Sei sulla pagina 1di 6

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/224611486

Comparison of state estimation algorithms for


extreme contingencies

Conference Paper · October 2008


DOI: 10.1109/NAPS.2008.5307380 · Source: IEEE Xplore

CITATIONS READS

22 83

3 authors, including:

Noel Schulz Anurag K. Srivastava


Kansas State University Washington State University
101 PUBLICATIONS 1,384 CITATIONS 148 PUBLICATIONS 1,613 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Real Time Applications Using Linear State Estimation Technology (DE-OE0000849) View project

Resilience of Power Distribution System View project

All content following this page was uploaded by Noel Schulz on 12 August 2014.

The user has requested enhancement of the downloaded file.


Comparison of State Estimation Algorithms for
Extreme Contingencies
Srinath Kamireddy, Student Member, IEEE, Noel N. Schulz, Senior Member, IEEE and
Anurag K. Srivastava, Member, IEEE
power system and the type of measurements that can be
Abstract- The electric power grid is a complex included to provide a better state estimator solution was
interconnected system that may be subjected to blackouts analyzed [8].
and external disasters. It is necessary for the utilities to Different state estimation algorithms may perform better in
restore their power system as quickly as possible during the case of a reduced data set when compared to the
extreme conditions. The exact state of the electric grid is availability of redundant data set. Identification of efficient
needed to perform any corrective or preventive action state estimation algorithms at different levels of data
during such conditions. It is possible to get a reduced set of redundancy is necessary during extreme conditions. It is also
data or corrupted data during extreme contingencies for required to explore and evaluate different state estimation
further analysis. Major loss of sensor data during extreme methods, when there are fewer amounts of data or no data
contingencies may not help to determine the exact state. from a system network. This paper focuses on analysis of
State Estimation (SE) helps to get a better picture of the performance of SE algorithms with loss of measurements.
power system with an available set of measurements. In Section II discusses the basic role of state estimation in a
the presence of Phasor Measurement Unit (PMU) data, power system. Section III discusses the state estimation
state estimation algorithm needs to be reformulated. The algorithms and implementation of algorithms on a six bus
objective of this research is to identify the most efficient system. Comparison of state estimation algorithms with loss
state estimation algorithm for different levels of data of clustered and scattered data sets are discussed in section IV
redundancy. This paper compares the performance of with summary and future work in section V.
three state estimation algorithms with scattered and
clustered loss of measurements. II. OVERVIEW OF STATE ESTIMATION
Index Terms— State estimation, extreme contingencies, Power system network is monitored by many sensors like
sensor data, least squares. current transformer, potential transformer, relay and phasor
measurement unit. Figure 1 shows the application of state
I. INTRODUCTION estimation in power system with sensors located in a

