Bialek, J., and Gustavsson, N. (2008). Identifying sources of damping
issues in the icelandic power system. Proceedings of the 16th Power Systems
Computation Conference (PSCC’08).
6. Castellanos, R.B., Messina, A.R., Gonzalez, M.R., and Guizar, G.C. (2018).
Assessment of frequency performance by grid integration in a large-scale power
system. Wind Energy 21: 1359–1371.
7. CIGRE Working Group C.6.08. (2011). Grid integration of wind generation
(February 2011).
8. Browne, T.J., Vittal, V., Heydt, G.T., and Messina, A.R. (2008). A comparative
assessment of two techniques for modal identification from power system
measurements. IEEE Transactions on Power Systems 23 (3): 1408–1415.
9. Grebe, E., Kabouris, J., López Barba, S. et al. (2010). Low frequency oscillations in
the interconnected system of Continental Europe. Proceedings of IEEE PES
General Meeting.
10. Sanchez-Gasca, J.J., Miller, N.W., Price, W.W. (2004). A modal analysis of a two-
area system with significant wind power penetration. Proceedings of the IEEE PES
Power Systems Conference and Exposition.
11. Messina, A.R. and Vittal, V. (2007). Extraction of dynamic patterns from wide-area
measurements using empirical orthogonal functions. IEEE Transactions on Power
Systems 22 (2): 682–692.
12. Barocio, E., Pal, B.C., Thornhill, N.F., and Messina, A.R. (2015). A dynamic mode
decomposition framework for global power system oscillation analysis. IEEE
13. Feeny, B.F. and Kappagantu, R. (1998). On the physical interpretation of proper
orthogonal modes in vibrations. Journal of Sound and Vibration 211 (4): 607–616.
14. Goebel, C.J. and Epstein, S.T. (1980). Motion of damped oscillators: expansion in
normal modes. American Journal of Physics 48 (4): 289–291.
15. Hamdan, A.M.A. and Nayfeh, A.H. (1986). Coupling measures between modes and
state variables. International Journal of Control 43 (3): 1029–1041.
16. Kundur, P. (1994). Power System Stability and Control. New York, NY:
McGraw-Hill.
17. Hauer, J. (1991). Application of Prony analysis to the determination of modal
content and equivalent models for measured power system response. IEEE
18. Pierre, D.A., Trudnowski, D.J., and Hauer, J.F. (1990). Identifying reduced-order
models for large nonlinear systems with arbitrary initial conditions and multiple
outputs using prony signal analysis. Proceedings of the 1990 American Control
Conference.
19. Trudnowski, D.J., Johnson, J.M., and Hauer, J.F. (1999). Making Prony analysis
more accurate using multiple signals. IEEE Transactions on Power Systems 14 (1):
226–231.
20. Zhou, N., Pierre, J., and Trudnowski, D. (2013). Some considerations in using
prony analysis to estimate electromechanical modes. Proceedings of the 2013 IEEE
Power & Energy Society General Meeting.
21. Berti, E., Cardoso, V., González, J.A., and Sperhake, U. (2007). Mining information
from binary black hole mergers: a comparison of estimation methods for complex
exponentials in noise. Physical Review D 75 (124017): 1–17.
22. Borden, A. and Lesieutre, B. (2014). Variable projection and Prony analysis for
power system modal identification. IEEE Transactions on Power Systems 29 (6):
2613–2620.
23. Hildebrand, F.B. (1956). Introduction to Numerical Analysis. New York, NY: Dover
Publications, Inc.
24. Potts, D. and Tasche, M. (2013). Parameter estimation for nonincreasing
exponential sums by Prony-like methods. Linear Algebra and Applications 439:
1024–1039.
25. Koehl, P. (1999). Linear prediction spectral analysis of NMR data. Progress in
Nuclear Magnetic Resonance Spectroscopy 34: 257–299.
26. Kul’ment’ev, A.I. (2004). Analysis of positron lifetime spectra via a fast Prony
algorithm. The European Physical Journal of Applied Physics 25: 191–201.
27. Khazaei, J., Fang, L., Jiang, W., and Manjure, D. (2016). Distributed Prony analysis
for real-world PMU data. Electric Power Systems Research 133: 113–120.
28. Golpira, H., Atarodi, A., Amini, S. et al. (2020). Optimal energy storage system-
29. Barkhuijsen, H., De Beer, R., and Van Ormondt, D. (1987). Improved algorithm for
noniterative time-domain model fitting to exponentially damped magnetic
resonance signals. Journal of Magnetic Resonance 73: 553–557.
References 257
30. Kumaresan, R. and Tufts, D. (1982). Estimating the parameters of exponentially

damped sinusoids and pole-zero modeling in noise. IEEE Transactions on
Acoustics, Speech and Signal Processing 30 (6): 833–840.
31. Kumaresan, R., Tufts, D.W., and Scharf, L.L. (1984). A Prony method for noisy
data: choosing the signal components and selecting the order in exponential signal
models. Proceedings of the IEEE 72 (2): 230–233.
32. Almunif, A., Fan, L., and Miao, Z. (2020). A tutorial on data-driven eigenvalue
identification: Prony analysis, matrix pencil, and eigensystem realization
algorithm. International Transactions on Electrical Energy Systems: 1–17.
33. Susuki, Y. and Mezic, I. (Nov. 2011). Nonlinear Koopman modes and coherency
identification of coupled swing dynamics. IEEE Transactions on Power Systems
26 (4): 1894–1904.
34. Susuki, Y. and Mezic, I. (2012). Nonlinear Koopman modes and a precursor to
power system swing instabilities. IEEE Transactions on Power Systems 27 (3):
1182–1191.
35. Susuki, Y. and Mezic, I. (2014). Nonlinear Koopman modes and power stability
assessment without models. IEEE Transactions on Power Systems 29 (2): 899–907.
36. Hernandez-Ortega, M.A. and Messina, A.R. (2018). Nonlinear power system
analysis using Koopman mode decomposition and perturbation theory. IEEE
Transactions on Powers Systems 33 (5): 5124–5134.
37. Alassaf, A. and Fan, L. (2020). Dynamic mode decomposition in various power
system applications. Proceedings of the 2019 North American Power
Symposium (NAPS).
38. Ruhe, A. (1984). Rational Krylov sequence methods for eigenvalue computation.
Linear Algebra and its Applications 58: 391–405.
39. Tu, J.H., Rowley, C.W., Dirk, M. et al. (2013). On dynamic mode decomposition:
theory and applications. Journal of Computational Dynamics 1 (2): 31–421.
40. Bistrian, D.A. and Navon, I.M. (2015). An improved algorithm for the shallow
water equations model reduction: dynamic mode decomposition vs POD.
International Journal for Numerical Methods in Fluids 78 (9): 552–580.
41. Kutz, J.N., Brunton, S.L., Brunton, B.W., and Proctor, J.L. (2016). Dynamic Mode
Decomposition – Data-Driven Modeling of Complex Systems. Phildelphia, PA: SIAM.
42. Sakata, I., Nagano, Y., Igarashi, I. et al. (2020). Normal mode analysis of a
relaxation process with Bayesian inference. Science and Technology of Advanced
Materials 21 (1): 67–78.
43. Tissot, G., Cordier, L., Benard, N., and Noack, B.R. (2014). Model reduction using
dynamic mode decomposition. Comptes Rendus Mécanique 342 (6-7): 410–416.
44. A Study of Reactive Power Compensators for High-Voltage Power Systems. (1982).
Advanced systems technology division and transmission and distribution systems
engineering department, westinghouse electric corporation, contract 4-L60-6964P,
Final Report (12 May 1981).
45. Roman-Messina, A. (2015). Wide-Area Monitoring of Interconnected Power Systems,

Power and Energy Series 77, Stevenage. UK: IET, The Institution of Engineering and
Technology.
46. WECC (2019). Converting REEC_B to REEC_A for solar PV generators WECC
REMTF (13 June 2019). https://www.wecc.org/Reliability/Converting%
20REEC_B%20to%20REEC_A%20for%20Solar%20PV%20Generators.pdf
(accessed 2020).
47. Richwine, M.P., Sanchez-Gasca, J.J., and Miller, N.W. (2014). Validation of a
second generation type 3 generic wind model. Proceedings of the IEEE PES
General Meeting Conference & Exposition.
48. Byerly, R.T., Sherman, D.E., and McLain, D.K. (1975). Normal modes and mode
shapes applied to dynamic stability analysis. IEEE Transactions on Power
Apparatus and Systems PAS-94 (2): 224–229.
49. Román-Messina, A. (2020). Data Fusion and Data Mining for Power System
259
Solar and Wind Integration Case Studies
In previous chapters, techniques to analyze and characterize transient behavior in

power systems with high renewable power generation were introduced. This chap-
ter describes the experience in the analysis of wind and solar integration in large-
scale power grids with complex dynamics and operating characteristics.
Studies to evaluate the impact of geographically dispersed wind and solar gen-
eration on system dynamic behavior are conducted, and challenges related to the
coordination of wind and hydro generation are described. The study assesses the
impact of large clusters of wind and solar photovoltaic (PV) generation on system
dynamic performance and identifies factors affecting small and large-signal stabil-
ity and frequency regulation. A second objective is to evaluate transmission and
generation limitations.
Methods for interpreting and understanding the impact of distributed renewable
energy resources (RESs) on global oscillatory behavior are given, and tools for
studying wind and hydro generation are developed and tested. The issues of net-
work reactive control performance and tuning, inertia reduction, generation com-
mitment, grid stability, and power system frequency response are discussed in
detail. Steady-state eigenvalue analyses and detailed step-by-step simulations
are used for assessing issues regarding wind and solar integration.
It is shown that dynamic performance considerations related to voltage collapse,
oscillatory and frequency instability, and postdisturbance oscillations may limit
high wind and solar PV integration.
8.1 General Context and Motivation
Large-scale penetration of inverter-based RESs presents some challenges and

