Aggregation of Component-Based Grid-Feeding

DER and Load Models for Simulation of Microgrid

Islanding Transients
M.H. Roos∗ , P.H. Nguyen∗‡ , J. Morren∗† , J.G. Slootweg∗†
∗ Department
of Electrical Engineering
Eindhoven University of Technology, Eindhoven, the Netherlands
Email: [email protected]
† Asset Management,

Enexis Netbeheer, ’s-Hertogenbosch, the Netherlands

‡ Sustainable Energy Systems Group,

Luxembourg Institute of Science and Technology, Belvaux, Luxemburg

Abstract—The reliability of distribution networks can be im- performed periodically due to the variability of generation,
proved by islanding of part of the network when a short-circuit network structure and load in microgrids.
fault occurs. The voltage and frequency stability during transient To analyze the stability with time-domain simulations, ac-
can be evaluated with time-domain simulations using component-
based DER and load models. However, simulations with this curate DER and load models are required which represent
type of models have a relatively high computational burden the nonlinear behavior during transients. The dynamics of
which constrains periodical stability analysis of microgrids. different subsystems are strongly coupled in microgrids [6]
This paper proposes an aggregation methodology which signifi- requiring comprehensive DER and load models. Component-
cantly decreases the computation time of fault-initiated islanding based models can model the behavior of DERs and load
simulations with component-based models, while maintaining
the nonlinear dynamics. The methodology is first validated by devices with a very high level of detail, accurately describing
directly comparing the original and equivalent models during the electrical, mechanical, controllers and protection systems
large voltage and frequency transients. Secondly, a fault-initiated [7], [8]. Although component-based models can describe the
islanding case study of a residential distribution network is behavior of the DERs and loads with high accuracy, these
performed. For both validations, a sensitivity analysis of model models increase the required computation time which con-
parameters and voltage variations is performed. The results
indicate the methodology preserves the dynamics of DERs and strains the ability to periodically perform FII stability analysis.
loads, while significantly reducing the computational time. The computation time can be reduced by aggregation of
the DER and load models. Three main methodologies for
Index Terms—Dynamic equivalent model, microgrids, island-
ing, distributed generation, power system transients. load aggregation have been described in the literature: modal
analysis, coherency based aggregation and system identifica-
tion [8]. System identification and modal analysis techniques
I. I NTRODUCTION are not suitable for detailed component-based models. System
identification techniques are based on measurements to gener-
The increasing societal dependence on electrical energy ate models as opposed to device components. Modal analysis
and the occurrence of natural disasters fuels the need for requires linear models which neglects important nonlinear
more reliable and resilient power (distribution) networks [1]. dynamics.
The reliability of distribution networks can be increased by During coherency based aggregation, models with similar
fault-initiated islanding (FII), which forms islanded microgrids dynamics are replaced by an equivalent model. Coherency
in case of a contingency such as a short-circuit fault [2], based methods have been mostly used for aggregation of
[3]. During the FII transient, the voltage and frequency can synchronous and induction machines [9]–[11]. Aggregation of
significantly deviate from their nominal values due to inter- power electronic load models has been performed by [12],
actions between the phase-locked loop of distributed energy [13], however until now only the steady-state behavior of these
resources (DERs) and the load [4], and the lack of voltage and aggregated models have been validated.
frequency control. Therefore the stability of the microgrids In this paper, a component-based DER and load model
should be evaluated to enable FII, especially during and right aggregation methodology is proposed which reduces the com-
after the FII transient [5]. The stability analysis should be putation time of simulations of FII microgrids significantly,
while preserving the dynamic behavior for accurate stability
This work has received funding from the European Union’s Horizon 2020 analysis. The methodology is validated by direct comparison
research and innovation program under grant agreement N°773717. between the original and aggregated models, and a FII case

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
study in a modified version of the CIGRE European LV
residential distribution benchmark network.
The contributions of this paper are: (a)