T he control centers of utilities receive abundant raw substation. The data from sensors are transmitted to a control
measurements from their power system networks. The center by a communication network. In control center, the
measurements may be destroyed or lost while being data is fed to a state estimator program for further analysis.
transmitted to the control centers due to defects in sensors or State estimation helps to calculate the states (voltage and bus
loss of communication during a natural disaster. In such cases angles), as well as estimated values of the measurements in
the raw data has to be processed before being used for presence of error. It provides a snapshot of the power system
analysis by various Energy Management System (EMS) from available data.
applications.The operators need to take decisions based on the
analysis done by EMS applications to enable safe and secure
operation of the system. State estimation is a computer
program that helps to process the raw measurements and
determine the power system states from them.
The general concepts of state estimation in power system,
including formulation and implementation have been
discussed by the authors in [1, 2, 3]. Work has been done by
the authors in [4, 5] to incorporate constraints in state
estimation algorithms. Several inherent bad data processing
state estimation algorithms are presented in [6, 7]. The effect
Figure 1. Role of state estimation
of massive data loss from a control center in a large scale
Mathematically, measurements are related to the estimated
This work was supported by the Department of Homeland Security/ Oak value and error as shown in equation 1.
Ridge National Laboratory fund number 63886.
The Authors are with Department of Electrical and Computer Engineering,
zi=hi(x) + ei (1)
Mississippi State University, Mississippi State, MS 39762. Where, i=1, 2, 3….m
(srivastava@ece.msstate.edu, 662-325-5838)
m=number of measurements measurements with respect to states. The WLS algorithm is
ei = error in ith measurement implemented using MATLAB [13]. Table 1 shows the
zi =ith measurement estimated values of 15 measurements by WLS algorithm.
hi(x) =function relating state variables with measurements.
x=state variables (all bus voltages and angles except slack TABLE I
bus angle in this case) WEIGHTED LEAST SQUARE METHOD
Type of
All the state estimation algorithms aim to minimize the measurement
Measured Estimated
Residual
difference between measured and calculated values. value value
V2 1.0509 1.0361 0.0148
III. STATE ESTIMATION ALGORITHMS V6 1.0197 1.0110 0.0087
P64 -0.1113 -0.1178 0.0065
P43 0.0281 0.0283 -0.0002
State estimation algorithm based on following mathematical P61 -0.1851 -0.1939 0.0088
techniques has been developed. P65 0.0492 0.0415 0.0077
A. Weighted Least Squares (WLS) [9] Q64 -0.0811 -0.0798 -0.0013
B. Least Absolute Value (LAV) [10] Q43 -0.0466 -0.0479 0.0013
C. Iteratively Reweighted Least Squares (IRLS) Q61 -0.0727 -0.0764 0.0037
Q65 0.0266 0.0249 0.0017
implementation of Weighted Least Absolute Value
P2 0.3423 0.3409 0.0014
(WLAV) [11] P6 -0.2777 -0.2702 -0.0075
These algorithms are implemented on a Ward Hale 6 bus Q2 0.2136 0.2136 0.0000
[12] test system which has 2 generators, 5 loads and 7 Q4 0.0000 0.0014 -0.0014
transmission lines. The measurements are obtained by Q6 -0.1520 -0.1502 -0.0018
performing power flow and introducing a random error in the
measurements as shown in Equation 2. Voltage, power flow B. Least absolute value method:
and power injection measurements are used for state LAV algorithm minimizes the sum of the absolute value
estimation. Standard deviations (Ϭi) of 0.01, 0.01 and 0.04 are of the difference between measured and calculated values.
assumed for voltage, real power and reactive power The objective function ‘f’ is given by (4)
measurement errors respectively. m

zi= Ai*(1+RND*Ϭi) (2) f= ∑ | h (x) - z


i =1
i i | (4)
Where, zi = measured value
RND=Random number with normal distribution & zero mean i, m, hi(x), zi are same as defined in section II
Ai=Actual value from power flow Linear programming based implementation of LAV
i, m are same as defined in section II algorithm using simplex method is used here as in [14]. LAV
algorithm is implemented using the linear programming
function in MATLAB. Table 2 shows the estimated values of
the LAV algorithm for the 15 measurement case.
TABLE II
LEAST ABSOLUTE VALUE METHOD
Type of
Measured Estimated
measurement Residual
value value
V2 1.0509 1.0121 0.0388
V6 1.0197 0.9862 0.0335
P64 -0.1113 -0.1113 0.0000
Figure 2. Ward Hale 6-bus test case P43 0.0281 0.0281 0.0000
P61 -0.1851 -0.1850 -0.0001
P65 0.0492 0.0492 0.0000
A. Weighted least squares method: Q64 -0.0811 -0.0795 -0.0016
WLS algorithm minimizes the objective function which is Q43 -0.0466 -0.0466 0.0000
the sum of the squares of the difference between measured Q61 -0.0727 -0.0785 0.0058
and calculated values. The objective function ‘f’ is given by Q65 0.0266 0.0261 0.0005
P2 0.3423 0.4071 -0.0648
1

m P6 -0.2777 -0.2471 -0.0306
f= * ei2 (3)
i =1
σ i
2 Q2
Q4
0.2136
0.0000
0.2148
0.0000
-0.0012
0.0000
Where, σ i2 =variance of ith measurement Q6 -0.1520 -0.1501 0.0001

