Autonomous State Estimation Based Diagnostic

System in Smart Grid

Sungyun Choi, Student Member, IEEE, A. P. Sakis Meliopoulos, Fellow, IEEE, and Ratnesh K.
Sharma, Member, IEEE

Abstract--Power system state estimation has been used to filter

raw measurement data, providing reliable and accurate real time
model, i.e. the operating conditions of the system, mathematical
model and network topology. Recently, as the power system
operation and control focuses on the distribution system (e.g.,
Smart Grid) with high penetration of distributed renewable
generators and various inverter-based smart devices, the modern
power system requires a new approach of state estimation that
can adapt to dynamic conditions of grids. We propose a new state
estimation approach for this system that operates autonomously.
The autonomous state estimation consists of two concepts: 1) a
robotic preprocessor that autonomously creates the network
connectivity, states and measurement model and 2) state
estimation. This paper applies the autonomous state estimation to
extract the real-time model of Smart Grid and then to use the
real-time model to perform diagnostic of the system. First the
real-time model is validated with standard state estimation
procedures, i.e. chi-square test for expected errors in the state
estimates and confidence level of the validity of the real-time
model. The validated real-time model is eventually used to assess
whether the system operates within operating limits and to issue
diagnostic in case system components operate near limits or
exceed limits. This paper also presents the laboratory
demonstration of the proposed diagnostic system using the Smart
Grid Energy Systems, which include a PV system, a
programmable load that can emulate the daily load profile, and
an energy storage system that has three operational modes: 1) the
standby mode, 2) the inverter mode, and 3) the charger mode.
The proposed approach is tested and verified with the Smart
Grid Energy Systems, and proper test results are presented.
Index Terms--Laboratories, power distribution, power system
analysis computing, power system modeling, smart grids, state
estimation.

I. INTRODUCTION

n power systems, state estimation provides the most likely

operating conditions (i.e., system states) based on the
measurement data and thus has been the basic function of the
energy management system (EMS) [1]. In fact, various power
system applications such as optimal power flow, economic
dispatch, and security assessment rely on the state variables of
power systems under management that are filtered initially by
state estimation.
The work was supported by NEC Laboratories America, Inc.
Sungyun Choi and A. P. S. Meliopoulos are with the School of Electrical
and Computer Engineering, Georgia Institute of Technology, Atlanta, GA
30332-0250,
USA
(e-mails:
schoi35@mail.gatech.edu
and
sakis.m@gatech.edu).
R. K. Sharma is with NEC Laboratories America, Inc. (e-mail:
ratnesh@nec-labs.com).

978-1-4673-4896-6/13/$31.00 2013 IEEE

However, decades ago, the operation and control of power

systems focused mainly on the transmission system because of
centralized generation and passive distribution systems. The
EMS as well as the state estimation is performed in the
centralized control center. In contrast, nowadays, with the
recent increase of the penetration rate of distributed renewable
generations, and with the advent of various inverter-based
smart devices including plug-in electric vehicles and energy
storage systems, the need to operate, control, and protect the
distribution system such as Smart Grid has been increasing [2].
Indeed, the operation and control of Smart Grid should be
implemented in a distributed and autonomous way by the local
distribution management system (DMS) because the
complicated configuration and dynamic nature of Smart Grid
overburdens the centralized management system [3].
As a result, in order for the DMS to perform the distributed
and autonomous operation and control over Smart Grid, the
state estimation, which has been used in the centralized EMS,
needs to be modified properly. Because of the imbalance
operation and unsymmetrical structure of the distribution
system, the device model for the state estimation should be
designed in three phases. Moreover, newly emerging smart
devices need to be modeled. Indeed, some devices change
their operational modes according to the situation, thereby
requiring a new state estimation that can adapt to the dynamic
operation of Smart Grid.
Therefore, this paper proposes a diagnostic system that is
based on the real-time model, which is validated by
autonomous state estimation. [4]. The autonomous state
estimation combines two concepts: 1) a robotic preprocessor
that autonomously integrates device models and creates the
states and measurement model for the state estimation and 2)
state estimation that extracts accurate operating conditions in
real time. Then, the results are statistically evaluated by the
chi-square test, presenting the confidence level, which
indicates the consistency between measurements and device
models. Finally, the confidence level can be used to detect bad
data and to check the integrity of the system operation.
This paper also presents the laboratory demonstration of the
proposed diagnostic system using the Smart Grid Energy
Systems in NEC Lab America, Inc. The test system consists of
a PV system, a programmable load that can emulate the daily
load profile, and an energy storage system that can change its
operational mode, and the test system is connected to the main
grid, allowing the bi-directional power flow. The proposed
method was tested with this laboratory setup, and several test
results are presented.