opportunities for the future power system. The use of emerging technologies such
as inverter-based wind and PV resources and the increasing size and complexity of
260 8 Solar and Wind Integration Case Studies
modern power systems are expected to introduce new interactions and operational
challenges [1–6].
Typically, factors that lead to integration issues include vulnerability to major
perturbation, reduced system inertia, rotor angle, and voltage stability, transient
stability and frequency performance, and postdisturbance oscillations [7–13]. In
addition, the successful integration of distributed RESs requires an increased level
of visibility (and controllability) of these devices [14, 15]. A minimum level of
expected performance during power system disturbances is needed, which in turn
requires knowledge of power system dynamic characteristics and the aggregated
behavior of distributed RESs [16]. Sensitivity studies and data-based analysis tech-
niques are of interest since they can supplement information on the system behav-
ior, for instance, to understand the commitment and dynamic patterns, arising
from the integration of new wind and PV generation.
Some recent investigations have also revealed a significant decline in the fre-
quency performance of some major grids [10, 11, 17], prompting concerns on
the effect of increased wind and solar penetration on the future inertial frequency
performance. Understanding the nature of this trend is therefore critical to coor-
dinating operational and control actions.
Recent advances in inertia emulation techniques have allowed modern wind
turbines to provide inertial response functionalities under low-frequency condi-
tions [18, 19]. Other developments include the use of advanced schemes for inertia
emulation [20]. However, as discussed by several authors, the role of inverter-
based inertial response and wind generation frequency response characteristics
is not well understood [17], or knowledge of the real capabilities and limitations
of inertial response from wind turbines is limited [21].
Other factors affecting system dynamic response include [22, 23]:
• The radial interconnection of wind or solar farms to the bulk transmission

system,
• The lack of adequate dynamic voltage support near areas or zones with a high
concentration of renewable generation,
• The intermittent nature and variability in power flow patterns and a more com-
plex real and reactive power dispatch,
• Inertia reduction resulting from large increases in wind and solar generation
and the associated displacement of thermal and hydro generation, along with
high reactive power losses; and
• Control mode interactions.

Massive integration of (larger amounts of ) wind-based renewable energy gener-
ation in weakly interconnected systems is influenced by many factors and con-
straints [3, 17, 24–26]. Modern generating plants such as wind and solar power
plants consist of generating units that interface to the power system through
8.2 Study System 261
frequency converters. These converters, however, are most often controlled in

such a way that the operation of the generating plants is independent of the system
frequency, resulting in reduced system inertia. A similar phenomenon is observed
in loads and other resources that utilize power electronics [7].
Recent studies and operational experience suggest that the frequency response
of major transmission systems in North America and Europe is deteriorating
[7, 24]. In this sense, a recent study [27] showed that increased levels of wind pen-
etration may lead to a reduction of the minimum short circuit levels and a reduc-
tion in system inertia. Similar observations are made in recent wind/PV
integration reports [28].
Adding new solar PV and wind power generation might further degrade fre-
quency response and voltage control. Moreover, the overall system response
may be affected by the amount and type of generation committed and how it is
dispatched. In Ref. [29], it has been found that as PV increases, the damping of
dominant interarea modes decreases. A second observation is that the mode shape
varies with the control strategy and new oscillation modes may emerge under
inappropriate control settings.
Another trend is the connection of rotating motors to the power system through
frequency converters [7]. In addition, high imports on high-voltage direct current
(HVDC) connections to other synchronous systems are expected to displace tradi-
tional generation more often resulting in decreased system inertia. These factors
will impact the system kinetic energy and inertia distribution [30].
Weak grids can pose challenges for connecting new resources and particularly
for connecting inverter-based resources [24]. In particular, it has been observed in
Ref. [27] that the trip of multiple hydropower plants on the system may result in
poorly damped oscillatory behavior during the postcontingency conditions. While
the frequency of large disturbances may not be increasing, their consequences
seem to be growing with the size and complexity of the system involved [31]. Spe-
cial protection and control actions are required to stop the power system degrada-
tion and minimize the impact of disturbances.
Potential mitigation measures include improved governor response and inertial
control from wind plants, [32]. These issues are addressed in the following sec-
tions. First, the study system is briefly described.
8.2 Study System
Figure 8.1 is a simplified single-line diagram of the study system showing the main
areas of concern for this study. Key buses and major transmission paths are indi-
cated on the schematic map. The base system used in the studies described further
Areas 1–3
Wind-farms
20–26
PV farms
1–3
Area 5
Area 4
Area 7
± 300
MVAr
Hydro
SVC 2 ∼ Generation
Area 6 (4970 MW)
SVC 1
± 300
Wind-farms
MVAr 11–19
Wind-farms 2500
MW
1–10
Figure 8.1 Simplified system representation.
is an expanded version of the given test system in Chapter 7 and includes the
detailed representation of 635 generators, 26 explicitly modeled inverter-based
wind farms (WFs), 3 solar PV farms, 5449 buses, and 5292 transmission lines. Tur-
bine governors are represented on major machines. See Section 7.1 for detailed dis-
cussions of other modeling considerations. This model has been validated using
measurement data from synchronized phasor measurement units (PMUs) [33].
Two wind and solar penetration scenarios are of interest here: (i) grid integration
of wind energy in Area 6 and (ii) wind and solar integration in Area 4 and Area 5.
The emphasis in the following analyses is placed on the scenario (i).
To evaluate the performance impact of wind and solar PV farms, appropriate
models of wind and solar farms have been developed together with special reme-
dial action schemes. Simulation studies include the detailed representation of the
machines that provide governor response and discrete control devices such as
underfrequency load shedding (UFLS) schemes.
The integration studies described further have two specific objectives: (i) define
limits of integration of wind energy in the system, and (ii) evaluate actions to
improve frequency regulation and voltage control on the 400 kV grid.
8.3 Wind Power Integration in the South Systems
Because of its geographical location and the sparse nature of the bulk transmission
network, wind integration in Area 6 is of special concern. In this system, WFs 1–19
are located at the south extremity of Area 6; the current installed wind capacity in
the study region is about 50% of the local hydropower generation. As wind pene-
tration in the study area is expected to increase, hydro plants providing governor
response will likely be displaced or be required to operate at reduced output (spin-
ning reserve) to accommodate wind variations and provide primary frequency
response as discussed in the following subsections.
Further, faults in this system resulting in loss-of-generation events will affect fre-
quency performance and result in postdisturbance oscillations.
8.3.1 Study Region

The area under study represents portions of the south systems that include 19 WFs
and a significant portion of hydro generation in the system. Figure 8.2 is a simpli-
fied diagram of this system around major WFs showing the interconnection to the
400 kV system. Three levels of transmission voltages are represented in detail: 115,
230, and 400 kV. For clarity of illustration, the low-voltage network around the
WFs below 230 and 115 kV and small WFs are not represented in the diagram.
As shown in Figure 8.2, most of the existing hydropower generation is concen-
trated on four major hydro plants and amounts to about 4970 MW. By contrast, the
wind generating capacity is over 2500 MW and constitutes about 50% of the local
installed hydro generation capacity; the total area load is 5120 MW.
The transmission system at the WFs’ point of interconnection with the system
(POIS) consists of two radial 400 kV transmission lines and a large ±300 MVAr
static volt-ampere reactive (VAR) compensator (static VAR compensator [SVC]
1 in Figure 8.2). SVC voltage support at the POIS is used to enhance the power
transfer capability of this transmission system, prevent large postfault voltage
swings, and increase reactive power reserves.
Fourteen generators (Gens 1–14) within the study region are selected to deter-
mine the synchronous generation (SG) to be displaced by wind generation. The
main characteristics of the SG and WF plants within the study region are summar-
ized in Tables 8.1 and 8.2. Since the primary frequency response of some genera-
tors is represented in detail, it can be inferred from Table 8.1 that the choice of the
Area
To Area Gens 1,2 7
7
∼
∼ Gen 14
∼
∼ Gen 13
Vpois 4970 MW
∼
SVC 2 Gens 11,12
∼ Gen 10
SVC 1
Areas Hydro machines
4 and 5 Area 6
Wind
farms 1–19
(2500 MW)
Figure 8.2 Simplified schematic of the study region within Area 6. Dashed lines represent
interconnections to neighboring systems.
synchronous machines to be displaced by wind generation is likely to have a major

impact on system dynamics. In addition to type and capacity, the use of power sys-
tem stabilizer (PSS) and governor response for the main generators are determined
in Table 8.1.
As shown in Table 8.2, the normal operating mode of doubly fed induction gen-
erators (DFIG) considered in these studies is voltage regulation and active power
control. No inertial or governor control on wind plants is provided.
The procedure adopted to assess the impact of high wind penetration on system
dynamic behavior is as follows:
• Wind generation was uniformly increased in proportion to its base case condi-
tion and generation at local hydro generators was decreased. The rationale
behind the adopted approach is discussed in Section 8.5.2.
• Among the various generating sources, hydro generators 10–14 are used for bal-
ancing wind fluctuations and utilized as a spinning reserve for major
contingencies.
Integration studies described further examine the impact of increased penetra-

tion of DFIG-based wind turbines on both transient and small-signal stability.
Table 8.1 Summary of the main generation in Area 6.
Generator Type Capacity (MW) PSS Governor response
1 Nuclear 750 No No
2 Nuclear 750 No No
3 Hydro 7.5 No No
4 Hydro 66 No No
5 Thermal 46 No No
6 Thermal 46 No No
7 Thermal 7 No No
8 Hydro 122 No No
9 Thermal 385 Yes No
10 Hydro 191 No Yes
11 Hydro 315 Yes Yes
12 Hydro 315 Yes Yes
13 Hydro 218 No Yes
14 Hydro 115 No Yes
Table 8.2 Main characteristics of DFIG WFs in Area 6.
Operating Voltage Active power