1) The aggregation methodology for component-based DER

and load models which decreases computation time of
(b)
FII microgrid simulations while maintaining accurate
dynamics.
2) Verification of the accuracy of component-based load and P P
W (c) W
D1 D2 D3 Deq
DER device equivalencing under parameter variation and A A
voltage differences.
3) Verification of the accuracy and reduction of computa-
tional burden of the methodology with a FII case study Boundary Device Interior Empty

under different network impedance, and DER and load Fig. 1: Overview of the aggregation methodology. (a): Kron reduction
models parameter variations. of interior and empty nodes of original network. (b): Creation of com-
mon node with power weighted average voltage. (c): Equivalencing of
In the next section the aggregation methodology is de- DER and load devices of same type. PWA: Power weighted average,
scribed, the numerical validation of the methodology is dis- Deq: dynamic equivalent of devices D1, D2 and D3.
cussed in section III, the results are shown in section IV and
discussed in section V, and finally in section VI conclusions
are given. B. Network Reduction
1) Kron reduction: Kron reduction is a network reduction
II. AGGREGATION M ETHODOLOGY methodology which reduces the number of nodes and constant
The methodology consists of several steps to transform impedance loads in the network. The boundary and device
a network consisting of nodes and branches into a reduced nodes are critical for voltage stability analysis as they contain
network with a common node to which the grid-feeding all detailed, nonlinear models. Therefore, these nodes are
DERs and load devices are connected. An overview of the preserved and Kron reduction is utilized to eliminate interior
aggregation methodology is shown in Fig.1. First, the network and empty nodes. Since the load connected to boundary and
nodes are classified based on the connected DERs and load de- detailed load nodes does not impact the Kron equivalent net-
vices, which are classified based on the different device types. work impedances, the DERs and detailed loads are neglected
Second, a Kron reduction is applied to eliminate the interior during Kron network reduction.
and empty nodes while maintaining the boundary and device To perform Kron reduction, the network admittance matrix
nodes. Third, a common node is created to which all grid- YN et is build as described by equation 1. Where N is a set
feeding DERs and detailed load models are connected instead of boundary and detailed load nodes, M is a set of interior
of the network nodes. The power weighted average (PWA) of and empty nodes, K is a set of detailed load nodes, P is a
the devices is used to create equivalent current sources at the set of device types and Q is a set of devices of each type.
network nodes and an equivalent voltage source at the common Matrices YN N , YN M and YM M contain the node admittance
node. Final, dynamic equivalent models are created for each and admittance between nodes of set N , N and M , and M
type of device based on the device classification. respectively.  
Y YN M
YN et = NT N (1)
A. Classification YN M YM M
Network nodes are classified into four categories: boundary, The Kron reduced admittance matrix is determined by using
device, interior and empty. Nodes to which the point of equation 2 [14].
common coupling (PCC) and/or grid-supporting DERs are
connected are classified as boundary nodes. Nodes to which −1
Ykron = YN N − YN M YM T
M YN M (2)
grid-feeding DERs and/or detailed load models are connected
are device nodes. Nodes to which only constant impedance 2) Creation of common node: Before equivalent models
loads are connected are interior nodes, and nodes without can be developed for each type of device, the models should
DERs or load devices are empty nodes. be connected to the same node. Therefore, a common node
Devices are classified into different categories based on the is created to which all grid-feeding DER and detailed load
type of device to cluster devices with similar structure to allow models are connected. The common node is fed by a voltage
the development of accurate equivalent models. Additionally, source which provides the PWA of the voltage of the device
classification can be performed based on electrical, mechani- nodes as shown in Fig. 1. The device nodes are replaced by
cal, controller, and protection system parameters. A larger set empty nodes with current sources which absorb or inject the
of classes can increase accuracy, however this results in more PWA of the current of the device models connected to the
equivalent models, increasing the computation time. common node.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 Vdq
Pref
abc/
Rin Qref PC Iref CC Idq
dq θ PLL
PWM
Cdc Reqi R0 Lb Iabc
L1 L2 Vabc

Em Cdc Rd
Vdc
Fig. 2: Diagram of compact fluorescent lighting load device. PWM VC C1

Vdc Fig. 4: Diagram of three phase photovoltaic DER.