II. PROPOSED DIAGNOSTIC SYSTEM

The proposed diagnostic system is based primarily on the
autonomous state estimation, which provides the real-time
operating conditions of Smart Grid with distributed generation
and smart devices that can be plugged and unplugged to the
grid any time. As depicted in Fig. 1, whenever new devices
are plugged into the grid, or whenever operational conditions
of devices are changed, status of each model is updated
through communication networks so that the state estimation
can keep tracking of the frequent reconfiguration of Smart
Grid. Device models are integrated through the model
integrator and then used to create measurement model, h(x),
with measurement data from the actual field. Based on the
measurement model, state estimation finally extracts the realtime model of the grid.

diagnostic system can check if the component operates within

its operating limits and warn the system operator about the
status of the component. For example, the operating limits of a
distribution transformer can be determined by monitoring
currents that flow through windings; note that the currents can
be obtained from the real-time model, which is validated by
the autonomous state estimation. Then, based on the electrothermal model of the transformer [5], [6], the diagnostic
system can compute the loss of life of the transformer, thus
providing the time to the operating limit as described in Fig. 2.
Finally, the diagnostic system alarm if the overloading
conditions maintain for the time limit.

Fig. 2. Transformer dynamic loading.

Fig. 1. Diagnostic system based on autonomous state estimation.

The results of the state estimation are evaluated by wellknown chi-square test, which quantifies the goodness of fit of
the measurements to the model, providing the probability that
the measurement errors are distributed by the chi-square
distribution; this probability can be interpreted as confidence
level between measurements and the model. Then, the
confidence level can be used to validate the real-time model in
terms of 1) the existence of bad measurement data and 2) the
integrity of system operating conditions. In other words, the
high confidence level can be interpreted with two aspects: 1)
the measurement data is trustable to be used for the operation
and control over Smart Grid, and 2) the system operates
without malfunction. On the contrary, the low confidence
level indicates the existence of bad measurement data or the
faulty operation of the system.
Based on the validated real-time model of each component
in Smart Grid, the diagnostic system can draw conclusions
about the health of the grid or individual components. The

All in all, the proposed diagnostic system is based on the

real-time model of Smart Grid. Indeed, the network topology
of Smart Grid is dynamic and changes with time, and the
operational modes of each device are variable. Therefore, the
autonomous state estimation can adaptively provide the
validated real-time model of Smart Grid and diagnostics that
are reliable.
III. AUTONOMOUS STATE ESTIMATION
Autonomous state estimation consists of two concepts: 1) a
robotic preprocessor that autonomously defines system states
and creates the measurement model and 2) state estimation
that extracts accurate operating conditions in real time.
A. Robotic Preprocessor
In the robotic preprocessor, the model integrator play an
important role in autonomously detecting system changes of
Smart Grid and reconfiguring states and the measurement
model for state estimation. For integrating models, the
proposed approach uses the symbolic integration method [7],
which represents device models in a generalized symbolic
form; based on the generalized form, any kinds of power
devices in Smart Grid can be modeled and automatically
integrated into one system model. The generalized form of
device models can be expressed in the following algebraic

companion form (ACF):

iac
vac
I = K + Y V + f (t )
dc
dc
0
y

(1)

vac T vac

f (t ) = Vdc Qi Vdc
y
y

(2)

where iac is the vector of ac-current phasors, Idc is the vector of

dc-current values, vac is the vector of ac-voltage phasors
(external state variables), Vdc is the vector of dc-voltage values
(external state variables), y is the vector of internal state
variables, K is the constant vector, Y is the device model
matrix, f(t) is the vector of nonlinear terms, and Qi are the
quadratization matrices. Note that each phasor value in the vac
or iac vector can be divided into real and imaginary part of the
phasor.
Then, the integrated device model is used to create
measurement model, h(x), along with measurement data. The
measurement model for state estimation is expressed as
follows:

z = h(x ) +

(3)

where z is the vector of measured values, h is the vector of the

function of states, x is the vector of state variables, and is the
vector of measurement errors. It is necessary to point out that
state variables of the overall system comprise external states
(i.e., ac- and dc-voltages) and internal states of each device
model [refer to the generalized device model expressed as (1)
and (2)]. In the proposed method, the function of states can be
represented with a set of linear and nonlinear (quadratic at
most) combination of state variables, and thus, each
measurement model [i.e., one row in h(x)] can be expressed as
the following standard format:

z m = c + ai xi + b jk x j xk + m
i

algorithm can be used to solve above optimization problem.