WF Type voltage (kV) control control
1–2, 5–10, 14–15, Type 3 0.69 Yes Yes

17–19 DFIG
3, 11–13 Type 3 12 Yes Yes

DFIG
4, 16 Type 3 13.2 Yes Yes

DFIG
8.3.2 Existing System Limitations

Because of their location relative to the bulk 400 kV network and the
weak transmission system, the integration of new wind generation in Area 6
presents multiple technical challenges to system security that need to be
addressed [27]:
1) Wind power concentration: The concentration of wind generation and low load
density within the study region. As discussed earlier, a large amount of wind
power is concentrated on a small geographical region.
2) Wind variability: Wind power fluctuations ranging from 0 to 100 MW within an
hour are not uncommon, especially early in the morning (during off-peak
hours). Further, wind generation is out of phase with load demand thus adding
to system variability.
3) Relative size: As noted earlier, the size of the WFs relative to hydro generation in
the south systems is high (approximately 50%). As a result, the automatic gen-
eration control (AGC) system must compensate for fluctuations in wind power
and load variations,
4) Point of interconnection: WFs 1–19 are connected radially to the 400 kV trans-
mission system. A large ±300 MVar SVC is used to support local voltage. As a
result, loss of SVC voltage support or loss of major transmission lines may result
in system instability or poor postdisturbance system response.
5) Impact on reserve margin: Present levels of wind power penetration in the south
systems may affect operating reserve margins.
In addition to these limitations, power transfers to the bulk 400 kV system are
constrained by grid stability and voltage considerations.
8.4 Impact of Increased Wind Penetration on the

System Performance
Detailed transient stability studies were conducted to evaluate the impact of high
wind penetration on system stability and performance. The following discussion
describes system response in the context of grid integration studies.
8.4.1 Study Considerations and Scenario Development

Table 8.3 summarizes the contingency scenarios selected for analysis. These con-
tingencies represent various operating scenarios that result in poorly damped
power oscillations.
Table 8.3 Contingency scenarios selected for the study.
Contingency
scenario Remarks
CS01 Outage without fault of transmission line 7415–8307 in Area 6
CS02 Three-phase stub fault at bus 153 in Area 3
CS03 Three-phase fault at the POIS of WFs 1–19 in Area 6
CS04 Outage without fault of transmission line 8024–8052 in Area 6
CS05 The simultaneous outage of two 400 kV lines 8024–8520 and

7415–7426 in Area 6
CS06 The outage of generator 1 unit 1 (750 MW) in Area 6
CS07 The outage of generator 10 units 1 and 2 (382 MW) in Area 6
In these analyses, three scenarios of wind penetration were selected for study:
Wind scenario 1 (WS1): This is the base operating case with about 400 MW wind
generation in the south systems, representing about 4% wind penetration rela-
tive to the installed hydro generating capacity in generators 10–14 in Figure 8.2.
Wind scenario 2 (WS2): A low wind generation scenario.
Wind scenario 3 (WS3): A 20% wind scenario.
In all cases, flow patterns were adjusted to reflect current operating policies in
the system. Note that, because some of the hydropower plants are equipped with
PSSs, adding wind generation may affect the damping of major interarea modes
associated with the study region.
8.4.2 Base Case Assessment

Figures 8.3–8.7 show the system time response of the base system to a three-phase
short circuit at the POIS located close to the WFs (wind scenario CS03 in
Table 8.3). Other contingency scenarios are discussed further.
As shown in these plots, critical faults in the neighborhood of the POIS can
result in marginally damped power, frequency, and voltage oscillations in which
wind generators participate in the (slowest) 0.394 Hz interarea mode 1; these oscil-
lations are observed in the bus terminal frequency, the WFs’ active (and reactive)
power output and nearby SVCs. Refer to Table 7.15, in Chapter 7, for details about
the main interarea modes in the system.
80
WF 10
70
60
50
WFs 3,9,17,19
Power [MW]
40
30 WFs 14,15
20
10
WFs 4–8, 11–13,16,18
0 WFs 1,2
0 5 10 15 20 25 30
Time [sec]
Figure 8.3 WFs’ active power output following a stub three-phase fault at the POIS
(contingency scenario CS03).
1.04
1.03
1.02
Voltage [pu]
1.01
SVC 1
1
0.99
0.98
0 5 10 15 20 25 30
Time [sec]
Figure 8.4 Wind farms terminal voltages following a three-phase stub fault at the POIS
For reference, Table 8.4 presents a comparison of the frequency and damping of
the dominant mode for the speed deviations of WFs for the contingency scenarios
in Table 8.3. In all cases, WFs are found to have dominant participation in the
200
100
Reactive power [MVAr]
–100
–200
SVC 1
–300 SVC 2
SVC 3
0 5 10 15 20 25 30
Time [sec]
Figure 8.5 SVC reactive power output following a three-phase stub fault at the POIS
0.394 Hz mode. From these results, contingency scenarios CS03, CS06, and CS07
are selected for analysis in the discussion that follows.
8.4.2.1 System Oscillatory Response

From Figure 8.6, the observed bus frequencies at the generator and WF locations
exhibit a dominant mode at about 0.396 Hz showing that both WFs in Area 6 and
60.1
60.08
60.06
Frequency [Hz]
60.04
WFs 20–26 WFs 1–19 0.396 Hz
60.02
60
59.98
0 5 10 15 20 25 30
Time [sec]
Figure 8.6 WFs’ bus frequencies following a three-phase stub fault at the POIS
600 WF 12
500
400 WF 1
300 WF 14
Qwind [MVAr]
200
100 WFs 2,4–5,9–11,13–19
0
–100
WF 8
–200
WF 7 WFs 3,6
–300
–400
0 5 10 15 20 25 30
Time [sec]
Figure 8.7 Wind farms reactive power output (contingency scenario CS03).
Table 8.4 Koopman analysis of rotor speed deviations for different contingency scenarios
(wind scenario WS1). The bold entries indicate critical contingency scenarios with the lowest
post-disturbance damping characteristics.
Contingency scenario Frequency (Hz) Damping (%)
CS01 0.397 0.555
CS02 0.395 0.475
CS03 0.396 0.309
CS04 0.395 0.445
CS05 0.395 0.687
CS06 0.395 0.186
CS07 0.396 0.296
WFs in areas 1–3 have an important participation in the slowest interarea mode in
the system.
Insight into the nature of the WFs’ contribution to these oscillations can be
gleaned from the analysis of the mode phase for selected machines. Table 8.5
shows the amplitude and phase associated with the 0.394 Hz mode for WFs 1,
5, 24, and 25 and Gens 10–13 in Table 8.1, extracted using Koopman mode anal-
ysis. It is observed that WFs 1 and 5 swing in opposition to WFs 24 and 25 and Gens
Table 8.5 Phase relationships for dominant system generators extracted from generator
speeds (contingency scenario CS03).
Generator Area Frequency (Hz) Amplitude Phase ( )
WF 1 6 0.396 0.0690 0.00
WF 5 6 0.395 0.0680 4.22
Gen 10 6 0.395 0.0130 104.75
Gen 11 6 0.395 0.0138 104.83
Gen 12 6 0.395 0.0136 104.86
Gen 13 6 0.395 0.0137 105.40
WF 24 3 0.397 0.0229 207.56
WF 25 3 0.397 0.0228 203.40
10–13 suggesting a local exchange of energy through the 400 kV transmission sys-
tem in Figure 8.2. WFs 1 and 5 are found to have the largest participation in the
0.394 Hz mode.
8.4.3 High Wind Penetration Case

In this case, a worst-case scenario involving a critical contingency at the POIS was
selected for analysis. Figures 8.8 and 8.9 show selected dynamic performance
20
WFs
3,17,19
15
10 WFs
4,5,7,8,16
5
Q [MVAr]
0 WFs 1,2 WFs 11–14,18
–5
WFs 9,10
–10
–15
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
Figure 8.8 WFs reactive power output (contingency scenario CS03).

700
SVC 12
600
500
400
Q [MVAr]
300
200
SVC 1
100
0 SVC 8
–100 SVC 7
–200
0 2 4 6 8 10 12 14 16 18 20
Time [sec]
Figure 8.9 SVC reactive power output (contingency scenario CS03).
simulations for contingency scenario CS03. In this analysis, it is assumed that wind
generation displaces hydroelectric generators equipped with PSSs.
From the results of Chapter 7, simulated data are approximated by a Koopman
representation of the form
m
yk = λkj ψ j x o v j 81
j=1
Close analysis of simulations for high wind generation cases in Figures 8.8 and
8.9 shows undamped reactive power oscillations involving two major electrome-
chanical modes: a local mode at about 1.36 Hz and the 0.394 Hz interarea mode
1 (m = 2 in Eq. (8.1)). The 1.36 Hz mode is of special interest since it has no coun-
terpart in linear (small signal) studies; similar findings are described in Ref. [29].
Simulation results highlight three key points: (i) careful analysis of Figures 8.8
and 8.9 shows that reactive power outputs at WFs swing in phase with nearby
SVCs (SVCs 1,7, 8, and 12); (ii) while not explicitly addressed, the WFs swing
out of phase with nearby hydro generators (Gens 10–14); and (iii) the oscillations
manifest themselves as nonlinear oscillations with a dominant component at
1.36 Hz.
As a first step toward assessing the nature of energy exchange associated with
these modes, phase relationships for several generators and wind generators were
determined using dynamic mode decomposition (DMD) analysis in Chapter 7. In
the interest of simplicity, Gens 1–14 in Table 8.1 and 19 WFs (Gens 1–19) were
selected for the study.
Toward this end, the measurement matrix, Xω, associated with wind and SG in
Area 6 is represented as a row-concatenated matrix
X ω = X ωwind X ωgen 82
where X ωwind = ΔωWF1 ΔωWF2 ΔωWF19 T

