State Estimation Accuracy of Tuned Least Measurement Rejected Estimator PDF

You might also like

You are on page 1of 6

State Estimation Accuracy of Tuned Least

Measurement Rejected Estimator


Farhan Ammar Ahmad Mohammad Shoaib Shahriar Ibrahim Omar Habiballah Aun Haider
Department of Electrical Engineering Department of Electrical Engineering Department of Electrical Engineering Department of Electrical Engineering
University of Management and King Fahd University of Petroleum King Fahd University of Petroleum University of Management and
Technology Lahore, and Minerals Dhahran, and Minerals Dhahran, Technology Lahore,
Sialkot Campus, Pakistan Kingdom of Saudi Arabia Kingdom of Saudi Arabia Sialkot Campus, Pakistan
farhan.sayal@post.umt.edu.pk shoebeee05@gmail.com ibrahimh@kfupm.edu.sa aun.haider@skt.umt.edu.pk

Abstract—Weighted Least Square (WLS) and Weighted Least Value (LTV), Least Measurement Rejected (LMR), Maximum
Absolute Value (WLAV) estimators are most commonly employed Constrained Satisfaction (MCS) estimator etc. LMS estimator
in electric power industry. WLS fails in the presence of bad data was proposed as a robust estimator but it suffers from poor
and WLAV estimator has large computational burden. Least
Measurement Rejected (LMR) is a robust estimator which can performance in terms of asymptotic efficiency and LMS has
handle bad data efficiently and has lower computational burden. very slow convergence rate [6].
LMR estimator has an important tolerance parameter which LMR is one of the robust estimators which not only can
is used to reject bad data during estimation process. In this deal with the bad data during estimation process but can
paper, an iterative tuning approach for the tolerance parameter
also provide reliable estimation results in the presence of
of LMR has been proposed. The accuracy and computational
efficiency of the proposed approach have been compared with leverage points and also has the lower computational burden.
most commonly used WLS and WLAV estimators. The accuracy The basic idea of LMR estimator is to reject all unreliable
has been computed in terms of Cumulative Estimation Error measurements during the estimation process and this rejection
(CEE) indicator for IEEE-30 and IEEE-118 bus systems. is based upon the tolerance value of LMR. For example; a
power flow meter has an actual value of 5MW, the upper
Keywords: State Estimation, Weighted Least Square, tolerance limit is 0.7MW and lower tolerance value is 0.4MW.
Weighted Least Absolute Value, Least Measurement Re- If the measurement lies within this range, it will be accepted
jected, Bad Data. otherwise, it will be rejected. The accuracy of a meter, which is
provided by the manufacturer, can be modeled as the tolerance
I. I NTRODUCTION
parameter of LMR [7].
Power System State Estimation (PSSE) is a key component In this paper, the tolerance parameter of LMR estimator
in Energy Management System (EMS). State Estimation (SE) has been tuned iteratively, to achieve the best SE results. The
filters out the measurement data, provided by Remote Terminal upper and lower tolerance of LMR can have symmetric or
Unit (RTU) through a communication channel, and process asymmetric values. In this paper, a single symmetric toler-
this data to provide a complete picture of overall power system ance value has been assigned to all of the measurements.
network [1]-[3]. Weighted Least Square (WLS) estimator is The accuracy and computational efficiency of the proposed
most commonly employed estimator in today’s modern power approach have been compared with the WLS and WLAV in
industry but WLS suffers in term of accuracy and cannot the presence of single and multiple bad data. The accuracy
provide reliable SE results in the presence of gross error or has been measured by Cumulative Estimation Error (CEE)
bad data in the measurements. Bad data in a power system indicator, which is the sum of absolute errors between the
can be due to poor meter calibration, noise due to inductance, actual and estimated values. The best tolerance value has been
Analog-to-Digital conversion (ADC) of measurement data or chosen based upon lowest CEE indicator because lower CEE
telemetry issues etc. WLS cannot handle bad data during indicator corresponds to better estimation results.
estimation process, so there are post-estimation techniques The paper is organized as follows: section II presents the
available in the literature to deal with bad data after estimation literature review about different SE techniques, the mathemati-
process. The most popular techniques are Largest Normalized cal formulation of SE techniques has been presented in section
Residual (LNR), Chi-Square method and Hypothesis Testing III. The simulation results have been discussed in section IV
Identification method [4]. Weighted Least Absolute Value and concluded in the last section.
(WLAV) is a robust state estimator and can handle differ-
ent types of bad data during estimation process but WLAV II. L ITERATURE R EVIEW
cannot provide reliable estimation results in the presence of
leverage points and also has higher computational burden WLS is most commonly deployed estimator in power indus-
[5]. There are many robust state estimators proposed in the try throughout the world in power control centers. Basically,
literature i.e. Least Median Square (LMS), Least Trimmed SE is a highly non-linear problem and it is usually solved