Φ
DTC T SC ωref
Ldc SVM
Rdc ω
these devices should be aggregated into multiple models as
discussed in section II-A.
Cdc M 1) Electrical parameters: The electrical parameters are de-
termined according to the parallel impedance. Resistance Req ,
impedance Leq and capacitance Ceq values are determined
with equation 5 where D is the number of parallel devices.
Fig. 3: Diagram of variable frequency drive load device.
D
1 1 X
Req = PD , Leq = PD , Ceq = Ci (5)
The PWA of the voltage (Veq ) is determined by equation 1 1
k i=1 Ri i=1 Li i=1
3, while the PWA of the current injected at node k (Ieq ) is
determined by equation 4. Where Vk is the voltage at node 2) Mechanical load parameters: The equivalent inertia Jeq ,
k, Sk is the total apparent power absorbed by the devices at friction Feq and torque Teq are equal to the sum of the inertia,
node k, Sp,k is the total apparent power absorbed by devices friction and torque of the devices as described by equation 6.
of type p connected to node k. D D D
X X X
Jeq = Ji , Feq = Fi , Teq = Ti (6)
K
X |Sk | i=1 i=1 i=1
Veq = Vk PK (3)
k=1 k=1 |Sk | 3) Controller parameters: DER and load devices are gen-
erally controlled by PI controllers. The equivalent proportional
P Q (kP ) and integral (kI ) controller parameters are determined by
k
X |Sp,k | X
Ieq = PK Ip,q (4) the PWA values as described by equation 7.
p=1 k=1 |Sp,k | q=1
D D
X |Si | |Si | X
C. Device equivalencing kP,eq = kP,i PD , kI,eq =
kI,i PD
i=1 d=1 |Sd | i=1 d=1 |Sd |
As shown in Figs. 2, 3 and 4, component-based models (7)
of DER and load devices consist of several components. By 4) Protection parameters: Mainly three types of in-
scaling the parameter values of the components, the models ternal protection systems are implemented in DER and
can be scaled to different power ratings with equal dynamics. load devices: overcurrent, under-/overvoltage and under-
Since two parallel devices with identical parameters can be /overfrequency. Overcurrent protection usually limits the peak
represented as a single device with two times the power value of the current and therefore only requires a maximum
rating, this concept is fundamental for the device equivalencing setting. The setting of the equivalent model is equal to the
methodology. For example, a constant impedance load can be sum of the overcurrent protection settings of the devices.
scaled to different power ratings by changing the impedance. The AC under-/overvoltage and under-/overfrequency pro-
Likewise, two parallel constant impedance loads can be repre- tection of DERs are dictated by the anti-islanding and fault-
sented by a single constant impedance load with the equivalent ride-through requirements in the local grid codes. Although
impedance. this is often equal for all DERs, different requirements may
This paper considers electrical, mechanical, controller and be specified for different DERs (e.g. power rating). This can
protection system parameters for device equivalencing. The be solved by classifying DERs in different types.
remainder of this section describes how these parameters The DC undervoltage protection of load devices causes
are determined. Note that as the models in this paper are temporary or permanent load drop-off. The setting of DC
nonlinear, the accuracy of the equivalent model depends on undervoltage protection commonly consists of a minimal
how similar the parameters of the original devices are. When threshold value, the setting of the equivalent model is therefore
two devices have very different parameters e.g. mechanical equal to the PWA value of the DC undervoltage protection
loads with similar power ratings and a large inertia mismatch, settings of the devices.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 TABLE II: Sensitivity analysis parameter variation in pu.
PWA Device 0.2pu 0.6pu 1pu 1.4pu
I1 I2 I3 I4 Ieq
1 1.1 1.3 1.5 1.7
V1 D1 V2 D2 V3 D3 V4 D4 Veq Deq 2 1.05 1.15 1.25 1.35
3 0.95 0.85 0.75 0.65
Fig. 5: Setup for validation of device equivalencing. D1 , D2 , D3 , D4 : 4 0.9 0.7 0.5 0.3
original models, Deq : equivalent model.
TABLE III: Sensitivity analysis voltage levels during steady
TABLE I: Model parameters for each device for the given power state/voltage transient in pu.
rating. Device 0.1pu 0.2pu 0.3pu 0.4pu
Par. Value Par. Value Par. Value 1 1.05/0.15 1.1/0.2 1.2/0.3 1.3/0.4
LEDL 2.4W Cdc 840µF R0 61.9mΩ 2 1.025/0.125 1.05/0.15 1.125/0.225 1.2/0.3