i, m, ei , hi(x) are same as defined in section II


C. Iteratively reweighted least squares implementation of
WLAV method:
Equation (3) represents a non-linear function and
minimized using Newton Raphson (NR) [12] method This method aims to minimize the objective function given
iteratively to obtain the most recent states. The jacobian used by equation (5)
is a measurement jacobian which is the change in value of
m that belongs to different parts of the system. Clustered and
f= ∑W * | h (x) - z
i i i | (5) scattered data points are shown by Figure 3. Different levels
i =1 of data redundancy are obtained by loss of various clustered
Wi=reciprocal of the variance data sets. The same levels are also obtained for scattered data
i, m, hi(x), zi are same as defined in section II loss.

The implementation of the IRLS WLAV method is similar to M-M1c=R1 (6)


that of the WLS method, except that the weights are updated M-(M1c+M2c) =R2 (7)
in every iteration based on difference between measured and M-M1s=R1 (8)
calculated value. In Table 3 the estimates of the 15 M-(M2s+ M1s) =R2 (9)
measurements by IRLS WLAV method are shown. The M= Total Number of available measurements
algorithm is implemented by developing a code in MATLAB. M1c ,M2c =Clustered data sets
M1s ,M2s =Scattered data points
TABLE III R1, R2 = Data sets at different redundancy levels
IRLS WLAV METHOD
Type of Data sets like R1, R2 are used for both the clustered and
Measured Estimated
measurement Residual
value value scattered data loss. WLS, LAV, IRLS WLAV algorithms are
V2 1.0509 1.0484 0.0025 tested with measurement sets obtained from clustered and
V6 1.0197 1.0197 0.0000 scattered data loss. L1, Infinity and Euclidean norms [15] are
P64 -0.1113 -0.1206 0.0093 calculated for every case of data loss.
P43 0.0281 0.0281 0.0000
P61 -0.1851 -0.1945 0.0094
P65 0.0492 0.0468 0.0024
Infinity norm= Maximum (|z_measured - z_calculated|) (10)
Q64 -0.0811 -0.0869 0.0058 L1-norm=Σ|z_measured-z_calculated| (11)
Q43 -0.0466 -0.0643 0.0177 Euclidean norm=√ (Σ (z_measured-z_calculated) ^2) (12)
Q61 -0.0727 -0.0727 0.0000
Q65 0.0266 0.0283 -0.0017 Values of different norms for WLS, LAV, IRLS WLAV
P2 0.3423 0.3496 -0.0073
P6 -0.2777 -0.2682 -0.0095
algorithms with loss of clustered data sets are shown by tables
Q2 0.2136 0.2860 -0.0724 4, 5 & 6.
Q4 0.0000 0.0055 -0.0055
Q6 -0.1520 -0.1506 -0.0014 TABLE IV
LOSS OF CLUSTERED DATA (WLS METHOD)
Values of L1-norm (calculated from Equation 11 in the %
Infinite Euclidean Time(sec)
Redundan L1 Norm
next section) are found to be 0.065, 0.2089 and 0.145 for the cy
Norm Norm
WLS, LAV and IRLS WLAV methods respectively. It can be 270 0.0254 0.1425 0.0016 0.4303
observed from tables 1, 2, 3 and the values of L1-norm that 227 0.0250 0.1097 0.0013 0.4061
WLS method has an overall better estimate for most of the 182 0.0121 0.0536 2.950e-4 0.5391
measurements compared to LAV and IRLS WLAV methods. 136 0.0128 0.0458 3.450e-4 0.0711
All the methods are implemented for the same set of 100 1.2717e-5 1.2739e-5 0.000 0.0360
measurements and test case.
TABLE V
LOSS OF CLUSTERED DATA (LAV METHOD)
IV. LOSS OF CLUSTERED AND SCATTERED DATA
%
Infinite Euclidean Time(sec)
Redundan L1 Norm
Norm Norm
cy
270 0.0372 0.1889 0.0042 0.3000
227 0.0383 0.1485 0.0032 0.1936
182 0.2563 0.0876 0.0016 0.3000
136 0.0216 0.0591 9.9578e-4 0.1310
100 1.2717e-5 1.2739e-5 0.0000 0.1200