978-1-5386-0922-4/18/$31.00 2018
c IEEE
by iterative methods e.g. Newton’s method. WLS is a non- but it was computationally expensive and exhaustive approach.
robust estimator i.e. it fails to produce reliable estimation In the proposed approach of this paper, LMR estimator has
results when measurements are contaminated with large errors been applied to larger systems i.e. IEEE-30 and IEEE-118 bus
or bad data. Numerous robust estimators have been formulated systems and it has been solved by mixed integer programming
in literature to minimize a non-quadratic objective function e.g. approach, a numerical method which is computationally effi-
Quadratic-Constant (QC), Quadratic-Linear (QL), Huber M- cient, instead of Genetic Algorithm.
estimator and Least Absolute Value (LAV) estimator [8]. The
implementation and performance evaluation of these proposed III. M ATHEMATICAL F ORMULATION
robust estimators (QC, QL and Huber M-estimator) have been In this paper, three state estimators have been chosen for
presented and analyzed in [9]-[11]. performance evaluation in terms of the accuracy and com-
WLAV has proved itself as a robust estimator if compared putational performance; WLS, WLAV and LMR estimator.
with WLS, but it is vulnerable to leverage points for certain WLS and WLAV are chosen because they are most commonly
measurement meter configurations [12]. WLAV has higher employed in practical power systems throughout the world
computational burden when deployed for larger power sys- and the proposed approach of tuned LMR can be tested and
tems. For numerical stability and computational efficiency in compared with these two commonly used estimators. In the
Linear Programming (LP), the scaling has been extensively following sub sections, mathematical formulation of these
used and its effectiveness for the WLAV estimator has been estimators has been presented in details.
proved in [13], where it was shown that the scaling helps
in reducing the effect of leverage points. In [14], a robust A. Weighted Least Square
WLAV-T estimator has been proposed to alleviate the effect A power system with specific network topology and meter
of leverage measurements. This approach is based upon the configuration has N buses, where m meters can be placed at
optimal transformation of associated rotation angles and scal- different locations to make the overall system observable for
ing factor in a systematic way, instead of heuristic approach. estimation process. The measurement vector z of size (mx1)
WLAV-OT has been proposed in [15] to deal with leverage is fed into an estimator to obtain the system’s state variable
measurements. It has better performance in comparison with vector x of size n = (2N − 1). The non-linear function
WLAV and WLAV-T because WLAV may depreciate its relating measurements and state variables can be expressed
performance in the presence of leverage points. The scaling as following;
factor and rotation angle was chosen by heuristic approach z = h(x) + e (1)
for WLAV-T, so there was need of a systematic approach
where:
for optical transformation which was provided by WLAV-OT.
h(x) is the non-linear function relating measurements with
There are numerous methods for identification of leverage
state vectors;
measurements which have been explained in [16][17]. WLAV
e is the measurement error vector.
also has auxiliary variables which can reduce the convergence
rate of the estimator.
SE is a problem of overdetermined nonlinear equations and
Least Median Squares (LMS) and Least Trimmed Squares
must be solved as unconstrained WLS problem. For WLS, we
(LTS) are also robust estimators developed by Rousseeuw
need to minimize the sum of the square of residuals:
and Huber, respectively [18][19]. These robust estimators have
shown high computational burden during their first implemen- n
X |(zi − hi (x))|2
tation [20] but they have the ability to alleviate the effects minJ(x) = (2)
Ri
of leverage measurements [21]. Several techniques have been i=1
proposed in [22][23] to improve the computational efficiency
of LMS and LTS. Rii = diag(σ12 , σ22 , ..., σm
2
) (3)
A mixed integer programming based robust estimator has minJ(x) = [z − h(x)]T [R−1 ][z − h(x)] (4)
been proposed and developed in [5] by M. R. Irving, which
can reject unreliable measurements with gross errors, based where:
on specified tolerance range, during the estimation process. Rii is diagonal element of Residual Covariance matrix
The proposed approach is not vulnerable to leverage points σm is standard deviation of mth measurement.
and it minimizes the number of rejected measurements during
the estimation process. The approach had been applied to a In this paper, Largest Normalized Residual (LNR) post-
very small system and it was mentioned as a limitation of the processing technique has been used for the identification and
approach. Actually, this limitation was due to the tolerance removal of bad data. It has been adopted to make the perfor-
value assigned to each measurement in a larger system. This mance comparison logical because WLS is a non-robust state
is known as Least Measurement Rejected (LMR) estimator and estimator while WLAV and LMR are robust state estimators.
somehow similar to Maximum Constraints Satisfaction (MCS) Any measurement having residual above a threshold value of
proposed in [6] and was solved by Genetic Algorithm (GA) 3 will be considered as bad data [24]. For single bad data,
to maximizes the number of satisfied uncertain measurements only a single measurement will be removed and for multiple