Rin 300Ω Rs 0.172Ω Lb 21µH 3 0.975/0.075 0.95/0.05 0.975/0.075 1/0.1
Rin2 235kΩ Rr 0.176Ω Cdc 800uF 4 0.95/0.05 0.9/0 0.9/0 0.9/0
Cin 517nF Ls 793µH L1 10.8mH
Cdc 47µF Lr 793µH L2 477µH
VFD 11kW Lm 51.4mH C1 10uF scaled CFL models have a power rating of 2.7kW, 5.4kW,
Rdc 0.5Ω TPPV 8kW Rd 1.44Ω
Ldc 3mH Em 48V 10.8kW and 21.6kW divided over three phases. The scaled
VFD models have a power rating of 11kW, 22kW, 33kW and
44kW. The scaled TPPV models have a power rating of 4kW,
III. N UMERICAL VALIDATION 8kW, 16kW and 32kW. Thus, the equivalent CFL, VFD and
TPPV models have a power rating of 40.5kW, 110kW and
The methodology described in the last section is validated
60kW respectively.
by comparing the original and equivalent models during large
voltage and frequency transients, and by comparing the voltage Z T
1
waveforms during FII in a case study. e(t) = |Ieq (t) − Ior (t)|dt (8)
Iˆn T t=0
A. Devices The NMAE is determined over a period of 500ms during
Three types of devices are considered: compact fluorescent which a voltage or frequency step change takes place. Voltage
lighting (CFL) load, variable frequency drive (VFD) load and transients are performed by step changing the voltage from
three phase photovoltaic (TPPV, 8kHz switching frequency) 1.0pu to 0.1pu and back to 1.0pu after 200ms. Frequency
DER as shown in Figs. 2, 3 and 4. The electrical parameter transients are performed by a step change from 50Hz to
values of the models are shown in table I. The CFL has a 20Hz and back to 50Hz over the same period. Pre- and post-
constant current equivalent resistance Reqi with a drop-off disturbance performance are thus analyzed over 150ms before
voltage of 35V. The VFD has a direct torque controller with and after the transient. This mimics the temporary voltage
controller gains kpt = 0.005, kit = 100 and a speed controller and frequency deviations which are expected during the FII
with controller gains kps = 10, kis = 4 to drive an 11kW transient. The transients are initiated at a phase angle of 90◦ ,
symmetrical induction machine modeled in the dq reference where the voltage waveform of phase a is at its peak value.
frame as described by [15]. The VFD drop-off voltage is As devices in a network may have different parameters and
420V with a hysteresis of 30V. The TPPV has a synchronous the nodes to which the devices are connected may have voltage
reference frame phase-locked loop (PLL) with controller gains differences, a sensitivity analysis is performed by varying the
kppll = 30.87, kipll = 1029 and a current controller in the parameters and voltage difference between the devices. The
dq reference frame with gains kppi = 1.63, kipi = 204. difference between the electrical, mechanical and controller
The passive anti-islanding protection trips when the voltage parameters of the devices are varied between 0.3pu and 1.7pu
V ≤0.7pu for 10ms, V ≥1.05pu or V ≤0.8pu for 1.95s, or according to table II (additional to scaling due to the different
if the frequency f ≥75Hz or f ≤25Hz for 1ms or f ≥51Hz power ratings). The voltage differences are varied between
or f ≤48Hz for 2s. The original are scaled to higher power 0.1pu and 0.4pu according to table III. In this scenario, the
ratings with equal dynamics as described in section II-C. equivalent model is connected to the PWA of the voltages of
devices 1, 2, 3 and 4.
B. Device equivalencing
The device equivalencing methodology is validated by de- C. Case study
termining the normalized mean absolute error (NMAE) of the A modified version of the CIGRE European LV residential
absorbed/injected current in the setup shown in Fig. 5. The distribution benchmark network is used to validate the load
NMAE is determined for CFL, VFD and TPPV by comparing aggregation methodology as shown in Fig. 6. The network con-
the sum of the current of four scaled models to the current tains two three-phase battery energy storage systems (BESS)
of the equivalent model by using equation 8. Where Iˆn is the which are modeled as TPPV normally operating in grid-
nominal peak current, Ieq (t) is the current of the equivalent feeding operation and switch to grid-supporting operation
model and Ior (t) = I1 (t) + I2 (t) + I3 (t) + I4 (t) is the sum when the anti-islanding detection trips. The grid-feeding and
of the current of the original (scaled) models repectively. The grid-supporting controllers are modeled as described by [16].

