Load Modeling and State Estimation

Methods for Power Distribution Systems:

Final Report
Prepared For:

United States Department of Energy

SBIR Grant No. DE-FG02-06ER84647
DoE Project Officer:
Eric M. Lightner
OE / Forrestal Building
U. S. Department of Energy
1000 Independence Avenue, SW
Washington, DC 20585
Email: [email protected]
Phone: (202) 586-8130
Principal Investigator:

Team Member:

Thomas E. McDermott, P.E., Ph.D.

MelTran, Inc.
90 Clairton Blvd, Suite A
Pittsburgh, PA 15236-3917
Email: [email protected]
Phone: (412) 653-0407

Mesut Baran, Ph.D.

North Carolina State University
Campus Box 7911
Raleigh, NC 27695-7911
Email: [email protected]
Phone: (919) 515-5081

EnerNex Project # 1055-0001

May 7, 2010

Project 1055-0001

Distribution System State Estimation

Table of Contents

Executive Summary ................................................................................................................ 1

Accomplishments vs. Goals .................................................................................................... 2

Summary of Activity............................................................................................................... 3
3.1

Branch Current State Estimation ...................................................................................... 3

3.2

Feeder Measurements ....................................................................................................... 5

3.3

CIEE / Southern California Edison Project ...................................................................... 8

3.4

CEATI / Southern Company Project ................................................................................ 8

Products Developed .............................................................................................................. 10

References ............................................................................................................................. 11
List of Figures

Figure 1 - Line Section and State Variables for Branch Current State Estimation......................... 4
Figure 2 - Test Feeder Instrumented with Substation and Feeder Meters, Line Post Sensors,
Wireless Current Sensors, and AMI Voltage Measurements ......................................................... 6
Figure 3 - Data Historian and Feeder Electrical Model Interfaced to State Estimation ................. 7
List of Tables
Table 1 - Project Tasks, Goals, and Accomplishments .................................................................. 2

SBIR/STTR RIGHTS NOTICE

These SBIR/STTR data are furnished with SBIR/STTR rights under Grant No. DE-FG0206ER84647. For a period of 4 years after acceptance of all items to be delivered under this grant,
the Government agrees to use these data for Government purposes only, and they shall not be
disclosed outside the Government (including disclosure for procurement purposes) during such
period without permission of the grantee, except that, subject to the foregoing use and disclosure
prohibitions, such data may be disclosed for use by support contractors. After the aforesaid 4-year
period the Government has a royalty-free license to use, and to authorize others to use on its
behalf, these data for Government purposes, but is relieved of all disclosure prohibitions and
assumes no liability for unauthorized use of these data by third parties. This Notice shall be
affixed to any reproductions of these data in whole or in part.

May 7, 2010

Project 1055-0001

Distribution System State Estimation

1 Executive Summary
State estimation is a key enabler for any number of smart grid applications on the distribution
system; these include reactive power management, outage management, loss reduction, demand
response, adaptable over-current protection, condition-based maintenance, distributed generation
dispatch, integration with transmission system operations, and more. At a February, 2008 DOE
meeting hosted by Pacific Northwest National Laboratory (PNNL), state estimation was listed as
one of eight non-prioritized requirements for modeling and simulation. State estimations
importance was reinforced in DOEs first biannual Smart Grid report [1].
The project objective was to provide robust state estimation for distribution systems, comparable
to what has been available on transmission systems for decades [2]. Classical methods work
poorly on distribution feeders for several reasons:

Very few measurements are available, sometimes only the voltage and current at the
substation.

Switch states, capacitor bank states and transformer/regulator taps may not be directly
monitored, as they typically are on transmission systems.

Many of the feeder measurements are current, rather than power (P and Q).

Three-phase unbalances and low X/R ratios complicate the measurement function

In addition, its necessary to use historical load data as pseudo-measurements. Due to the radial
structure of must feeders, load and state estimation are practically synonymous for most North
American distribution systems.
This project used an algorithm called Branch Current State Estimation (BCSE), which is more
effective because it decouples the three phases of a distribution system, and uses branch current
instead of node voltage as a state variable, which is a better match to current measurement. Some
benefits of distribution system state estimation are:
1. Improved reliability
a.
b.
c.
d.

Locate faults quicker

Restore power quicker, to more customers with less risk of creating more problems
Avoid overloads
Adaptable over-current protection settings as the load or weather varies

2. Better asset utilization

a.
b.
c.
d.