Once the solution of the state estimation determines the
best estimate of system states, the results can be evaluated by
chi-square test, which provides the probability of how well the
measurement are consistent with the system model (i.e., the
goodness of fit). This probability can be interpreted as the
confidence level. Typically, the high confidence level
indicates the measurement data are consistent with the system
model. On the contrary, the low confidence level indicates the
inconsistency between the measurements and the model, and
therefore, it implies the existence of bad data or the incorrect
operation of the system.
IV. LABORATORY TEST
The proposed approach was implemented with the Smart
Grid Energy Systems in NEC Lab America, Inc.
A. System Description
The Smart Grid Energy Systems is designed to test the
intelligent power management system (IPMS), which operates
and controls the energy system that consists of renewable
energy resources and the energy storage system in accordance
with various operational objectives (e.g., economic dispatch or
environmental dispatch). The overall system diagram is
described in Fig. 3. The PV system has a total of 21 PV panels,
generating maximum 6kW, and a programmable load can
simulate the daily load profile. An energy storage system with
a maximum output of 10kW provides insufficient power to the
system according to an optimization algorithm of IPMS. The
test system is connected to the main grid (i.e., the utility),
exporting or importing energy. The power flow can be
measured by a grid meter at the point of connection to the
main grid. Furthermore, an Arbiter relay, a PV inverter, an
energy storage system, a load, and Tigo Maximizers
(maximum power point tracking devices) are streaming
measurement data to a database system of IPMS in real time.

(4)

where zm is the measured value, c is the constant term, ai are

the linear coefficient terms, bjk are the nonlinear coefficient
terms (i.e., quadratic terms), xi, xj, and xk are the state variables
(i, j, and k are the indices of states), and m is the measurement
error.
B. State Estimation and Statistical Evaluation
The proposed autonomous state estimation is based on the
weighted least squares (WLS) algorithm [8], and the objective
function, which minimizes the weighted, squared residuals of
measurements, can be formulated as follows:

Minimize J = [ z h( x)]T W [z h( x)]

(5)

where W is the weight matrix whose diagonal entries are the

inverse of the variance of measurement errors. The iterative

Fig. 3. Diagram of the Smart Grid Energy Systems.

The energy storage system with four battery compartments

has three modes: 1) the standby mode, 2) the inverter mode,
and 3) the charger mode. In the standby mode, the dc-to-ac
inverter and the ac-to-dc charger are disconnected from both
the batteries and the grid; the switches at both sides of the
inverter and the charger are open (see Fig. 4). Only small
amount of power is consumed by the user display, which is
represented by the conductance, GB. In the inverter mode, the
switches at both sides of the inverter are open, and in the
charger mode, the switches at both sides of the charger are
open. Note that the robotic preprocessor automatically
changes the device model of the storage system according to
its operation mode. In this model, the conductances, G1 and G2, Fig. 5. Daily load profile.
are added to take into account the average power loss of the
inverter and the charger, respectively.

Fig. 6. On and off status of the operation modes of the energy storage system.

Fig. 4. Details of the energy storage system.

B. Test Scenarios
The proposed approach was tested with the Smart Grid
Energy Systems during one and a half days. With the
programmable load, a usual daily profile is given during the
test time as depicted in Fig. 5. Furthermore, Fig. 6 shows the
on and off status of the inverter and charger modes; when the
mode is on, the value is one, and when the mode is off, the
value is zero. As described in Fig. 6, the inverter mode is on
around 11am and 10pm on the first day, and then, the mode
turns off after a certain time. Right after the inverter mode
becomes off, the charger mode is on around 11:40am and
23:50pm on the first day. After several hours, the charger
mode becomes off. It is necessary to point out that if both the
charger and inverter mode are off in Fig. 6, the operation
mode is standby. The optimization algorithm to operate the
energy storage system is out of scope of this paper.
For the autonomous state estimation, the PV inverter, the
load, and the energy storage system are modeled. The state
estimation can be performed with each single device and the
corresponding measurements or with the integrated model of
all three devices and the corresponding measurements. Finally,
the estimated results can be evaluated by two points of view: 1)
validation of measurements and 2) validation of system
operating conditions.