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 TABLE V: Network cable lengths.
4
1-7 2-10 3-7 4-8 5-9 6-10 7-8 8-9 9-10
6 Length (m) 70 70 30 135 30 30 35 70 105

PCC
1 7 8 9 10 2 0.06

Normalized mean absolute error

External grid
B1 0.04
3 5
B2
0.02
Fig. 6: Modified CIGRE European LV residential distribution bench-
mark network. B1, B2: battery energy storage systems.
0

TABLE IV: Power rating and parameter multipliers of grid-feeding 0 0.2 0.4 0.6 0.8 1 1.2 1.4
DER and load devices in the network. CZ: constant impedance load, Parameter variation (p.u.)
PV: grid-feeding photovoltaic DER, Mult.: parameter multiplier of :CFL :VFD :TPPV
devices connected to node (only used in scenario 2). Fig. 7: Normalized mean absolute error of the current of the equiva-
Node CFL (kW) VFD (kW) CZ (kW) PV (kW) Mult. lent CFL, VFD and TPPV models with different parameter variations
1 2.7 22 1.25 - 0.7 in case of a voltage transient.
2 - 44 1.25 4 0.8
3 5.4 - 1.25 32 0.9
4 10.8 - 2.5 - 1.1 IV. R ESULTS
5 - 33 2.5 8 1.2
6 21.6 - 1.25 16 1.3 A. Device equivalencing
7 - - 5 - -
8 - - - - - As discussed before, large differences between the parame-
9 - - 2.5 - - ters of devices and the voltages of different network nodes may
10 - - - - -
impact the accuracy of the aggregated model. The parameter
and node voltage variations are simulated as described in
section II-C. The NMAE of the equivalent CFL, VFD and
A three phase fault is simulated in the external grid at t = 0.1s
TPPV models with different parameter variations in case of
by reducing the voltage of the external grid to 0.1pu. At
a voltage and frequency transients is shown in Figs. 7 and
t = 0.15s the network is islanded by opening the point of
8 respectively. The NMAE is below 0.1pu in almost all cases
common coupling. After islanding, the islanding detection
which indicates that the equivalent models accurately represent
systems of the BESS will trip and switch the BESS from grid-
the dynamics of the original models and that a larger variation
feeding to grid-supporting operation.
in parameters generally leads to an increasing error. The
The network nodes are classified in the following way: impact of parameter variations is most significant in case of
nodes 1, 2 and 5 are boundary nodes, nodes 3, 4 and 6 are VFDs, however even with a variation of 1.4 p.u. the maximum
device nodes, nodes 7 and 9 are interior nodes, and nodes 8 and NMAE is smaller than 0.05. The NMAE does not increase
10 are empty nodes. The grid-feeding DER and load devices linearly with the parameter variation due to nonlinear effects.
at each node are shown in table IV. The constant impedance For example, when the parameter variation is 0.6pu or larger
loads are modeled as parallel RL loads with a power factor the smallest VFD drops off during a voltage transient, while
of 0.95. The network consist of 240mm2 AL cables modeled the largest VFDs and the equivalent model remain operational.
as series RL branches with a resistance and reactance of The NMAE of the current absorbed injected by the equiva-
0.15Ω/km and 0.071Ω/km respectively. The parasitic cable lent CFL, VFD and TPPV models with different node voltages
capacitance is neglected due to the short length. The network in case of a voltage and frequency transients is shown in Figs.
cables lengths are shown in table V. 9 and 10 respectively. The NMAE increases when the node
Three scenarios are analyzed to evaluate the performance of voltage variation increases in three cases, while the NMAE re-
the aggregation methodology in different conditions. Scenario mains approximatly constant in three other cases. Node voltage
1 is the baseline where the parameters of the grid-feeding variation mostly impacts the accuracy of the equivalent TPPV
DER and load models of the same type are scaled from the due to PLL tracking issues during the frequency transients. The
models described in III-A based on power rating. In scenario equivalent model of the TPPV is therefore considered accurate
2, the electrical (impedance), mechanical (inertia and friction) during frequency transients when the voltage variation is
and controller (controller gains) parameters of the grid-feeding smaller than 0.3pu.
DER and load models are multiplied by the corresponding
multipliers shown in table IV. This scenario emulates a case in B. Case Study
which DERs and loads have different parameters. In scenario The voltage waveforms at node 2 in the original and
3, the network impedance is five times larger which emulates aggregated networks are shown in Figs. 11, 12 and 13 for
a network with larger voltage differences between DERs and scenario 1, 2 and 3 respectively. The dynamics during the
loads. FII transient are similar in all scenarios. At t = 0.1s the