, X ωgen = Δωg1 Δωg2
T
Δωg14 , and the Δω i represent speed deviations. The accuracy of the model
was improved by normalizing the individual data sets using the approach sug-
gested in Section 5.4.
The idea is to expand the data matrix, Xω, as a linear combination of functions of
space and time as:
X ω = x1 x2 … xd = ΦΛΓm t 83
or
X ω = x1 x2 … xd = ΦV vand c 84
where, for clarity of illustration, the DMD method is used in the analysis, and the
symbols in Eqs. (8.3) and (8.4) have the usual interpretation in Section 7.2.3.
In the studies presented further, generation was rescheduled to account for
increases in wind generation. Starting from the base case, wind generation was
increased, and active and reactive power was rescheduled.
For reference and validation, Table 8.6 shows the extracted modes using
Koopman/DMD analysis for the high wind penetration scenario. Similar results
are obtained for other operating conditions.
Further, Figure 8.10 shows the speed-based mode shape for the 1.36 Hz mode
extracted using this approach.
Analysis of the mode shape in Figure 8.10 reveals several significant results.
First, WFs 1–19 swing in opposition to machines 1–14 revealing the local nature
Table 8.6 The slowest modes of oscillation of the test system (contingency scenario CS03).
Eigenvalue Frequency (Hz) Damping in %
1 0.399 0.357
2 0.545 7.001
3 0.696 5.26
4 0.701 7.42
5 0.748 3.512
0.25
Gen 10
Gens 11,12
0.2
0.15
Real part of mode
Gens 13,14
0.1
Gen 8
0.05
WFs 1–19
0
Generator
WFs 9–11
–0.05
Generator/wind farm
Figure 8.10 Speed-based mode shape (contingency scenario CS03).
of this mode; Gens 10–12 (hydro machines) and Gens 13–14 are seen to have the
largest contribution to the mode. The WFs, on the other hand, show a similar con-
tribution in terms of amplitudes and exhibit the largest participation in the mode.
Physically, the form of the mode indicates the oscillatory mechanism involved in
the observed time-domain responses.
8.5 Frequency Response

Disturbances that result in the loss of generation within the study area are of spe-
cial concern. Three aspects of frequency dynamics are of interest in wind integra-
tion studies [24]: (i) inertial frequency response, (ii) the postdisturbance frequency
fluctuations (primary frequency response) following major generator outages, and
(iii) the amount and location of contingency reserves.
Inertial frequency response is inherent in the system and is associated with the
rotating characteristics of loads and SG [6]. As a result, the amount and location of
system inertia may affect it. Postdisturbance frequency fluctuations may be asso-
ciated with inappropriate control settings and other operating and structural issues
as discussed in the following section.
8.5.1 Frequency Variations

Figure 8.11 gives a conceptual representation of the system model used to assess
inertial system response. A measured frequency excursion is also shown caused by
a loss of generation event in the test system [33]. As discussed in Chapters 4 and 7,
the location of contingency reserves plays a critical role in the ability of the system
to limit frequency variations [34]. A simple criterion to estimate the effect of gen-
erator dropping on the frequency behavior at bus j, Δfj, can be obtained from the
sensitivity relationship [27]
∂ f j ∂Pgout
Δf j = ΔPwgen 85
∂Pgout ∂Pwgen
where Pgout and Pwgen are the amount of generator outage and WF active power,
respectively, and the sensitivities can be estimated from measurements or simula-
tion studies involving frequency behavior to selected generator outages and wind
generation scheduling strategies.
Another way of looking at frequency dynamics is by using two-area inertia-
based dynamic equivalents derived from the center-of-gravity (COG) formulation
in Chapter 6. Figure 8.11a gives a conceptual representation of this model, show-
ing a recorded frequency disturbance event following the loss of a large hydro gen-
erating plant in Area 6. Figure 8.11b, in turn, shows the equivalent system for the
analysis of wind and hydro coordination and frequency response analysis.
Using the same notation as in Section 6.2, the frequency deviation of the partial
centers of inertias (COIs) can be approximated as [35, 36]

Ptie MMG
M COG j
Mi ΔPmech − ΔPelec
Δ f COI i = Δ f COG 86
M COG 2πD
1+ Δ f COG
M i ΔPmech − ΔPelec
From (8.6), the relationship between the frequency of the COG and the motion
of the local centers of angle is first determined, and expressions to compute local
frequency deviations following major disturbances can be derived. Formally, the
frequency deviation at bus j can then be approximated as

Ptie MMG
Mi j
Mj ΔP j − Ptie
ij − η j ΔP i
MMG
Δf j = Δ f COI i 87
Mi 2πD
1+ Δ f COI i
M j ΔP j − P − η ΔPMMG
tie
ij j i
where the symbols have been already described.

This approach provides alternate, complementary information concerning the
COI frequency [10] and the local (cluster-based) frequency estimated using the
c-means approach in Chapter 7.
(a)
60 Hz
Intertial
response Area 2
Nadir
Local COI
dynamics

tie
PCOI 1,COG tie MCOG
PCOI2,COG [∆Pi – ∑ PCOI
tie
i,COG
]
MCOIi j
ΔfCOIi = ΔfCOG
MCOG 2πD
δCOI1, MCOI1 ΔP+ ΔfCOG
MCOIi
COG
tie
Area 6 PCOI j,COG δCOG , MCOG
tie
PCOI Magnitude of
n,COG
disturbance
δCOI6, MCOI 6 Area 7

δCOI7, MCOI 7
∂fi ∂fCOIi
Δfj(t) = Δf (t)
∂fCOIi ∂fCOG COG
(b)
P
COG SG
–C
OG
G
CO
F–
W
P
PWF–SG ∼ Gen 14
VPOIS
∼ Gen 13
∼ Gens 11,12
∼ Gen 10
WFs 1–19 SVC 1
Hydro machines
Figure 8.11 The COG dynamic equivalent adopted for assessing energy exchange and wind
and hydro coordination; (a) multiarea representation and (b) two-area dynamic equivalent.
Key observations are:
• Frequency sensitivities can be obtained for various values of transmission and

inertia in the system, which allows determining the maximum amount of wind
energy penetration as a function of the frequency nadir [35].
• As a by-product, this approach allows determining the minimum amount of

inertia required in the equivalent system (hydroelectric machines).
Once frequency sensitivities are estimated, the partial sensitivities in (8.5) can be
determined, as suggested in Section 8.5.3.
8.5.2 Wind and Hydropower Coordination

A key problem of interest in assessing inertial frequency response is that of the
coordinated operation of hydro and wind generation [28]. As noted previously
and suggested in Figure 8.11, adding wind generation in Area 6 displaces local
hydro generation (mainly Gens 10–14) making the development of commitment
and dispatch of hydro generation procedures critical to assess frequency response
and resulting in low system inertia and high rate of change of frequency (RoCoF)
during system disturbances.
The first key issue is the correlation between wind generation and the hydro gen-
eration that needs to be displaced [28]. In this discussion, two aspects are of inter-
est: the location of the machines where the spinning reserve is needed (inertial
support) and the magnitude and distribution of the required spinning reserve.
The first issue is addressed here using analytical models such as those in Eqs. (8.5)–
(8.7) and the use of correlation analysis. The second requires extensive dynamic
simulation studies.
Inspired by techniques that fuse or regress data, this section explores the use of
correlation techniques to determine suitable machines to maintain spinning
reserves. To introduce these ideas, consider two data sets X ᑬN × m1 and Y
ᑬN × m2 , where N is the number of snapshots, and m1, m2 denote the number of
sensors or measuring locations, which in the more general case are assumed to
represent distributed generation and conventional SG in (8.3), respectively [37].
In the present context, they could represent wind electric output power and
hydro (thermal) active generation output and are assumed to have the same length
N and have different measured signals or sensors, i.e. m1 m2. The relationship
between these two centered sets of variables is given by the covariance matrix
R ᑬN × m1 = Y T X.
Conceptually, these techniques decompose the data into a set of statistically
independent and orthogonal components. Each component is defined in time
by a unique series of weights assigned to each of the time samples; these weights
are called component loadings. The extent to which each principal component
(PC) contributes to a particular oscillation in the original data set is reported as a

component score [38].
The process can be summarized in three steps:
1) Create a concatenated data matrix from the individual models of distributed

generation (i.e. wind and solar generation) and conventional SG (see Eq. (8.2)),
2) Perform correlation analysis on datasets X, Y,
3) Extract correlation measures.
To examine the applicability of this approach, active power responses were cal-
culated for each wind generator and SG in the study system.
From partial least squares (PLS) regression analysis in Chapter 5, the measure-
ment data are defined as
T
X = X sm = P gsmj1 P gsmj2 … P gsmj14
T
88
Y = X wf = P gwfj1 P gwfj2 … P gwfj19
where for the study system, P gwfj = Pg j t 1 Pg j t 2 Pg j t N , j = 1, …, 19,

Pgsmj = Pg j t 1 Pg j t 2 Pg j t N , j = 1, …, 14 and N = 3603 samples.
For completeness, Figure 8.12 shows the time evolution of active power devia-
tions for contingency scenario CS07.
Using these assumptions, the correlation between data sets X, Y can be
expressed as
X = t 1 pT1 + E = TPT + E
89
Y = u1 qT1 + F = UQT + F
in which, as discussed in Chapter 5, the T, U are matrices of the dominant

extracted score vectors, matrices P, Q, represent the matrices of loadings, and
E, F are the matrices of residuals. The results for this scenario are shown in
Figure 8.13, using the approach in Ref. [37].
In this case, the percentage of variation (PCTVAR), explained by the three dom-
inant PCs, is
0 8899 0 0899 0 0101
PCTVAR = 0 1021 0 0190 0 0289 8 10
PC1 PC2 PC3
It is apparent from column 1 in Eq. (8.10) that, overall, synchronous machines in