1) Validation of Measurements
If there are bad measurement data, the statistical evaluation
of the autonomous state estimation indicates the existence of
bad data by producing the low confidence level. For this test
scenario, the individual component is tested with the
corresponding measurement data. Then, the integrated model
of three devices is tested.
2) Validation of System Operating Conditions
The statistical evaluation of the autonomous state
estimation is capable of checking if the system under
monitoring operates correctly or not. In the test-bed, the
energy storage system has three operational modes, and
therefore, the device model of the system should be able to be
changed to the corresponding mode. Otherwise, the
confidence level would be low, and then it can be concluded
that the actual operation of the system do not match the device
model. Six test scenarios are listed in TABLE I.
TABLE I
TEST CASES
Test Case
1
2
3
4
5
6

Actual Operation
Inverter mode
Inverter mode
Charger mode
Charger mode
Standby mode
Standby mode

Device Model
Standby mode
Charger mode
Standby mode
Inverter mode
Inverter mode
Charger mode

V. TEST RESULTS
A. Validation of Measurements
Each device of the PV inverter, the programmable load,
and the energy storage system was individually tested, and
then, the integrated model of three devices is tested.
1) PV Inverter
When testing with the PV inverter model, the results
indicate that the confidence level is 100% over the testing
period.
2) Programmable Load
When testing with the programmable load model, the
results indicate that the confidence level is 100% over the
testing period.
3) Energy Storage System
Fig. 7 represents the confidence level over the testing
period when testing with the energy storage system. The
figure indicates that the transient decrease of the confidence
level occurs when the inverter switches or the charger
switches operate. The first reason for this transient decrease is
that device models in use are based on the static model in the
frequency domain, and the second one is that the sampling rate
is not high enough to capture the transient moments. However,
the transient period is very short, and thus, the low confidence
level in the transients can be negligible unless the fast
response operations and controls are required.

two equations:
2

0 = V1P I1P + V1N I1N

I 1P
I
1N + P1
G1
G1

0 = V2 P I 2 P + V2 N I 2 N

I 2P
I
2 N + P2
G2
G2

(6)

(7)

Fig. 8. Confidence level and the measurement of the power input to batteries
of the energy storage system through the charger. CC represents the constant
current mode, and CV does the constant voltage mode.

4) Integrated Model
The device models of the PV inverter, the programmable
load, and the energy storage system are integrated into one
system model, which is then tested. As a result, Fig. 9
indicates that the confidence level is nearly 100% during the
testing time but decrease temporarily at the transient moment
of switch operations. However, unlike the test case with only
the energy storage system, the confidence level is 100% even
at the constant voltage charging mode. This is due to the fact
that the degree of freedom increases with the integrated model.

Fig. 7. Confidence level with the on and off status of the inverter and charger
mode when testing with the energy storage system.

Furthermore, it can be pointed out that when the charger

mode is on, the confidence level slightly decreases after a
certain period of time. This behavior is marked with red
circles in Fig. 8. Note that the confidence level is low when
the storage system are in the constant voltage charging mode,
where current that flows to batteries decreases significantly
while the voltage keeps a constant level. In fact, powerelectronics-based devices such as inverters or chargers
generate current harmonics, which eventually affect the
accuracy of current or power measurements especially when
the current magnitude or the active power is relatively low.
After all, in the constant voltage mode, where a small amount
of power (i.e., active power) flows to batteries through the
charger, the confidence level can decrease slightly because the
device model of the energy storage system, described in Fig. 4,
has several power-related equations including the following

Fig. 9. Confidence level when testing with the integrated model. The on and
off status of the inverter and the charger are represented together.

B. Validation of System Operating Conditions

A total of six test cases are tested with the integrated device
model of three devices. All results of six test cases are shown
in Fig. 10, indicating that in all test cases, the confidence level
is zero when the actual operation mode is not consistent with
the device model.

VIII. ACKNOWLEDGEMENT
The authors would like to thank NEC Laboratories America,
Inc. and Yanzhu Ye. for supporting this work.
IX. REFERENCES
[1]

[2]

[3]

[4]

[5]
[6]

[7]

[8]
Fig. 10. Test results for system operation health index.

VI. FUTURE WORK

The determination of data validity is based on the chisquare test that provides a confidence level of data validity.
Using the valid data the proposed system compares the realtime model of each component and compares the operation to
the recommended range of operating parameters for the device.
Diagnostics can be issued if the operating conditions of each
device are close to the operating limits or they are exceeded.
Future work can include improvements of the autonomous
state estimation as well as improvements in the diagnostic
system. For example, more redundant measurement data can
help the autonomous state estimation to filter out the bad data.
Improvements in the diagnostic system can be performed by
adding analytics that will predict when operating limits may
be violated by a device. An example will be a transformer
operating at overload; additional analytics can determine how
long it will take for the transformer temperature to reach the
permissible value.
VII. CONCLUSION
The proposed diagnostic system is based on the
autonomous state estimation, which can update the state
estimation problem in real time to adapt to dynamic operations
of Smart Grid. Then, the statistical evaluation of the state
estimation results yields the confidence level that quantifies
the goodness of fit of the model to measurement data. Then,
the measurement data and the system operating conditions can
be validated. The proposed approach was tested with the
Smart Grid Energy Systems in NEC Lab America, Inc.