Balance loads among feeders and phases

Control and dispatch of distributed generation
Volt/VAR control on feeders
Just-in-time feeder upgrades and maintenance

3. Predict load response to market or other signals

Two pilot projects are underway at Southern Company (co-funded by CEATI, DALCM 5085)
and at Southern California Edison (co-funded by CIEE, PODR01-X06). Code to support the
algorithm has been added to the OpenDSS simulator (http://sourceforge.net/projects/electricdss/).

May 7, 2010

Project 1055-0001

Distribution System State Estimation

2 Accomplishments vs. Goals

Table 1 summarizes the tasks and goals as originally planned, with actual results. In general, the
changes reflect a normal process of iterating to the best methods and platforms to implement a
new algorithm. It was more difficult than expected to find a host utility for field trials. The two
basic reasons are that utilities dont have load information correlated to their electrical models,
and that it would take significant engineering and crew time to support this project, even though
EnerNex requested no utility funding. After sustained effort, two pilot projects were secured as
described in Section 3 of this report.
Table 1 - Project Tasks, Goals, and Accomplishments
Task

Goal

Result

Phase I 1

IEEE Power Quality Data Interchange Format

(PQDIF) extensions to support state estimation

Ongoing standards participation; PQDIF

measurements not used in pilot project

Phase I 2

OSIsoft PI data historian platform development

PI-ACE training and development;

this platform not selected for a pilot project

Phase I 3

MatLab platform development

Using OpenDSS and C code instead

Phase I 4

Algorithm specification for phase II

Selected BCSE from NC State University;

converted MatLab code to C

Phase II 1

Load modeling

Based on AMI data from CEATI project

Phase II 2

State estimation voltage measurement functions,

topology identification, and testing

Completed and published, section

3.1 of this report and [6-9]

Phase II 3

Bad data identification

Testing of meter phase identification, see

section 3.4 of this report

Phase II 4

Define data historian tabs for real-time feeder

monitoring

Outlined in section 3.2 of this report

Phase II 5

Feeder model interface

Developed and tested CIM interface to

OpenDSS at the first distribution CIM
interoperability tests in December 2009.
MultiSpeak interface development underway.

Phase II 6

Software build management

Code is maintained with OpenDSS on

SourceForge

Phase II 7

Administration and review

Completed Peer Review in October 2008

Phase II 8

Marketing initiate field trials

Two projects underway, see sections 3.3 and

3.4 of this report

May 7, 2010

Project 1055-0001

Distribution System State Estimation

3 Summary of Activity
During the three years of Phases I and II, emphasis shifted from data historians to AMI projects,
which offer the best opportunity for collecting detailed load data. EnerNex contributed to the
MultiSpeak and Common Information Model (CIM) data exchange formats, which offer the best
opportunity for interfacing advanced algorithms into a utilitys existing IT infrastructure.
EnerNex also participated in open-source modeling efforts GridLab-D and OpenDSS, which
offer the most accessible simulation platforms for state estimation. The Branch Current State
Estimation (BCSE) algorithm, developed by Mesut Baran at North Carolina State University,
was selected during Phase I as the best candidate for commercial implementation. These efforts
were documented in previous project progress reports [3-5]. The rest of this section describes the
BCSE algorithm, and two current projects using it.

3.1 Branch Current State Estimation

State estimation is the prediction of all voltages and currents in the system, from a limited set of
actual measurements. It must account for missing or bad data, load variations, and local control
operations such as capacitor switching, voltage regulator operation, or automatic switch
operation. Transmission state estimation uses weighted least squares to estimate all voltage
magnitudes and phases, built around a load flow solution
The problem is tougher on distributions systems because of high feeder resistance (or low X/R
ratio), unbalanced loads and impedances, predominance of current magnitude measurements
rather than real and reactive power, and relatively fewer measurements than on a transmission
system.
Equation (1) expresses that state estimation minimizes the estimated error by weighted least
squares. The state variables, X, are the node voltage phasors (magnitude and phase). For
transmission they are positive sequence, or balanced three-phase. The measurement functions, h,
are typically load flow solution outputs corresponding to measurements from the SCADA
system. For example, h could be a node voltage (simplest) or a branch current. Or it could be P
and Q in a branch, determined by node voltages at each end, and knowing the branch (line or
transformer) impedance. Therefore, state estimation requires an electrical model of the power
system, too. The measurement values are z. The weighting functions, w, can be adjusted for the
type or quality of individual measurements.
m

min J (x) wi ( zi hi (x))2

i 1

(1)

If you have only the substation voltage and feeder current, then you have to estimate all the other
quantities by scaling the load models, based on archived data. Sometimes the archived data
accounts for season, day of the week, hour of the day, etc., but these are relatively coarse
adjustments. Inherently, the pseudo-measurements will be less accurate than real measurements.
The resulting state estimates are also relatively coarse. Existing distribution system state
estimators operate this way. The results are used to identify overloads and assist in service
restoration, but they are not good enough for smart grid applications.