Area 6 (first row) have the largest participation in mode 1 relative to WFs 1–19
(second row). In sharp contrast with this, PC modes 2 and 3 (columns 2 and 3
in Eq. (8.10)) are found to have a minor contribution to the observed oscillation.
(a)
1000
900
Power [MW]
Gen 1
800
700
600
2 4 6 8 10 12 14 16 18 20
Time [sec]
(b)
50
Power [MW]
40 Gen 6
30 Gen 9
20
0 5 10 15 20
Time [sec]
(c)
400
Gen 11
300
Power [MW]
200 Gen 13
Gen 14
100
Gen 10
0
2 4 6 8 10 12 14 16 18 20
Time [sec]
Figure 8.12 Generator active power response following the simultaneous outage of two
generators (Contingency scenario CS07) for: (a) nuclear generation, (b) thermal generation,
and (c) hydro generation.
In addition, one can infer from Figure 8.13 that WFs 3, 9–10, and 17, 19 have the
largest contribution to the postdisturbance scenarios. In turn, nuclear generators 1
and 2 and hydro machines 9–13 show the largest contribution to the oscillation;
the opposite is true for machines 3–8. Because nuclear generators are not expected
to provide governor response, hydro generators 9 through 13 are selected for
analysis.
(a)
200
150
Magnitude
100
50
0
0 5 10 15 20
Wind farm
(b)
300
Nuclear
generators
Magnitude
200 Hydro
generators
100
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Generator
Figure 8.13 Partial least squares regression between WFs and generation sources in the
south systems (contingency scenario CS07); (a) WFs 1–19 and (b) selected generators in
Area 6.
From the system viewpoint, this information helps to easily identify the location
and distribution of generation to be displaced; these findings can also be used to
coordinate operational actions. A related issue is that of inertia support in hydro
machines, as some of them could be operated as synchronous condensers.
8.5.3 Response to Loss-of-Generation Events

Large loss-of-generation events may result in complex postdisturbance oscillations
and low frequency nadirs [27, 33]. To investigate inertial and primary frequency
response, three-generation trip scenarios are considered: (i) loss of the largest gen-
erating unit in the system (750 MW), (ii) loss of a hydro generation without PSSs
(315 MW), and (iii) loss of a large generation unit equipped with a PSS (315 MW).
For illustration, two wind penetration scenarios are considered for analysis: (i) a
base case with 400 MW wind generation and (ii) a high wind generation case. In
both cases, the increase in wind generation was offset by disconnecting main
hydro generators located near the cluster of WFs 1–19 as suggested by analytical
60
Base case
59.98 Base case with 7% wind penetration
Base case with 15% wind penetration
59.96
59.94
Frequency [Hz]
59.92
59.9
59.88
59.86
59.84
59.82
59.8
0 5 10 15 20 25 30
Time [sec]
Figure 8.14 Frequency response at the POIS following the outage of a 750 MW nuclear
unit (contingency scenario CS06).
results in Section 8.5.2. The metrics of interest are the RoCoF, the frequency nadir,
and the settling frequency [10, 11].
Figure 8.14 is an example of postdisturbance frequency oscillations following a
750-MW generation loss (contingency scenario CS06). As shown in this plot, large
loss-of-generation events result in negatively damped oscillations with a frequency
of about 1.36 Hz (5–25 seconds). The mode at 1.36 Hz is identified as a local oscil-
lation between the western portion (WFs 1–19) and the eastern portion (hydro
generators) of Area 6.
For the base case, the frequency response reaches a nadir of about 59.9 Hz at
approximately 3.8 seconds after the generation trip; while not shown in
Figure 8.14, the frequency settles back to the prefault (nominal) condition at about
40 seconds. Several differences are obvious when comparing system performance.
It can be seen that the slope of the inertial response increases as the amount of
connected wind generation increases. Additional studies show that this effect is
more pronounced in the case of stressed operating conditions.
Table 8.7 shows the eigenvalues computed using Koopman mode analysis for
the contingency operating conditions in Figure 8.4. In this analysis, the study win-
dow is chosen to coincide with the frequency primary response period (5–15
seconds).
Analytical experience with other operating scenarios shows that frequency
nadirs range from 59.92 Hz for low wind penetration levels to 59.65 Hz for high
wind penetration levels for the cases with double- and triple-generation
Table 8.7 Prony results for frequency deviation signals in Figure 8.14 following the loss of
critical system generators for contingency scenarios CS06 and CS07 in Table 8.1 (time
interval 15–30 seconds).
CS07-loss of two
CS06-loss of unit 1 of generating units
Gen 1 (750 MW) (180 MW each)
Operating scenario Frequency Damping Frequency Damping
Base 0.394 0.412 0.397 0.419

Base case + 20% increase in generation 0.393 −0.452 0.397 0.211
Maximum wind condition 0.393 −0.452 0.398 −0.397
1.366 −0.010 1.378 0.010
contingency scenarios as shown in Figure 8.15. As expected, the critical mode is

affected significantly depending on the amount of generation disconnected.
The analysis also suggests the need to coordinate turbine governor settings and
other controllers to damp out the postdisturbance oscillations.
59.95
59.9
300 MW generator tripping
Frequency nadir [Hz]
59.85
59.8
59.75 600 MW generator tripping
59.7
59.65
1000 1500 2000 2500 3000
Wind penetration level [MW]
Figure 8.15 Frequency nadir as a function of the magnitude of generator tripping.

8.6 Effect of Voltage Control on System Dynamic

Performance
The advent of modern, inverter-based wind and solar farms offers a direct means
for controlling system voltage. Specifically, the WFs in the south systems are
expected to provide local voltage support (Figure 8.16). Two main factors influence
the effect of voltage control on system dynamic performance: (i) the location, size,
and control characteristics of reactive support sources near the wind/solar PV
plants; and (ii) the control characteristics of WFs.
8.6.1 Voltage Support and Reactive Power Dispatch

Because of the geographical location of wind generators, and the connection of
large SVCs around the POIS, the need arises for coordinated voltage control. As
illustrated in Figure 8.2, a large (400 kV, ±300 MVAr) SVC is connected near
the WFs’ POIS.
Major sources of reactive support in the area also include major hydro generat-
ing units in the area.
Three control strategies are suggested:
1) Individual voltage control of WFs

2) Hierarchical voltage control of the 19 WFs at the POIS
3) Hierarchical, voltage control involving the local SVC and WF control
The first issue is studied here. Based on these studies, the need for coordinated
voltage control is established.
8.6.2 Effect of Voltage Control Characteristics

Assessing the impact of voltage control characteristics on system dynamic perfor-
mance is also integral to the wind and solar PV integration studies. As discussed in
the introductory section, voltage limitations in the south systems impose further
constraints on wind generation and may be compounded by a weak connection
grid at the POIS [24].
In the literature, a variety of different voltage control strategies are considered at
WFs and dynamic reactive power compensators. Three operating modes are con-
sidered here [8]: (i) constant power factor control, (ii) constant Q control, and
(iii) bus voltage control. Operating in voltage control mode is often considered
to be the most efficient way to use their static and dynamic reactive power
capability, which can reduce the burden on nearby SGs and SVCs on the
power system [14].
Vref
Ibranch +
1
∣Vreg – (Re+ jXc)Ibranch∣ 1
∑
1 + sTstfr –
Vreg 0 qmax
emax Qext
+ K
Qbranch + KPq + qi 1
∑ s 1 + sTp
Kc
emin
qmin
Qbranch 1
∑
1 + sTstfr –
+
Qref
Figure 8.16 Simplified diagram of the WF reactive power control loop.
Figure 8.16 depicts the WF reactive power control loop used in the simulations.
The primary voltage control loop includes a proportional–integral controller and
droop control; the adjustable droop is obtained by the feedback of the WF reactive
current and voltage and modulates the reference voltage, Vref. The difference value
is then processed through the PI controller to obtain the reactive command for the
output, Qext.
Further, through the modification of the reactive power reference, Qref, the WFs
can maintain constant reactive power output. A slow reactive control loop, auto-
matic reactive power regulator (AQR) can also be utilized for the WFs. A suitable
coordination control between the SVCs and the WFs [39] may be necessary as dis-
cussed in Chapter 5.
As a first step, extensive power flow analyses were conducted to determine the
impact of transmission system limitations on the energy deliverability of the wind
plant output. In these simulations, static compensation with constant reference
voltage control is considered at SVCs 1 and 2 in Figure 8.2. Table 8.8 shows the
SVC output reactive power as a function of the WFs’ reactive power output
Table 8.8 SVC reactive power as a function of WFs’ loading. The voltage control loop in
Figure 8.16 is enabled.
SVC WS1 (base case)a WS2 (20%)a WS3a
1 −200.78 −127.33 208.8

2 4.55 −6.05 11.44
a MVAr.
obtained from a steady-state power flow solution for the case when WFs 1−19
operate on voltage control mode. Three cases are considered, namely (i) wind sce-
nario 1 (WS1) as the base case, (ii) wind scenario 2 (WS2), and (iii) wind scenario
3 (WS3).
By increasing the WFs’ output, the reactive losses on the transmission system
connecting the WFs with the 400 kV system (see Figure 8.11) increase, and the
margin of dynamic reactive power to support voltage recovery decreases. Increased
reactive losses, in turn, increase the possibility of voltage collapse as discussed
further.
Based on these results, studies were conducted to assess voltage stability con-
straints. Emphasis was placed on the impact of solar and wind generation on
power system performance. From the perspective of system modeling, three main
aspects are of interest [1]: (i) optimizing voltage control, (ii) tuning voltage/VAR
control loops [40], and (iii) providing sufficient static and dynamic reactive power
capability [41]. These features are interrelated as discussed further.
To pursue these concepts, Figure 8.17 shows plots of terminal voltage at SVC 1
for various control strategies. The postdisturbance oscillatory response shows a
marginally stable ( f = 1.366 Hz, ξ = 0.011% oscillation and a negatively damped
oscillation at 0.396 Hz (ξ = − 0.759%). Modal analysis discloses higher harmonics
at 2.732 Hz (ξ = − 0.012%) revealing the nonlinear nature of the observed
oscillations.
(a)
1.03
Voltage [pu]
1.028
With SVC 1 and WFs in constant Q control mode

1.026 With SVC 1 and WFs in voltage control mode
1.024
0 5 10 15 20 25 30
Time [sec]
(b)
1.03
Voltage [pu]
1.028
With low gain at SVC 1 and WFs in Q control mode