F. C. Schweppe, J. Wildes, and D. B. Rom, "Power system static-state

estimation, part I, II, and III," IEEE Trans. Power App. Syst., vol. PAS89, no. 1, pp. 120-135, Jan. 1970.
M. Kezunovic, J. D. McCalley, and T. J. Overbye, "Smart grids and
beyond: Achieving the full potential of electricity systems,"
Proceedings of the IEEE, vol. 100, no. 13, pp. 1329-1341, May 2012.
A. P. S. Meliopoulos, G. Cokkinides, R. Huang, E. Farantatos, S. Choi,
Y. Lee, and X. Yu, "Smart grid technologies for autonomous operation
and control," IEEE Trans. Smart Grid, vol. 2, no. 1, pp. 1-10, Mar.
2011.
S. Choi, B. Kim, G. J. Cokkinides, and A. P. S. Meliopoulos,
"Feasibility study: Autonomous state estimation in distribution
systems," IEEE Trans. Power Syst., vol. 26, no. 4, pp. 2109-2117, Nov.
2011.
IEEE Guide for Loading Mineral-Oil-Immersed Transformers, IEEE
Standard C57.91-1995, Jun. 1995.
A. P. S. Meliopoulos, G. J. Cokkinides, and T. J. Overbye, "Component
monitoring and dynamic loading visualization from real time power
flow model data," in Proc. the 37th Annual Hawaii International
Conference on System Sciences, Jan. 2004.
A. P. S. Meliopoulos, G. J. Cokkinides, S. Choi, E. Farantatos, R.
Huang, and Y. Lee, "Symbolic integration and autonomous state
estimation: Building blocks for an intelligent power grid," in Proc. the
16th International Conference on Intelligent System Application to
Power Systems (ISAP), Sept. 2011.
A. Abur and A. G. Exposito, "Chapter 2: Weighted least squares state
estimation" in Power system state estimation: Theory and
implementation: Marcel Dekker, Inc., 2004.

X. BIOGRAPHIES
Sungyun Choi (S 09) received B.E. degree in Electrical Engineering from
Korea University, Seoul, Korea in 2002. And then, he worked as a network
and system engineer in Korea. Since 2007, he started graduate studies at
Georgia Institute of Technology, where he received the M.S. degree in 2009.
He is currently pursuing his Ph.D. degree, working in Power System Control
and Automation Laboratory at Georgia Institute Technology. His research
interest lies in distributed power system automation and control, power system
protection, communication networks and systems in substation automation.
He is a student member of IEEE.
A. P. Sakis Meliopoulos (M 76, SM 83, F 93) was born in Katerini,
Greece, in 1949. He received the M.E. and E.E. diploma from the National
Technical University of Athens, Greece, in 1972; the M.S.E.E. and Ph.D.
degrees from the Georgia Institute of Technology in 1974 and 1976,
respectively. In 1971, he worked for Western Electric in Atlanta, Georgia. In
1976, he joined the Faculty of Electrical Engineering, Georgia Institute of
Technology, where he is presently a Georgia Power Distinguished Professor.
He is active in teaching and research in the general areas of modeling,
analysis, and control of power systems. He has made significant contributions
to power system grounding, harmonics, and reliability assessment of power
systems. He is the author of the books, Power Systems Grounding and
Transients, Marcel Dekker, June 1988, Lightning and Overvoltage Protection,
Section 27, Standard Handbook for Electrical Engineers, McGraw Hill, 1993.
He holds three patents and he has published over 220 technical papers. In
2005 he received the IEEE Richard Kaufman Award. Dr. Meliopoulos is the
Chairman of the Georgia Tech Protective Relaying Conference, a Fellow of
the IEEE and a member of Sigma Xi.
Ratnesh K. Sharma (M '11) leads the Energy Management Department at
NEC Laboratories America. He has a PhD degree from University of
Colorado at Boulder and BTech. (Hons.) degree from Indian Institute of
Technology, Kharagpur. His research interests span sustainable energy
management in electricity, buildings and transportation sectors including
energy conversion, power systems, communications and analytics. He has
authored more than 150 papers/technical reports and holds over 60 US patents.