May 7, 2010

Project 1055-0001

Distribution System State Estimation

Sometimes there will be a few downstream current measurements available, such as from an
automated recloser. While helpful, these are not as good as the P, Q measurements typical of
transmission state estimators. One reason is that current magnitude measurements convey no
direct information about phase angles (actually, there are non-unique solutions for the phase
angles). Another reason is that equation (1) is ill conditioned for current measurements on
lightly loaded lines, because of the nearly equal voltage magnitudes at each end.
A typical transmission state estimator decouples the solutions real and imaginary parts, taking
advantage of the fact that real power flow depends mainly on phase angles, while reactive power
flow depends mainly on voltage magnitudes. This doesnt work very well on distribution feeders
with relatively high resistance, or low X/R ratio. The solution cannot be decoupled. Also,
distribution feeders can be significantly unbalanced, and the state estimator should account for
that by providing estimates by phase.
BCSE still uses weighted least squares, but estimates the branch currents instead of node
voltages. Any voltage measurements are ignored, except for the substation bus voltage, which
becomes the reference. If you know the substation voltage plus all the branch currents, then you
know all of the downstream node voltages as well.
Figure 1 and equation (2) show the line voltage drop as a function of branch current. BCSE still
needs an electrical model of the system. There is a measurement function for each of the three
phases of a branch, but it depends only on the currents in that phase. So the state estimation
problem is decoupled by phases. The current magnitude and phase angle solutions are still
coupled, because of the low X/R ratio. Overall, the solution is easier than if using node voltages
as the state variable.
A current magnitude measurement has a simple measurement function that fits right into the
BCSE. This gets around all of the problems with current measurements in traditional SE. The
pseudo-measurements from archived load data, namely power and power factor, are converted
into load current magnitude and angle for BCSE. The missing data and topology identification
issues still apply to BCSE. These may add more variables to be estimated, such as capacitor and
switch on/off status.
Vt

Vs
ph. 1

Il,1

ph. 2

Il,2
ph. 3
grnd.

Il,3

S t,1
S t,2

Il+1,1

Vt,1 Vs,1
z11
V V g z
t,2
s,2
l 21

z31
Vt,3 Vs,3

z12
z22
z32

S t,3

z13 Il,1
z23 Il,2

z33
Il,3
(2)

Figure 1 - Line Section and State Variables for Branch Current State Estimation

Measurements of P and Q, either real or pseudo-measurements, lead to linear measurement

functions in the real and imaginary parts of the current. The most recent estimated values of node
voltage appear in the denominator of (3), and (4) combines the real and imaginary parts.

(3)
4

May 7, 2010

Project 1055-0001

Distribution System State Estimation

(4)
Current magnitude measurement leads to non-linear measurement functions, coupling real and
imaginary parts of current in equations (5) (7).

(5)

(6)
(7)
Voltage drop measurement couples both phases and real / imaginary parts of the current in
equations (8) (10). This leads to some loss in efficiency when BCSE processes voltage
measurements from AMI. In a pilot project, more emphasis was placed on AMI demand interval
measurements, which can be converted to equivalent P and Q, or to current magnitude.

(8)

(9)

(10)
Once all of the measurement functions have been defined, they are used to update the gain
matrix, G, and measurement Jacobian, H. Equations (11) and (12) are used to iteratively refine
the estimated branch currents in x.