2
bad data, multiple measurements will be removed but one by investigated thoroughly in this paper and it has been tuned
one because there are many residuals greater than 3. LNR is iteratively for a specific power system network and it will
a sequential approach, where contaminated measurements are remain fix for that power system. The objective function of
removed one after another. the LMR estimator is to minimize the number of rejections
with a fixed value of tolerance.
B. Weighted Least Absolute Value
The minimization objective function for WLAV is: n
X
m K = min ki (7)
X |(zi − hi (x))| i=1
f (x) = (5)
i=1
σi
It can be transformed into LP problem [25]: zi − ti − M ki ≤ hi (x) ≤ M ki + ti + zi (8)

min cT .Y (6) where:


n is number of measurements
Subject to: hi (x) is measurement equation of ith meter
ti is tolerance value of ith measurement
A.Y = b ki is a binary value and indicates whether the measurement
error is within a specified tolerance limit or not
Y ≥0 M is arbitrary large scalar value.
The equations 7 and 8 have been transformed into mixed
where: integer programming problem and the formulation can be
1 1 written as [25]:
cT = [0n 0n ]
σp σp
mincT .Y (9)
0n = [0, ..., 0]
Subject to

σp = [σ1 , σ2 , ..., σm ]
A.Y = b

Y T = [xu xv u v] where:

A = [h − h Im − Im ] cT = [0n 1m ]

b = 4z A = [h − M ; −h − M ]
xu and xv are the components of state variables and both of
these variables are size of n = (2N −1), where N is number of
buses. The difference between xu and xv will provide values B T = [b + t b − t]
of state variables at each iteration [25][26]. Im is identity
matrix of size mxm, where m is number of measurements.
b = 4z
C. Tuned Least Measurement Rejected
A novel robust state estimator based on mixed integer
programming had been proposed by M. R. Irving in [5] in Y T = [4x k]
which a specific tolerance is associated with each measurement
and this approach can reject unreliable measurements based In this paper, M = 50000 has been chosen. If M is very
on the defined tolerance value during the estimation process. small, the erroneous measurements will not be rejected and if
The tolerance parameter can be the accuracy of a meter, M is very large, it will make the numerical ill-conditioning
which is specified by the manufacturer of a meter. For every [15]. A randomly chosen small value can be a starting point for
measurement, there is upper and lower tolerance value limit, tolerance parameter and for each iteration, a step increment or
while the tolerance value can be symmetrical or asymmetrical. decrement can be made in the initial or previous value. In this
In this paper, symmetrical tolerance value has been chosen paper, the starting point was 0.01 and with step decrement, the
by an iterative process for all measurements. This approach LMR has been tuned. The tuned tolerance value for IEEE-30
is trivial and effective to reject all kind of bad data during and IEEE-118 bus system has been shown in their respective
the estimation process. The tolerance parameter has been results section.