TABLE VI
LOSS OF CLUSTERED DATA (IRLS WLAV METHOD)
%
Infinite Euclidean Time(sec)
Redundan L1 Norm
Norm Norm
cy
Figure 3. Clustered and Scattered data sets 0.1511 0.3829
270 0.2413 1.1238
227 0.2036 0.6706 0.0756 0.0727
Clustered data refers to the data associated with a particular 9.859e-4 0.0830
182 0.0193 0.0774
part of power system. Clustered data loss is simulated by 1.704e-4 0.0880
136 0.0122 0.0189
removing set of measurements that belong to sensors of a 0.2944 0.0197
100 0.5421 0.5884
particular bus. Scattered data is a measurement or data point
with lower norm and convergence time. At the other lower
The variation of L1-norm versus percentage redundancy data redundancy level (100%) LAV method has less error as
and time to obtain state estimation solution versus percentage well as convergence time compared to the other methods. The
redundancy for clustered data loss is shown by figures 4 and 5 computer used for simulation consists of Intel Pentium 4, 3.4
respectively. The performance of the algorithms is thus GHz processor. The time calculated for every case of data loss
compared in terms of measurement error by summing the includes convergence time for algorithm, calculation of
absolute value of each residual at different data redundancy estimated values and norms.
levels.

Figure 4. L1-Norm Vs % redundancy for clustered data loss


Figure 6. L1-Norm Vs % redundancy for scattered data loss

Figure 5.Time (sec) Vs % redundancy for clustered data loss


Figure 7.Time (sec) Vs % redundancy for scattered data loss
From figures 4 & 5 it can be observed that at higher levels
of data redundancy (270%, 227%), WLS and LAV methods V. SUMMARY AND FUTURE WORK
have approximately the same norm, but with less time of
convergence for LAV method. At medium data redundancy Three state estimation algorithms are tested on a 6 bus
level (182%), all the methods have nearly the same norm system for a particular set of measurements. They are also
while IRLS WLAV method converged faster than that of the tested with loss of clustered and scattered measurements. In
other two methods. At lower data redundancy (136%, 100%), the event of data loss, the performance of algorithms is
LAV and WLS methods have nearly the same norm with less evaluated based on error and time for convergence. The
time of simulation for WLS method. A procedure similar to operator can thus be provided an option to choose the best
that of clustered data loss is followed for loss of scattered data state estimation algorithm based on the data loss.
points and the variation of L1-norm and time with percentage Future work includes testing more state estimation
data redundancy are plotted in figures 6 & 7. It is evident algorithms with insufficient measurements on a bigger test
from figures 6 & 7 that at higher redundancy level (270%, case. Future work also includes comparing the performance of
227%), both LAV and WLS methods have lower norm with state estimation algorithms with both conventional
less convergence time for LAV method. It can also be measurements and PMU data.
observed that at medium (182%) and one of the lower (136%)
data redundancy levels IRLS WLAV method performed better
Technological Institute, India in 1997. He is working as Assistant Research
Professor at Mississippi State University since September 2005. Before that,
VI. REFERENCES
he worked as research assistant and teaching assistant at IIT, Chicago, USA
and as Senior Research Associate at Electrical Engineering Department at the
[1] Fred C. Schweppe and J.Wildes, “Power system static-state estimation,
Indian Institute of Technology, Kanpur, India as well as Research Fellow at
part I: exact model”, IEEE Transactions on Power Apparatus and
Asian Institute of Technology, Bangkok, Thailand. His research interest
Systems, Vol.PAS-89, pp.120-125, January 1970.
includes real time simulation, power system modeling, power system security,
[2] Fred C. Schweppe, “Power system static-state estimation, part III:
power system deregulation and artificial intelligent application in power