(11)

(12)

3.2 Feeder Measurements

Figure 2 shows a radial feeder with measurement of real power (P), reactive power (Q), voltage
(V), current magnitude (I), and temperature (T) at the feeder breaker or substation bus. This is
supplemented by wireless current sensor measurements (I) at key branch points. While some
automated devices (e.g., reclosers) could also provide such measurements, they are much more
5

May 7, 2010

Project 1055-0001

Distribution System State Estimation

expensive to deploy. Wireless sensors on switches and capacitor banks can assist greatly with
network topology identification. Line post sensors at key points provide a complement of P, Q,
I, and V measurements out on the feeder. Figure 2 also shows a large number of voltage
measurements (V) from customer meters; these meters also collect load information for
statistical analysis and modeling.
v

v
v
T, V
P, Q, I

P, Q, I, V

I
I

P, Q, I

Figure 2 - Test Feeder Instrumented with Substation and Feeder Meters, Line Post Sensors, Wireless Current
Sensors, and AMI Voltage Measurements

Figure 3 shows how several data sources can be linked to BCSE through a data historian. Several
commercial SCADA and EMS products interface with Osisofts PI already.
Level 1 tags (or points, in SCADA terminology) cover what most utilities have today; although
they might not be using it for distribution state estimation.
1. Substation voltage
2. Feeder current, power, and reactive power
3. Ambient temperature
Level 2 adds the direct measurement of local control states that would otherwise have to be
estimated. For example, feeder capacitor banks could either be centrally dispatched, or
communicate their on/off state back to the substation. Disconnect switches can be automated and
their actual state would then be a measurement. Likewise for line voltage regulators. Most
automated products outside the substation fence are not compliant with IEC 61850, so
integration of them into a SE could be a bit more work.
1. Switch, capacitor, and tap changer status
2. Switch and regulator currents
Level 3 represents an attempt to flesh out the missing feeder measurements. Some utilities have
a limited number of power quality meters that could provide some data. AMI has the potential of
providing a measurement of some type at every customer load. Various low-cost sensors are also
under development; these may provide current or other measurements.
6

May 7, 2010

Project 1055-0001

Distribution System State Estimation

1. Power quality monitors (utility or customer)

2. Automated meter reading devices
3. Modern sensors
BCSE still requires a redundant set of measurements to work. Whatever measurements are still
missing have to be filled in with pseudo-measurements. These can be better utilized than they are
now, by incorporating real-time measurements such as temperature, or similar loads that have
instrumentation. This requires a statistical processing model, based on the listed types of data.
Pseudo-measurements can also be helpful in forecasting load response in real time. For example,
one might wish to predict the price signal necessary to achieve a certain load reduction in a
certain period of time. Useful load model data includes:
1. Load survey data
a.
b.
c.
d.

Customer classes
Load profiles by day and hour
kW, and either kVAR or power factor
corresponding weather data (temperature)

2. Feeder section loads

a. Customer numbers or size, and class
b. Monthly energy (kWh) for each load, with corresponding temperature

Figure 3 - Data Historian and Feeder Electrical Model Interfaced to State Estimation

May 7, 2010

Project 1055-0001

Distribution System State Estimation

3.3 CIEE / Southern California Edison Project

The California Institute for Energy and Environment (CIEE) funds this project under sub-award
PODR01-X06. It includes the use of LightHouse MV current sensors from Tollgrade
Communications [www.tollgrade.com/lighthouse]. These are clamp-on devices powered
inductively, transmitting current measurements through a wireless network. The cost is relatively
low, at about $700 per sensor. Tollgrade is based in Pittsburgh and has been in regular
communication with EnerNex about these devices.
SCE is primarily interested in the outage management aspect of state estimation, in a learning
platform for advanced distribution management system (DMS) applications, and in the
possibility of improved fault location. The LightHouse MV sensors have current waveform
capture capability that will aid in fault location. One overhead feeder served from a single
substation will be selected by SCE. The substation meters will be supplemented with
approximately 25 LightHouse current measurements linked to a single aggregator, and many
AMI voltage measurements.
SCE uses the following software and hardware that will be interfaced to the BCSE:

CYMEs CYMDIST analysis software, for the feeder electrical models

GE SmallWorld, for GIS interface and display

CGI (to become M3I) for the Outage Management System (OMS)

Emeter, for meter data management (MDM)

Itron, for automated meters

InStep eDNA, for data historian

The project was delayed starting due to state budget constraints in California, and in the
meantime the responsible personnel at SCE were transferred to other responsibilities. Site
selection is now underway to complete the project during 2010.