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 Normalized mean absolute error
Normalized mean absolute error 0.06

0.1
0.04

0.05
0.02

0
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
0 0.2 0.4 0.6 0.8 1 1.2 1.4 Voltage variation (p.u.)
Parameter variation (p.u.) :CFL :VFD :TPPV
:CFL :VFD :TPPV
Fig. 10: Normalized mean absolute error of the current of the
Fig. 8: Normalized mean absolute error of the current of the equiva- equivalent CFL, VFD and TPPV models with different voltage
lent CFL, VFD and TPPV models with different parameter variations variations in case of a frequency transient.
in case of a frequency transient.
400
Normalized mean absolute error

Original (V)
0.015
200 Aggregated (V)

0.01 0

-200
0.005

-400
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 Time (s)
Voltage variation (p.u.) Fig. 11: Comparison of the voltage at node 2 in the original network
:CFL :VFD :TPPV and the aggregated network in scenario 1.
Fig. 9: Normalized mean absolute error of the current of the equiv-
alent CFL, VFD and TPPV models with different voltage variations
in case of a voltage transient.
developed in section II-C increases when the devices of
the same classification have a larger variation in parameters
fault causes the voltage to decrease significantly. When the and when the devices are connected to nodes with different
network is islanded at t = 0.15s the voltage increases while voltages. It should be noted that the device equivalencing
the frequency slightly decreases. Islanding is detected by the negates possible interactions between devices of the same type.
BESS around t = 0.21s in all scenarios, recovering the voltage However the impact on the simulation results is limited since
and frequency to near nominal values. The voltage waveforms these interactions mainly occur between droop controlled grid-
remain partly distorted after control-mode switching due to the supporting DERs [6], which are preserved in the proposed
nonsinusoidal current waveforms of power electronic loads. methodology. Additionally, a small-signal stability analysis
As shown in Figs. 11, 12 and 13, the original and aggregated approach is often preferred to analyze the high frequency
voltage waveforms are very similar during and after the FII (voltage and current controller) interactions in microgrids.
transient. Slight differences are visible during islanding before The reduction in computational burden enables periodical
the BESS switch to grid-feeding control. The results from (bi-hourly) analysis of the voltage and frequency stability
scenario 2 show that parameter differences between models analysis during and right after FII, even on a laptop computer.
impact the accuracy of the aggregated model. The increased Detailed component-based DER and load models are re-
network impedance has no significant negative effect on the quired to develop the aggregated model. This information may
accuracy of the aggregated model. not be available to the microgrid operator, which requires the
The simulations are performed in Matlab/Simulink using load to be identified. A suitable methodology to identify load
the ode23tb solver with a maximum step size of 1µs on a devices based on aggregated measurements is proposed by
PC with an Intel i7-4700MQ processor and 8GB of RAM. [17]. Alternatively, smart meters can be utilized to identify
The computation time of the three scenarios is averaged. The load devices as described by [18].
average computation time to simulate a single scenario is The accuracy of the aggregated model is mainly impacted by
36837 seconds using the original model and 4909 seconds parameter variations between different DER and load models