3
IV. S IMULATION R ESULTS TABLE II
IEEE-30 B US S YSTEM - M ETER D ISTRIBUTION
In this section, the simulation results of WLS, WLAV
and LMR estimator are presented and discussed in terms of Measurement Meter Type Number of Measurements
accuracy and computational efficiency. As already mentioned, Real power flow meters 41
Real power injection meters 16
the accuracy of the state estimators has been computed in Reactive power flow meters 40
terms of the CEE indicator beside, computational efficiency Reactive power injection meters 15
of the estimators is presented, which has been calculated in Voltage magnitude meters 14
seconds. The lower the CEE indicator of a state estimator, the
better is the accuracy of the state estimator. In this paper,
four types of bad data have been simulated at randomly robustness as its CEE indicator is almost around the value
chosen meter locations to prove the robustness of the proposed of White Noise case, in all kind of bad data. But WLAV
approach. While choosing a location for bad data simulation, has higher CEE indicator if compared with tuned LMR and
it has been considered that the location should not be a critical WLS with LNR post-processing. While WLS without any
location. If a meter is removed from a critical location, then post-processing has the highest indicator and it has failed to
the overall system will become unobservable. If a bad data provide reliable estimation results in the presence of different
occurs in a critical location, then the LNR approach and bad data. WLS with LNR post processing has provided reliable
most of the robust estimators cannot identify. Here is the estimation CEE indicator, but not better than tuned LMR and
description of four types of bad data; 1) Single bad data as it also has more a computational burden with steps for LNR
power flow meter represented as (SBD1) 2) Single bad data as post-processing. It can be concluded for IEEE-30 bus system
power injection meter (SBD2) 3) Single bad data as voltage that LMR has best state estimation performance irrespective
magnitude meter (SBD3) 4) Multiple non-interacting bad data of bad data type or location.
as a power flow, a power injection and a voltage magnitude
meter (MNI). Power flow or injection meter was chosen for TABLE III
bad data simulation can be real or reactive. MATPOWER [27] IEEE-30 B US S YSTEM - CEE I NDICATOR
package has been integrated into MATLAB R2015a and used
Bad Data Cumulative Estimation Error (CEE)
for the state estimation. WLS has been solved using Newton’s
method, WLAV has been solved by LPSOLVE package while WLS Tuned
Type Location WLS with WLAV LMR
LMR has been solved by mixed integer programming approach LNR
in MATLAB. The computational time has been calculated on WhiteNoise NA 0.6485 0.6485 1.2107 0.3537
Core i3 processor with 4GB RAM. The proposed approach 0.3784
SBD-1 PF 6-9 3.4075 0.6639 1.2157
has been applied on IEEE-30 and IEEE-118 bus systems. The
SBD-2 PG-5 7.1949 0.6477 1.2108 0.3668
standard deviation of measurements for IEEE-30 and IEEE-
118 bus systems are given in Table I. SBD-3 Vm-3 1.1374 0.6385 1.2108 0.3569
PF 4-6,
MNI PG-16, 13.366 0.8303 1.3648 0.7533
TABLE I Vm-21
TABLE M EASUREMENT

Measurement Meter Type Standard Deviation


Active power flow or injection 0.02
Reactive power flow or injection 0.04 The computational efficiency of the state estimators for
Voltage magnitude 0.01 IEEE-30 bus system has been shown in Table IV. It is
noticeable that WLS without LNR has best computational
time, but unfortunately, WLS failed in the presence of single or
multiple bad data as the results have shown in Table III. WLS
A. IEEE-30 Bus System
with LNR has better computational efficiency than WLAV
The meter distribution details are given in Table II for the and LMR but it requires a higher number of iterations to
test case. The reference bus is slack bus 1 and its phase converge. While LMR requires less number of iterations and
angle value is zero degrees. In IEEE-30 bus system, the full good convergence time if compared with WLS and WLAV.
redundancy value is 4.3 with 254 SCADA meters. The test
case considered in this paper has only 126 SCADA meters
with a global redundancy value of 2.13.
B. IEEE-118 Bus System
The meter distribution for the test case has been shown
Table III contains the CEE indicators of WLS, WLAV and in Table V. The full redundancy value is 4.67 with 1098
LMR estimator for the IEEE-30 bus system. The best-tuned SCADA meters in the IEEE-118 bus system. The test case
tolerance value for this measurement configuration is 0.001. considered in this paper has only 441 SCADA meters with a
It can be observed from the results that WLAV has proved its global redundancy of 1.87.