3.4 CEATI / Southern Company Project

CEATI (formerly Canadian Electric Association Technology Institute) funds this project as
DALCM 5085, under the Distribution Asset Life Cycle Management group. The overall goal is
to identify which phase each customer load is actually connected to, using wireless AMI data.
(Note: AMI hardware that transmits data over distribution wires would positively identify the
phase, but much AMI hardware is wireless). The application can use archived data; it does not
have to work in real time. The feeder model was converted from CYMDIST to OpenDSS for the
BCSE to run, using AMI and distribution SCADA measurements.
Results to date indicate that:
1. Some AMI meters dont have sufficient precision in their voltage measurements. For
state estimation, the hardware and configuration need to be specified ahead of time.
2. Most AMI systems dont sample or transmit voltage measurements as often as load
measurement. Some AMI systems only transmit voltage alarms.
3. A model of the service drop and transformer is necessary to link feeder voltages to AMI
voltages. This requires additional data and assumptions beyond the feeder model.
8

May 7, 2010

Project 1055-0001

Distribution System State Estimation

4. With line post sensors, AMI can be allocated to zones with improved estimation
results.
5. Further improvement is possible if using AMI demand interval measurements, either 15minute or hourly, to serve as pseudo-load measurements. Selected AMI voltage
measurements can then be checked against the estimated feeder model. However,
precision of the demand measurements can still be an issue for shorter demand intervals.
The final report on this project is due within the next month or two.

May 7, 2010

Project 1055-0001

Distribution System State Estimation

4 Products Developed
Four conference papers have been written and presented [6-9]. These all acknowledge
government support through this project. All are available through IEEE Xplore.
The project has fostered and supported many other collaborations:

MultiSpeak Initiative proposed extensions to facilitate feeder model exchange; these

were adopted in MultiSpeak version 4.

International Electrotechnical Commission (IEC) Working Group 14 contributing to an

international standard on distribution system model exchange, including improvements to
the Common Information Model (CIM) for distribution.

Electric Power Research Institute performed a gap analysis of the CIM, facilitated the
release of their Distribution System Simulator (OpenDSS) on SourceForge, and provided
a new sparse matrix solver for OpenDSS.

Pacific Northwest National Laboratory development partner for GridLab-D, working

on OpenDSS and MultiSpeak /CIM interfaces.

Tollgrade Communications helped with business case development for Lighthouse MV

wireless current sensor.

North Carolina State University is a subcontractor on this project. Also, EnerNex

supported their successful proposal to NSF establishing an engineering research center.

Elster Electric discussions of pilot project with their Advanced Grid Initiative (AGI).

OSIsoft EnerNex is a development partner for the PI data historian.

CEATI and Southern Company / Alabama Power EnerNex and NC State have nearly
completed a pilot project using AMI and DSCADA for distribution system phasing. Use
of the state estimation algorithm was a key part of this project.

CIEE and Southern California Edison EnerNex, Tollgrade, and NC State are
conducting a pilot project incorporating advanced state estimation with wireless current
sensors into a distribution management system.

10

May 7, 2010

Project 1055-0001

Distribution System State Estimation

5 References
1. U. S. Department of Energy, Smart Grid System Report, July 2009.
2. A. Abur and A. G. Exposito, Power System State Estimation: Theory and
Implementation, Marcel Dekker, 2004.
3. Mesut Baran and Jaesung Jung, Branch Current State Estimation with Voltage
Measurements, May 29, 2008.
4. T. E. McDermott, Load Modeling and State Estimation Methods for Power Distribution
Systems: Phase II Continuation Report, June 13, 2008.
5. T. E. McDermott, RDSI Peer Review Summary: Load Modeling and State Estimation
Methods for Power Distribution Systems, October 30, 2008.
6. M. Baran and T. E. McDermott, "Distribution system state estimation using AMI data,"
IEEE Power Systems Conference and Exposition, pp.1-3, 15-18 March 2009, Seattle,
WA.
7. M. Baran and T. E. McDermott, "State estimation for real time monitoring of distribution
feeders," IEEE Power & Energy Society General Meeting, pp.1-4, 26-30 July 2009,
Calgary, AB.
8. M. E. Baran, Jaesung Jung, and T. E. McDermott, "Including voltage measurements in
branch current state estimation for distribution systems," IEEE Power & Energy Society
General Meeting, pp.1-5, 26-30 July 2009, Calgary, AB.
9. M. E. Baran, Jaesung Jung, and T. E. McDermott, "Topology error identification using
branch current state estimation for distribution systems," IEEE Transmission &
Distribution Conference & Exposition: Asia and Pacific, pp.1-4, 26-30 Oct. 2009, Seoul,
Korea.

11

May 7, 2010