using the aggregated model, resulting in a reduction of 87%. of the same type. This is evident as models with very different
dynamics are difficult to represent with an equivalent model.
V. D ISCUSSION This can be solved by increasing the number of grid-feeding
The results indicate that the proposed methodology accu- DER and load classifications, however this also increases
rately aggregates component-based grid-feeding DER and load the computation time as this results in additional equivalent
models which greatly reduces the computational burden of models. Optimization of the number of DER and load type
FII simulations. The error of the dynamic equivalent models classifications should therefore be treated by future research.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020
 400 [2] Z. Wang and J. Wang, “Self-healing Resilient Distribution Systems based
Original (V)
Aggregated (V) on Sectionlization into Microgrids,” IEEE Transactions Power Systems,
200 vol. 30, no. 6, pp. 3139–3149, 2015.
[3] C. Chen, J. Wang, F. Qiu, and D. Zhao, “Resilient Distribution System
0 by Microgrids Formation after Natural Disasters,” IEEE Transactions on
Smart Grid, vol. 7, no. 2, pp. 958–966, 2016.
[4] D. Dong, B. Wen, D. Boroyevich, P. Mattavelli, and Y. Xue, “Ana-
-200
lysis of Phase-Locked Loop Low-Frequency Stability in Three-Phase
Grid-Connected Power Converters Considering Impedance Interactions,”
-400 IEEE Transactions on Industrial Electronics, vol. 62, no. 1, pp. 310–321,
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
jan 2015.
Time (s)
[5] W. Zheng, P. Crossley, B. Xu, and H. Qi, “Transient stability of a distri-
Fig. 12: Comparison of the voltage at node 2 in the original network bution subsystem during fault-initiated switching to islanded operation,”
and the aggregated network in scenario 2. International Journal of Electrical Power and Energy Systems, vol. 97,
no. August 2017, pp. 418–427, 2018.
[6] IEEE PES Power System Dynamic Performance Committee IEEE and
400
Original (V) IEEE PES Task Force on Microgrid Stability Analysis and Modeling,
Aggregated (V) “Microgrid Stability Definitions, Analysis and Modeling,” Tech. Rep.
200 April, 2018.
[7] JOINT WORKING GROUP C4/C6.35/CIRED, “Modelling of Inverter-
0 Based Generation for Power System Dynamic Studies,” CIRED, Tech.
Rep. May, 2018.
-200 [8] Cirgré Working Group C4.605, “Modelling and Aggregation of Loads
in Flexible Power Networks,” Cigre, Tech. Rep. February, 2014.
-400
[9] D. C. Franklin and A. Morelato, “Improving dynamic aggregation of
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 induction motor models,” IEEE Transactions on Power Systems, vol. 9,
Time (s) no. 4, pp. 1934–1941, 1994.
Fig. 13: Comparison of the voltage at node 2 in the original network [10] F. Milano and K. Srivastava, “Dynamic REI equivalents for short circuit
and transient stability analyses,” Electric Power Systems Research,
and the aggregated network in scenario 3. vol. 79, no. 6, pp. 878–887, 2009.
[11] X. Liang and C. Y. Chung, “Bus-Split Algorithm for Aggregation of
Induction Motors and Synchronous Motors in Dynamic Load Modeling,”
Large voltage differences between network nodes also impact IEEE Transactions on Industry Applications, vol. 50, no. 3, pp. 2115–
2126, may 2014.
the accuracy of the aggregated model, however this error is [12] A. J. Collin, S. Member, J. L. Acosta, S. Member, B. P. Hayes, and
less significant since the voltage differences in microgrids are S. Member, “Component-based Aggregate Load Models for Combined
small, even with high network impedance. Power Flow and Harmonic Analysis,” in 7th Mediterranean Conference
and Exhibition on Power Generation, Transmission, Distribution and