1.026
0 5 10 15 20 25 30
Time [sec]
Figure 8.17 Bus voltage magnitudes at the POIS for various control and dynamic support
characteristics (contingency scenario CS06); (a) SVC voltage support at the POIS and nominal
SVC gain and (b) SVC voltage support at the POIS with low SVC gain.
Bus 81
1.1 Bus 132
Bus 136
Bus 73
Bus 40
Bus voltage magnitude [pu]
1.05
0.95
0.9
0 5 10 15 20 25 30
Time [sec]
Figure 8.18 Bus voltage magnitudes at selected buses in Area 6 (contingency

scenario CS06).
A crucial observation is that the study results in Figure 8.17a show that constant
Q control may result in mid-term voltage instability. In turn, it can be seen in
Figure 8.17b that improper SVC tuning can also result in degraded performance
and severe voltage fluctuation or instability. Voltage instability is observed in
Figure 8.17a at about 28 seconds into the simulation. The effect becomes more pro-
nounced as the system is further stressed.
The need for additional (or more coordinated) dynamic reactive support is evi-
dent and deserves further investigation.
Based on the preceding results, system studies were conducted to identify modal
characteristics in the observed oscillations. Figure 8.18 illustrates the results of
transient stability simulations for contingency scenario CS06.
The nature of these oscillations becomes evident from the Fourier spectra of the
observed oscillations in Figure 8.19 and Prony analysis of selected simulations in
Table 8.9. From this Table, hydro plant generators and WFs in Area 6 (WFs 1 and
8) are seen to swing in opposition to WFs in Area 3 (WFs 23 and 26) and solar PV
farm 1 in Area 4 as expected from physical considerations. It should be noted that
the larger modal voltage swings for the dominant mode are located in zones or
buses without voltage regulation (Bus 73) in agreement with Fourier spectral anal-
ysis in Figure 8.19.
Bus 81
40 Bus 136
Bus 73
Bus 40
20
Magnitude [dB]
–20
–40
–60
–80
0 0.5 1 1.5 2 2.5 3
Frequency [Hz]
Figure 8.19 Fourier spectra of bus voltage magnitude deviations in Figure 8.18.
Table 8.9 Prony analysis of selected bus voltage magnitudes.
Generator/bus Area Amplitude Phase ( )
Gen 10 6 0.0120 −59.71
Gen 11 6 0.0119 −59.70
Gen 12 6 0.0121 −58.49
WF 1 6 0.0105 −85.22
WF8 6 0.0105 −78.96
POIS 6 0.0106 −61.93
SVC 2 6 0.0102 −72.65
Bus 73 6 0.144 −57.71
WF 23 3 0.0059 32.12
WF 26 3 0.0061 39.04
PV 1 4 0.0032 43.05
8.7 Summary
In this chapter, studies to evaluate the impact of high wind penetration levels on
the dynamic performance of a large, realistic power system model have been con-
ducted. The studies examine the effect of geographically disperse wind and solar
generation on transient stability and frequency regulation. The major emphasis
was directed toward the problem of wind and hydro coordination and the study
of the effect of coordinated voltage control on system dynamic performance.
The applicability of correlation techniques to determine the conventional SG gen-
eration to be displaced by WF integration is also pointed out.
System stability studies have shown that dynamic performance considerations
including voltage collapse, oscillatory and frequency instability, and postdistur-
bance oscillations may limit high wind and solar PV integration. The combination
of these constraints makes the analysis of the high integration of wind and solar PV
penetration difficult.
References
1. Miller, N.W., Larsen, E.V., and MacDowell, J.M. (2004) Advanced control of wind
turbine-generators to improve power system dynamic performance. Proceedings of
the IEEE International Conference on Harmonics and Quality of Power.
2. CIGRE Working Group C.6.08. (2011). Grid integration of wind generation
(February 2011).
3. Alberta Electric System Operator, Wind Integration Impact Studies Phase 2. (2006).
Assessing the impacts of increased wind power on AIES operations and mitigating
measures (18 July 2006), Revision 2.
4. Miller, N.W., Guru, D., and Clark, K. (2008). Wind generation applications for the
cement industry. Proceedings of the 2008 IEEE Cement Industry Technical
Conference Record.
5. Winter, W. (ed.) (2010). European wind integration study (EWIS) – EWIS final
report, Brussels. http://www.wind-integration.eu (accessed 2020).
6. Sharma, S., Huang, S.H., and Sarma, N., System inertial frequency response
estimation and impact of renewable resources in ERCOT interconnection.
Proceedings of the 2011 IEEE Power and Energy Society General Meeting.
7. European Network of Transmission System Operators for Electricity (ENTSO-E).
Future system inertia – Report prepared by Energienet.dk, Fingrid, Stanett and
Svenska krafnät, Brussels. https://eepublicdownloads.blob.core.windows.net/
public-cdn-container/clean-documents/Publications/SOC/Nordic/
Nordic_report_Future_System_Inertia.pdf (accessed 2020).
References 289
8. Vittal, E., O’Malley, M., and Keane, A. (2010). A steady-state voltage stability
analysis of power systems with high penetrations of wind. IEEE Transactions on
9. Vittal, V. and Ayyanar, R. (2013). Grid Integration and Dynamic Impact of Wind
Energy. New York, NY: Springer.
10. Tan, J., Zhang, Y., You, S. et al. (2018). Frequency response study of a U.S. Western
interconnection under extra-high photovoltaic generation penetrations.
Proceedings of the IEEE Power and Energy Society General Meeting.
11. Tan, J., Zhang, Y., Veda, S. (2017). Developing high PV penetration cases for
frequency response study of U.S. Western Interconnection. Proceedings of the
IEEE Green Technologies Conference.
12. Quintero, J., Vittal, V., Heydt, G.T., and Zhang, H. (2014). The impact of increased
penetration of converter control-based generators on power system modes of
oscillation. IEEE Transactions on Power Systems 29 (5): 2248–2256.
13. Clark, K., Freeman, L.A., Jordan, G.A. et al. (2010). Impact of high levels of wind
and other variable renewable generation on the grid operation: summary of major
US studies. CIGRE 2010, C5_209_2010.
14. AEMO. (2019). Australian energy market operator, maintaining power system
security with high penetrations of wind and solar generation. International
Insights for Australia (October 2019).
15. MIGRATE – Massive InteGRATion of Power Electronic Devices. (2016).
Deliverable D.1.1, Report on Systemic Issues. Tech. Rep. https://www.h2020-
migrate (accessed 2020).
16. European Network of Transmission System Operators for Electricity (ENTSO-E).
(2014). Dispersed generation impact on CE region security: dynamic study – 2014
report update – report of ENTSO-E SG SPD. http://www.entsoe.eu (accessed 2020).
17. Gevorgian, V., Zhang, Y., and Ela, E. (2015). Investigating the impacts of wind
generation participation in interconnection frequency response. IEEE Transactions
on Sustainable Energy 6 (3): 1004–1012.
18. Caldas, D., Fischer, M., and Engelken, S. (2015). Inertial response provided by full-
converter wind turbines. Proceedings of the Windpower 2015 Conference and
Exhibition.
19. Laudahn, S., Seidel, J., Engel, B. et al. (2016). Substitution of synchronous
generator based instantaneous frequency control utilizing inverter-coupled DER.
Proceedings of the 2016 IEEE Seventh International Symposium on Power
Electronics for Distributed Generation Systems (PEDG).
20. Hydro-Québec TransÉnergie (2009). Transmission provider technical
requirements for the connection of power plants to the hydro Québec transmission
system. http://www.hydroquebec.com/transenergie/fr/commerce/pdf/
exigence_raccordement_fev_09_en.pdf (accessed February 2009).
21. Fischer, M., Engelken, S., Mihov, N., and Mendonca, A. (2016). Operational
experiences with inertial response provided by type 4 wind turbines. IET Renewable
Power Generation 10 (1): 17–24.
22. Gautam, D., Vittal, V., and Harbour, T. (2009). Impact of increased penetration of
DFIG-based wind turbine generators on transient and small signal stability of
power systems. IEEE Transactions on Power Systems 24 (3): 1426–1434.
23. MIT Energy Initiative. (2011). Managing large-scale penetration of intermittent
renewables. An MIT Energy Initiative Symposium (20 April 2011).
24. NERC. (2012). Frequency response initiative report - the reliability role of
frequency response. Draft. http://www.nerc.com/docs/pc/FRI%20Report%209-30-
12%20Clean.pdf (accessed September 2012).
25. North American Electric Reliability Corporation. (2017). Integrating inverter based
resources into low short circuit strength system, reliability guideline
(December 2017).
26. Milligan, M., Lew, D., Corbus, D. et al. (2009). Large-scale wind integration studies
in the United States: preliminary results. Proceedings of the Eighth International
Workshop on Large Scale Integration of Wind Power and on Transmission
Networks for Offshore Wind Farms.
27. Bustamante, R.C., Roman Messina, A., Gonzalez, M.R., and Guizar, G.C. (2018).
Assessment of frequency performance by grid integration in a large-scale power
system. Wind Energy 21: 1359–1371.
28. Acker, T. (2011). National Renewable Energy Laboratory, IEA Wind Task 24,
Integration of Wind and Hydropower Systems, Volume 1: Issues, Impacts, and
Economics of Wind and Hydropower Integration. Tech. Rep. NREL/TP-5000-50181
(December 2011).
29. You, S., Kou, G., Liu, Y. et al. (2017). Impact of high PV penetration on the inter-
area oscillations in the U.S. Eastern Interconnection. IEEE Access 5: 4361–4436.
30. Tuttelberg, K., Kilter, J., Wilson, D., and Uhlen, K. (2019). Estimation of power
system inertia from ambient wide area measurements. Proceedings of the 2019
IEEE Power & Energy Society General Meeting.
31. Begovic, M.M. and Messina, A.R. (2010). Wide area monitoring, protection and
control, special issue on wide-area monitoring, protection and control. IET
Generation, Transmission & Distribution: 1083–1085.
32. Miller, N.W., Shao, M., D’aquila, R. et al. Frequency response of the US Eastern
Interconnection under conditions of high wind and solar generation. Proceedings
of the 2015 Seventh Annual IEEE Green Technologies Conference.
33. Martinez, E. and Messina, A.R. (2011). Modal analysis of measured inter-area
oscillations in the Mexican Interconnected System: The July 31, 2008 event.
Proceedings of the 2011 IEEE Power and Energy Society General Meeting.
34. Muljadi, E., Gevorgian, V., Singh, M., and Santoso, S. (2012). Understanding
Inertial and frequency response of wind power plants. Proceedings of the IEEE
Symposium on Power Electronics and Machines in Wind Applications.
References 291
35. Golpira, H. and Messina, A.R. (2018). A center-of-gravity-based approach to