4
TABLE IV TABLE VII
C OMPUTATIONAL E FFICIENCY OF S TATE E STIMATORS C OMPUTATIONAL E FFICIENCY OF S TATE E STIMATORS

State State
Convergence Time (sec) Maximum iterations Convergence Time (sec) Maximum iterations
Estimator Estimator
Min Max Min Max
WLS 0.09 0.14 8 WLS 1.28 1.44 8
WLS with LNR 0.14 0.20 19 WLS with LNR 2.98 5.93 32
WLAV 0.30 0.34 4 WLAV 5.16 5.51 6
Tuned LMR 0.14 0.16 2 Tuned LMR 1.90 3.68 6

TABLE V
IEEE-118 B US - M ETER D ISTRIBUTION V. C ONCLUSION
Measurement Meter Type Number of measurements In this paper, an iterative approach has been proposed
Real power flow meters 134 to tune the tolerance parameter of LMR and this proposed
Real power injection meters 55
approach has been compared with two commonly used estima-
Reactive power flow meters 134
Reactive power injection meters 56 tors, WLS and WLAV, in terms of accuracy and computational
Voltage magnitude meters 61 efficiency. LMR has been tuned iteratively and best tolerance
value has been chosen based upon lowest CEE indicator.
Largest Normalized Residual (LNR) post-estimation technique
has been incorporated with WLS to filter out bad data after
The tuned tolerance value for this measurement configu- estimation process and to make the comparison logical and
ration is 0.001. The CEE indicators for IEEE-118 bus sys- justified with robust estimators. It has been observed and
tem have been shown in Table VI and it can be seen that discussed in results section that the tuned LMR has overall
WLAV has highest CEE indicator in the absence or presence best performance indices if compared with other estimation
of bad data, if compared with other estimation techniques. techniques. LMR has rejected all kinds of bad data coming
WLS without any post-estimation technique could not provide from different types and locations of meters with better com-
good and reliable estimation results, but WLS with LNR has putational efficiency. Multiple bad data is strong enough to de-
successfully identified and remove the bad data and it has teriorate the performance of a state estimator and WLS without
better CEE indicator if compared with WLS and WLAV. The any post-processing has worst performance in the presence of
results of this paper proved that the tuned LMR estimator has MNI bad data, while tuned LMR has best estimation accuracy
overall best performance indicator in comparison with other with better computational efficiency, even in the presence of
estimation techniques. MNI bad data.
In this paper, the tolerance parameter of LMR has been
TABLE VI assigned symmetrical values to all of the measurements. To
IEEE-118 B US S YSTEM - CEE I NDICATOR minimize the effects of bad data on LMR estimator, from any
meter type and location, it is recommended from simulation
Bad Data Cumulative Estimation Error (CEE)
experience that there should be more logical technique to
WLS Tuned assign a tolerance value of each measurement meter, instead
Type Location WLS with WLAV LMR
LNR of assigning a fixed symmetrical value, so that this approach
WhiteNoise NA 5.2015 5.2015 9.6025 4.5407 can be applied in practical systems reliably. The accuracy of
PF 4.6002
each meter can be modeled individually and can be assigned
SBD-1 7.6132 5.2073 9.6403
12-14 as tolerance value to each measurement. Historically recorded
SBD-2 PG-15 17.533 5.2312 9.6026 4.9776 data of the measurements can be helpful in the modeling of
SBD-3 Vm-40 6.1169 5.2087 9.3997 4.5429 tolerance parameter, as well. In this way, effects of bad data
PF12- can be minimized on LMR estimator.
14, 4.8461
MNI 9.4755 5.2189 10.796
PG-24, ACKNOWLEDGMENT
Vm21
The authors acknowledge the support provided by King
Table VII shows the convergence time and total number Fahd University of Petroleum and Minerals, Dhahran, King-
iterations required for IEEE-118 bus system to converge for a dom of Saudi Arabia.
global solution. WLS estimator has lowest convergence time R EFERENCES
but it has failed in the presence of bad data while WLS with
LNR has better convergence time, if compared with WLAV. [1] F. C. Schweppe and J. Wildes, ”Power system static state estimation. I
- exact model,” Power Industry Computer Applications Conference, pp.
The LMR estimator has much better computational efficiency 8591, 18-21 May 1969.
if compared with WLS or WLAV. [2] F. C. Schweppe and D. B. Rom, ”Power System Static-State Estimation,
Part II: Approximate Model,” IEEE Trans. Power Appar. Syst., vol. PAS-
89, no. 1, pp. 125130, 1970.