Implementation”, IEEE Transactions on Power Apparatus and Systems,
system. Dr. Srivastava is member of IEEE, IET, Power Engineering Society,
Vol.PAS-89, pp.130-135, January 1970.
Sigma Xi and Eta Kappa Nu. He is recipient of several awards and serves as
[3] Robert E. Larson, William F. Tinney, Laszlo P. Hajdu and Dean S. Piercy
reviewer for IEEE Transactions, international journals and conferences.
“State estimation in power systems, part II: implementation and
applications”, IEEE Transactions on Power Apparatus and Systems
vol.PAS-89, pp.355-363, March 1970.
[4] Ali Abur, Mehmet K. Celik “Least absolute value state estimation with
equality and inequality constraints” IEEE Transactions on Power
Systems, vol.8, pp.680-688, May 1993.
[5] Kevin A. Clements, Paul W. Davis, Karen D.Frey, “Treatment of
inequality constraints in power system state estimation” IEEE
Transactions on Power Systems, vol.10, pp.567-574, May 1995.
[6] L. Mili, M. G. Cheniae, N. S. Vichare, and P. J. Rousseeuw, “Robust state
estimation based on projection statistics”, IEEE Transactions on Power
Systems, vol.11, pp.1118-1127, May 1996.
[7] R. Baldick, K. A. Clements, Z. Pinjo-Dzigal, and P. W. Davis,
“Implementing non-quadratic objective functions for state estimation and
bad data rejection”, IEEE Transactions on Power Systems, vol.12,
pp.376-382, February 1997.
[8] Carlos Gonzalez-Perez and Bruce F. Wollenberg, “Analysis of massive
measurement loss in large-scale power system state estimation”, IEEE
Transactions on Power Systems, vol.16, pp.825-832, November 2001.
[9] Fred C. Schweppe and D.Rom, “Power system static-state estimation,
part II: approximate model”, IEEE Transactions on Power Apparatus and
Systems, vol.PAS-89, pp.125-130, January 1970.
[10] M. R. Irving, R. C. Owen and M. J. H. Sterling, “Power system state
estimation using linear programming”, IEEE Proceedings, Part C,
vol.125, pp.879-885, September 1978.
[11] R. A. Jabr and B.C. Pal, “Iteratively re-weighted least absolute value
method for state estimation”, IEEE proceedings on Generation,
Transmission and Distribution, Vol.151, Issue 1, pp. 103-108, Jan
2004.
[12] Mariesa Crow, “Computational Methods for Electric Power System”,
(Power Engineering Series, vol. 9), CRC press. 2003.
[13] Website for “Matlab User Guide.”, Mathworks Inc.,
http://www.mathworks.com/access/helpdesk/help/helpdesk.html.
[14] A. Abur and Antonio Gomez Exposito, “Power System State Estimation:
Theory and Implementation,” (Power Engineering Series, vol. 24) Marcel
Dekker, New York, 2004.
[15] http://en.wikipedia.org/wiki/Norm_%28mathematics%29.

VII. BIOGRAPHIES
Srinath Kamireddy is pursuing his master’s degree in Electrical and
Computer Engineering at Mississippi State University. He received B.tech
degree in Electrical and Electronics Engineering from Jawaharlal Nehru
Technological University, Andhra Pradesh, India in 2006. His areas of interest
include power system state estimation, wide area monitoring of electric power
grid.

Noel N. Schulz received her B.S.E.E. and M.S.E.E. degrees from Virginia
Polytechnic Institute and State University in 1988 and 1990, respectively. She
received her Ph.D. in EE from the University of Minnesota in 1995. She has
been an Associate Professor in the ECE department at Mississippi State
University since July 2001. She currently holds the TVA endowed
professorship in power systems engineering. Her research interests are in
computer applications in power system operations including artificial
intelligence techniques. She is a NSF CAREER award recipient. She has been
active in the IEEE Power & Energy Society and served as Secretary for 2004-
2007 and Treasurer for 2008-2009. She was the 2002 recipient of the
IEEE/PES Walter Fee Outstanding Young Power Engineer Award. Dr. Schulz
is a member of Eta Kappa Nu and Tau Beta Pi.

Anurag K. Srivastava received his Ph.D. degree from Illinois Institute of


Technology (IIT), Chicago, in 2005, M. Tech. from Institute of Technology,
India in 1999 and B. Tech. in Electrical Engineering from Harcourt Butler

View publication stats

Potrebbero piacerti anche