estimate slow power and frequency variations. IEEE Transaction on Power Systems
33 (1): 1026–1035.
37. Roman-Messina, A. (2020). Data Fusion and Data Mining for Power System
38. Rosipal, R. and Kramer, N. (2006). Overview and recent advances in partial least
squares. In: Subspace, Latent Structure and Feature Selection: Statistical and
Optimization Perspectives Workshop (SLSFS 2005), Revised Selected Papers (Lecture
Notes in Computer Science 3940), 34–51. Berlin: Springer-Verlag.
39. Irokawa, S., Andersen, L., Pritchard, D., and Buckley, N. (2008). A coordination
control between SVC and shunt capacitor for windfarm. CIGRE B4-307.
40. Bevrani, H., Watanabe, M., and Mitani, Y. (2014). Power System Monitoring and
Control. New York, NY: IEEE-Wiley Press.
41. Order 661. (2005). Interconnection for wind energy, issued by Federal Energy
Regulatory Commission (FERC) of United States (2 June 2005).
293
Index
a Markov clustering 159, 162, 163

acceleration torque 189, 203 temporal clustering 163, 241
adjacency matrix 158 COG formulation 28, 48, 189, 275
AGC. See Automatic generation control 2, COG representation 36, 41
72, 266 COG-based equivalent model 37, 41,
analog simulator 67 42, 46
angle control 2, 3, 8, 11, 14, 233 coherency identification techniques 28
aperiodic mode 90 COI frequency 33, 34, 35, 39, 75, 78,
area control error (ACE) 2 80, 275
automatic generation control (AGC) 2, combustion delay 56
72, 266 communication delay 5
automatic voltage regulator (AVR) 3, 200 companion matrix 225, 226, 228
auxiliary reactances 50 complex pole 91
computational burden 46, 50, 150
b concatenated data 177, 178, 179, 278
beat voltage 193 constant power load 26
best fitness value (BFV) 46 control loops 2, 3, 105, 113, 166, 233, 285
conventional droop characteristics 61
c correlation analysis 176
capital recovery factor 121 two-way correlation analysis 179, 181,
cascading events 42 277, 278
center of angles 42 critical modes 91, 92, 93, 136, 137, 165
center of gravity (COG) 28, 79, 186, 275 current references-based VSG control 107
center of inertia (COI) 28, 124, 188 current transition 192
centroid 152, 164, 242, 243
cluster d
cluster centroid 152 damping coefficient 12, 33, 76, 105
clustering damping ratio 90, 91, 92, 93, 94, 109, 136,
C-means 164, 173, 175, 243, 275 137, 234
direct clustering 164, 175, 154 damping torque 90, 203
fuzzy clustering 164, 175, 243 data-driven approaches 148, 153, 211, 213,
k-means 171, 173 235, 237, 239

294 Index
DC exciter field 56 averaged energy 230

DC field voltage 56 energy storage systems (ESSs) 25, 104
delta frequency detection 74, 77, 78, 84 equivalent inertia 28, 76, 115
diffusions maps (DMs) 177 equivalent system 12, 31, 38, 187, 188, 189,
multivie DMs 177, 179 190, 192, 275, 277
dimensional space 115, 158 ESS, see energy storage systems 25, 27, 47,
dimensionality reduction 154, 155, 156, 48, 104, 108, 113, 114, 120, 121, 124,
161, 181 242 130, 131, 132, 134, 136, 137, 138, 191,
dispatchable inertia 104, 113, 139 194, 195, 196, 197, 199, 200, 204
distance function 156 excitation system 6, 35, 128, 147, 231
distance matrix 156, 157, 159, 178 expansion 122, 123, 162, 223, 225
distributed energy resources (DERs) 25, 71, external system 27
145, 185, 212, 253, 254
distributed generation (DG) 25, 71, 103, f
185, 213, 222, 254, 277, 278 FACTS, see flexible alternating current
DM, see Diffusion map 156, 159, 160, 161, transmission system 6, 7, 103, 151
162, 164, 171, 173, 174, 175, 179 feedforward 56
doubly fed induction generators fictitious reactances 33, 40, 50, 51
(DFIGs) 108, 233, 264 fictitious tie-lines 29
droop characteristic 13, 61, 65, 84, 85, 125 fictitious transmission lines 28
dynamic equivalencing 25, 26, 27, 28, 61, first-swing oscillations 38
67, 239 flexible AC transmission systems
dynamic frequency indices 114, 122 (FACTS) 103, 152
dynamic mode decomposition flexible Inertia 186, 192, 203, 205, 206, 207
(DMD) 175, 215, 222, 223, 229, 272 flexible loads 27
DMD modes 228, 240 frequency bias 5, 77, 78
dynamic trajectories 156, 157, 162, frequency control
163, 164 primary 2, 114
secondary 2, 6, 7, 146
e tertiary 2, 7
economic modeling 119 frequency divider 36, 39
eigenvalue 41, 90, 91, 92, 93, 94, 116, 137, frequency dynamics propagation 189
159, 160, 163, 215, 217, 222, 223, 225, frequency nadir 14, 43, 50, 74, 75, 80, 82,
226, 227, 228, 229, 234, 235, 248, 249, 86, 88, 108, 109, 110, 113, 114, 121, 122,
259, 273, 281 123, 124, 126, 129, 189, 190, 191, 195,
eigenvectors 90, 116, 159, 161, 162, 163, 196, 197, 198, 199, 200, 277, 280, 281
171, 215, 226, 227, 228, 229, 230 frequency propagation 36, 37, 39, 43, 44,
electrical distances 149, 167, 168 45, 46
electrical vehicles (EVs) 108 frequency regulation 4, 5, 27, 108, 113,
electrical zones 168 259, 263, 288
electromagnetic transient (EMT) frequency response 5, 6, 12, 13, 33, 39, 46,
models 27 49, 65, 72, 73, 74, 75, 76, 77, 78, 84, 86,
electromechanical low-frequency 114, 123, 185, 188, 189, 191, 259, 260,
dynamics 27 261, 263, 274, 275, 277, 280, 281
emergency control 9, 10, 13 frequency stability 2, 5, 27, 46, 71, 73, 74,
energy criteria 80, 112, 113, 114, 129
Index 295
Frobenius matrix 216 inter-unit synchronizing oscillations 28