There are many different types of DERs and loads connected Energy Conversion, no. November, 2010.
to distribution networks [19]. Additionally, these devices can [13] X. Xu, J. Gunda, and D. Fang, “Modelling and Aggregation of LED
have different dynamics e.g. due to the type of control [20], Lamps for Network Harmonic Analysis,” in 2018 Power Systems Com-
putation Conference (PSCC). IEEE, jun 2018, pp. 1–7.
[21], phase locked loop [22], anti-islanding detection [23], [14] L. Luo and S. V. Dhople, “Spatiotemporal model reduction of inverter-
power factor correction [19] and mechanical load [24]. Several based islanded microgrids,” IEEE Transactions on Energy Conversion,
component-based DER and load models have been proposed vol. 29, no. 4, pp. 823–832, 2014.
[15] P. Krause, O. Wasynczuk, and S. Sudhoff, Analysis of Electric Machin-
in the literature [7], [8], however more DER and load models ery and Drive Systems. IEEE Press, 2002.
should be developed and validated in future research to enable [16] J. Rocabert, A. Luna, F. Blaabjerg, and I. Paper, “Control of Power Con-
analysis of microgrids with different types of DERs and loads. verters in AC Microgrids.pdf,” IEEE Transactions on Power Electronics,
vol. 27, no. 11, pp. 4734–4749, 2012.
[17] X. Wu, X. Han, L. Liu, and B. Qi, “A load identification algorithm of
VI. C ONCLUSION frequency domain filtering under current underdetermined separation,”
IEEE Access, vol. 6, pp. 37 094–37 107, 2018.
This paper describes a methodology to aggregate grid- [18] H. Ahmadi and J. R. Marti, “Load Decomposition at Smart Meters Level
feeding DER and load models in microgrids to reduce compu- Using Eigenloads Approach,” IEEE Transactions on Power Systems,
tational burden while enabling accurate voltage and frequency vol. 30, no. 6, pp. 3425–3436, 2015.
[19] A. Collin, “Advanced Load Modelling for Power System Studies,” Ph.D.
stability analysis during and right after FII. The comparison dissertation, The University of Edinburgh, 2013.
of original and equivalent DER and load models, and the case [20] J. Rocabert, A. Luna, F. Blaabjerg, and I. Paper, “Control of Power
study show that the proposed methodology constructs accurate Converters in AC Microgrids,” IEEE Transactions on Power Electronics,
vol. 27, no. 11, pp. 4734–4749, 2012.
aggregated models. In the case study the computation time of [21] G. Buja and M. Kazmierkowski, “Direct Torque Control of PWM
the aggregated model is reduced by 87% while a small error Inverter-Fed AC Motors—A Survey,” IEEE Transactions on Industrial
is introduced, enabling accurate periodical stability analysis. Electronics, vol. 51, no. 4, pp. 744–757, aug 2004.
[22] S. Golestan, J. M. Guerrero, and J. C. Vasquez, “Three-Phase PLLs: A
Future research should treat the optimal classification of Review of Recent Advances,” IEEE Transactions on Power Electronics,
load types before equivalencing, and the development and vol. 32, no. 3, pp. 1894–1907, mar 2017.
validation of more component-based models. [23] C. Li, C. Cao, Y. Cao, Y. Kuang, L. Zeng, and B. Fang, “A review of
islanding detection methods for microgrid,” Renewable and Sustainable
Energy Reviews, vol. 35, pp. 211–220, 2014.
R EFERENCES [24] C. Cresswell, S. Djoki, and M. Ieee, “Representation of Directly
[1] Y. Wang, C. Chen, J. Wang, and R. Baldick, “Research on Resilience of Connected and Drive-controlled Induction Motors. Part 2 : Three-phase
Power Systems under Natural Disasters - A Review,” IEEE Transactions Load Models,” in Proceedings of the 2008 International Conference on
on Power Systems, vol. 31, no. 2, pp. 1604–1613, 2016. Electrical Machines, 2008, pp. 1–6.

21st Power Systems Computation Conference Porto, Portugal — June 29 – July 3, 2020
PSCC 2020