fuel command 55, 56 inverter-Based DG 58, 61, 103, 108,
fuel conversion ratio 56 109, 245
full model 27 inverter-based distributed generators 1
islanded mode 47, 53, 54, 59, 61, 67, 188
g
genetic algorithm (GA) 33, 125 j
Genset 54, 55, 57, 59, 66, 67, 81, 82, 86 Jacobian matrix 95, 96, 148, 149
global frequency 31, 33, 67 Jacobian sensitivities 95
global Positioning System (GPS) 4
goal programming 48, 49 k
governor response 123, 261, 262, 263, kernel
264, 279 Gaussian kernel 157, 159
GPS, see Global Positioning System 9 kernel bandwidth 158
graph kernel methods 157
undirected graph 157, 163 Koopman mode 211, 215, 223, 250,
graph clustering 153, 157 270, 281
grid code 85 Koopman mode decomposition 211,
grid-connected mode 26, 61, 63, 67, 215, 250
119, 188 Koopman operator 222, 223
grid-forming mode 53 Krylov sequence 224
grid-supporting mode 54 Kumaresan-Tufts 222
h l
Hadamard product 163 LCC, see life cycle cost 119, 120, 134
Hankel matrix 221, 222 least-squares 218, 222, 223
Hankel-SVD 221 life cycle cost (LCC) 119
hierarchical levels of voltage control 7 linear prediction model (LPM) 217
high voltage direct current (HVDC) 6, 27, load damping 33
103, 104, 261 load frequency control (LFC) 2
high-frequency switches 58 load shedding 9, 13, 44, 73, 104, 185,
197, 246
i local frequency 27, 28, 33, 35, 67, 185, 275
ideal voltage source 61 local loads 61
IEEE 50-machine test system 35, 44, 85, 89 long-term power frequency transients 186
impulse response 116 low-frequency equivalent models 27
inertia emulation 105, 114, 116, 124, low-frequency models 26, 27
185, 260
inertia-based control 104, 188, 189, 191, m
197, 199, 200, 201, 202, 203, 204 mapping 114, 160, 161, 162, 173, 224
inertia-weighted perturbation 187 Markov
inflation operator 163 Markov clustering algorithm
inner voltage loop 106, 107 (MCL) 162, 163
instantaneous frequency 82, 122 Markov matrix 158, 162, 163, 172
inter-area modes 7, 41, 91 matrix of attenuation 150
internal combustion (IC) 54 matrix pencil 221
296 Index
maximum penetration level 71, 80, 85, 86, oscillation damping 2, 3, 4, 5, 7, 8, 12,
88, 89, 93, 94, 95, 99 105, 108
MCL, see Markov clustering 159, 162, oscillation-free behavior 61
163, 176
measurement- based approach 28, 169 p
Mei-Sheila algorithm 159 parameter estimation 1
MG, see microgrid 1, 4, 5, 6, 11, 14, 25, 26, parametric uncertainty 5, 135
27, 28, 47, 53, 54, 61, 63, 64, 65, 67, 71, partial least squares
72, 73, 76, 78, 79, 80, 82, 84, 85, 88, 89, partial least squares correlation 179
90, 91, 92, 185, 188, 189, 195, 196, 197, partial least squares regression 181
198, 207 participation factor 11, 137, 138, 215
MGs-penetrated grid 26 percentage of variation (PCTVAR) 278
microgrids (MG) 1, 25, 71, 103, 145 permanent magnet synchronous generators
mid-term dynamics 39 (PMSGs) 108
MIMO, see multi-input multi-output 4 permissible frequency deviation 83, 84
minimization problem 47, 48, 137, 202 phase-locked loop (PLL) 191
minimum inertia 82, 84, 85, 88, 122, phasor measurement units (PMUs) 2,
123, 125 145, 262
modal analysis 5, 36, 41, 61, 126, 213, 222, phasor models 27
230, 254, 285 pilot bus 149, 151, 168, 175
modal decomposition 213, 215, 223, 228 pilot node 149, 150, 151, 152, 166, 168, 178
mode shape PMU, see phasor measurement units 3, 4,
mode shape identification 241 5, 6, 7, 8, 9, 12, 13, 145, 150, 153,
speed-based mode shape 234, 253, 273 154, 155
model point C criterion 82
deterministic 25, 47, 48, 50 point of common coupling 64
stochastic 25, 50, 207 point of interconnection with the system
modular multilevel matrix converter (POIS) 263
(M3C) 108 post-disturbance frequency fluctuation 73
modulation index 59 post-fault stable equilibrium points 46
multi-input multi-output (MIMO) 4 power angle-based stability index 129
multi-MGs (MMGs) 47, 82, 188 balance of plant 119
MUltiple Signal Classification power conversion system (PCS) 119
(MUSIC) 115, 221 power conversion system 119
multiterminal HVDC systems 108 Power references-based VSG control 108
MUSIC, see MUltiple Signal power system modeling 25, 28
classification 115, 116, 221 power system simulator 63
power system stability 1, 4, 14, 26, 53, 71,
n 72, 74, 77, 99, 104, 105, 207
New York New England (NYNE) test power system stabilizer (PSS) 200, 264
system 35, 78, 131, 194 power-electronic interfaced sources 27
primary protection 44
o primary reserve 72
online estimation 13 principal component (PC) 277
optimal placement 103, 104, 113, 114, 121, probability density function (PDF) 49, 135
125, 128, 136 Prony analysis
Index 297
Multisignal Prony analysis 215, 237 spinning reserve 263, 264, 277
Prony polynomial 216, 217, 220 state matrix 90
standard Prony analysis 215 state of charge (SOC) 115, 116, 121, 191
proportional-integral (PI) 55, 284 Static load bank (SLB) 57
pseudo-spectrum 116, 117 static switch 193, 194
pulse function 64 static VAR compensator (SVC) 6, 146, 263
stochastic approaches 46
q stochastic transition matrix 162
quasi-steady-state 132 study zone 27
supervisory control and data acquisition
r (SCADA) 154
rate of change of frequency (RoCoF) 13, swing equation 11, 13, 33, 62, 63, 64, 73,
43, 73, 104, 121, 188, 277 104, 105, 107, 114, 122, 187
reactive control zones 150 synchronism 90
reactive power dispatch 260, 283 synchronizing torque 90
reconstruction error 251 synchronous machine 3, 27, 35, 36, 54, 55,
reduced model 41 69, 90, 104, 105, 128, 264, 278
reliability 11, 26, 46, 71, 72, 74, 147, 206 synchronous-based DG 54, 196
renewable energy sources (RESs) 1, 105, synchronverter 104
146, 212, 259 system inertia 25, 29, 32, 35, 36, 37, 72, 73,
resiliency 48 86, 90, 92, 112, 113, 124, 136, 194, 196,
resonance 12, 193 260, 261, 274, 277
right half plane (RHP) 91 system instability 14, 38, 42, 112, 266
RMSE, see Root Mean Square Error 82 system topology 38, 151
RoCoF, see rate of change of frequency 13,
43, 50, 73, 74, 76, 77, 80, 85, 86, 104, 105, t
108, 109, 110, 113, 114, 121, 122, 123, TCC, see total capital cost 119, 120
126, 129, 188, 189, 191, 200, 277, 281 temporal structure 212, 227
root mean square error (RMSE) 82 three-phase voltage source 59
rotor oscillations 90 tie-line power exchanges 29
rotor-angle stability 27 total capital cost (TCC) 119
transient stability 51, 72, 86, 114, 129, 132,
s 146, 200, 211, 237, 246, 260, 266,
286, 288
sensitivity analysis 6, 90, 92, 134, 135
transient stability assessment (TSA) 86
short-term behavior 42
transition probability matrix 158
similarity matrix 156, 157
transmission system operators
singular value decomposition
(TSOs) 73, 104
(SVD) 179, 215
trial-and-error 72
slowest modes 36
TSOs, see transmission system
small-signal stability assessment
operators 73
(SSSA) 86, 90
turbine dynamics 28
solar photovoltaic 6, 71, 146, 211, 259
turbine-governor model 35
spatial patterns 161, 180, 254
spatial structure 229, 242
spatiotemporal 156, 162, 164, 212, 254 u
spectral graph theory 157 uncertainty analysis 50, 135
298 Index
v volt–ampere reactive (VAR) 145, 263

Vandermonde matrix 216, 229 VSG, see virtual synchronous generator 4,
VAR, see Volt–ampere reactive 145, 263 104, 105, 106, 107, 108, 109, 110, 112,
variable generations 45, 46, 47 114, 139, 185
vehicle-to-grid (V2G) 108
virtual governor 107 w
virtual inertia 3, 5, 10, 112, 113, 114, 116, washout filter 36, 43
120, 123, 124, 126, 128, 129, 132, 136, wide-area control (WAC) 3
138, 188, 191 wide-area measurement 7, 8, 10, 153
virtual Inertia modeling 114 wide-area monitoring and
virtual power flows 29 control (WAMC) 7
virtual reactance 31, 38 wide-area voltage control scheme 145, 150
virtual synchronous generator (VSG) 4, wide-area voltage monitoring
104, 185 Wide-area monitoring systems 153
voltage control 2, 3, 6, 7, 12, 96, 105, 145, wind coordination 277, 288
146, 147, 149, 150, 151, 153, 154, 155, wind energy
156, 166, 168, 171, 172, 174, 181, 261, generic wind energy model 233, 245
263, 283, 284, 285, 288 wind farms (WFs)
voltage control areas 145, 147, 148, 149, wind farms active power 267, 268
150, 153, 154, 168, 171, 172, 174, 181 wind farms reactive power 270, 271, 284
voltage references-based VSG control 107 wind integration 111, 259, 263, 274
voltage regulator 3, 56, 200
voltage sensitivities 95, 148
voltage sensitivity factors 95 z
voltage source 27, 59, 61, 108, 191, 233 Zones
voltage source converters (VSCs) 108 critical zones 147
voltage stability 4, 6, 7, 12, 145, 146, 153, voltage control zones 147, 148, 153, 156,
260, 285 166, 168, 174
WILEY END USER LICENSE AGREEMENT
Go to www.wiley.com/go/eula to access Wiley’s ebook EULA.

Psoc Res IV Year

Uploaded by

Copyright:

Available Formats

Psoc Res IV Year

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Psoc Res IV Year

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What are some of the main topics covered in the book?

What are some of the main topics covered in the book?

Renewable Integrated

Power System Stability

IEEE Press Editorial Board

Jón Atli Benediktsson David Alan Grier Elya B. Joffe

Library of Congress Cataloging-in-Publication Data

Cover Design: Wiley

Set in 9.5/12.5pt STIXTwoText by SPi Global, Pondicherry, India

2 MG Penetrated Power Grid Modeling 25

2.3.1.5 Simulation Results 35

3 Stability Assessment of Power Grids with High Microgrid

3.3.4 Minimum Inertia Related to Delta Frequency Detection 84

4 Advanced Virtual Inertia Control and Optimal Placement 103

4.3.1.6 Sensitivity Analysis 134

5 Wide-Area Voltage Monitoring in High-Renewable Integrated Power

6 Advanced Control Synthesis 185

7 Small-Signal and Transient Stability Assessment Using Data-Driven

7.4 Large-Scale System Study 244

8 Solar and Wind Integration Case Studies 259

Increased needs for electrical energy as well as environmental concerns besides

control methods in emulating virtual inertia are clarified. Afterwards, according to

Hêmin Golpîra, University of Kurdistan

X,Y data matrices

Pin injected power of ESS to the host grid

ξ+ deviations from the target value in positive direction

List of Abbreviations and Acronyms

1.1 Power System Stability and Control

• Frequency control: Since the frequency generated in an electric network is

• Voltage control: The generators are usually operated at a constant voltage by

• Virtual synchronous generator: Additional flexibility may be required from var-

value, analogous to the inertia of a conventional generator. This setup, which is

1.2 Current State of Power System Stability and Control

Figure 1.1 Conceptual structure of a virtual synchronous generator.

1.2.1 Frequency Control

1.2.2 Voltage Control

As mentioned above, several PMU-based voltage control methodologies have

1.2.3 Oscillation Damping

1.3 Data-Driven Wide-Area Power System Monitoring

Power System Data

Pre-filtering and De-noising

Emergency Control and Automatic

Figure 1.3 PMU-based wide-area measurement system and control.

algorithms. Otherwise, the estimated parameters such as scheduling parameters

1.4 Dynamics Modeling and Parameters Estimation

A complete understanding of reliability considerations via effective modeling/

1.4.1 Modeling of Frequency, Voltage, and Angle Controls

of system model for frequency control. Considering the frequency response

1.4.2 Parameters Estimation

mostly based on rolling averaging windows, can be employed to prepare the

synchrophasor measurements. IEEE Transactions on Power Systems 26 (2):

overview of the consolidated hierarchical system. IEEE Transactions on Power

MG Penetrated Power Grid Modeling

Power grids worldwide have experienced a significant transformation, which has

The increased demand for electrical power, environmental concerns, pressing

2.2 Basic Concepts

2.2.1 Dynamic Equivalencing

low-frequency models are formally utilized in simulating rotor-angle stability of