5
[3] F. C. Schweppe, ”Power System Static-State Estimation, Part III: Im-
plementation,” IEEE Trans. Power Appar. Syst., vol. PAS-89, no. 1, pp.
130135, 1970.
[4] S. Raghuraman and R. Jegatheesan, ”A survey on state estimation
techniques in electrical power system,” 2011 Int. Conf. Recent Adv.
Electr. Electron. Control Eng., pp. 199205, 2011.
[5] M. R. Irving, ”Robust state estimation using mixed integer programming,”
Power Syst. IEEE Trans., vol. 23, no. 3, pp. 15191520, 2008.
[6] A. K. Al-Othman and M. R. Irving, ”A Robust State Estimator Based on
Maximum Constraints Satisfaction of Uncertain Measurements,” vol. 40,
no. 3, pp. 585590, 2005.
[7] M. R. Irving, ”Robust Algorithm for Generalized State Estimation,” vol.
24, no. 4, pp. 18861887, 2009.
[8] A. Monticelli, ”Electric power system state estimation,” Proc. IEEE, vol.
88, no. 2, pp. 262282, 2000.
[9] F. Zhuang and R. Balasubramanian, ”Bad data suppression in power
system state estimation with a variable quadratic-constant criterion,”
IEEE Trans. Power Appar. Syst., vol. PAS-104, no. 4, pp. 857863, 1985.
[10] R. Baldick, K. A. Clements, Z. Pinjo-Dzigal, and P. W. Davis, ”Imple-
menting nonquadratic objective functions for state estimation and bad
data rejection,” IEEE Trans. Power Syst., vol. 12, no. 1, pp. 376382,
1997.
[11] R. A. Jabr, ”Power system Huber M-estimation with equality and
inequality constraints,” Electr. Power Syst. Res., vol. 74, no. 2, pp.
239246, 2005.
[12] Y. Xiaoli, H. Zongshuai, and F. Rusen, ”Weighted Least Squares State
Estimation Based on the Optimal Weight,” 2015 Third International
Conference on Technological Advances in Electrical, Electronics and
Computer Engineering (TAEECE), no. 3, pp. 1216, 2015.
[13] M. K. elik and A. Abur, ”Use of scaling in wlav estimation of power
system states,” IEEE Trans. Power Syst., vol. 7, no. 2, pp. 684692, 1992.
[14] M. K. elik and A. Abur, ”A robust wlav state estimator using transfor-
mations,” IEEE Trans. Power Syst., vol. 7, no. 1, pp. 106113, 1992.
[15] E. Caro and A. J. Conejo, ”State estimation via mathematical pro-
gramming: a comparison of different estimation algorithms,” in IET
Generation, Transmission Distribution, vol. 6, no. 6, pp. 545-553, June
2012. doi: 10.1049/iet-gtd.2011.0663
[16] L. Mili, M. G. Cheniae, N. S. Vichare, and P. J. Rousseeuw, ”Robust
State Estimation Based on Projection Systems,” IEEE Trans. Power Syst.,
vol. 11, no. 2, pp. 11181127, 1996.
[17] Rousseeuw P. J. and Leroy A. M., ”Robust Regression and Outlier
Detection, book, John Wiley and Sons, 1987.
[18] S. M. Mahaei and M. R. Navayi, ”Power System State Estimation with
Weighted Linear Least Square,” vol. 4, no. 2, 2014.
[19] M. G. Cheniae and L. Mili, ”Identification of multiple interacting bad
data via power system decomposition,” IEEE Trans. Power Syst., vol. 11,
no. 3, pp. 15551563, 1996.
[20] P. J. Rousseeuw, ”Least Median of Squares Regression,” J. Am. Stat.
Assoc., vol. 79, no. 388, pp. 871880, 1984.
[21] A. Abur and A. Exposito, ”Power system state estimation: theory and
implementation”. 2004.
[22] L. Mili, M. G. Cheniae, and P. J. Rousseeuw, ”Robust state estimation
of electric power systems,” IEEE Trans. Circuits Syst. I Fundam. Theory
Appl., vol. 41, no. 5, pp. 349358, 1994.
[23] L. Mili, V. Phaniraj, and P. J. Rousseeuw, ”Least median of squares
estimation in power systems,” IEEE Trans. Power Syst., vol. 6, no. 2, pp.
511523, 1991.
[24] M. G and A. Abur, ”LAV based robust state estimation for systems
measured by PMUs,” IEEE Trans. Smart Grid, vol. 5, no. 4, pp. 18081814,
2014.
[25] Xia Yuanhai, ”Robust Estimators in Modern Power System Network”,
Master thesis in Dec 2013.
[26] C. Xu and A. Abur, ”Robust linear state estimation using multi-level
power system models with different partitions,” 2017 IEEE Manchester
PowerTech, Manchester, 2017, pp. 1-5. doi: 10.1109/PTC.2017.7980918
[27] R. D. Zimmerman, C. E. Murillo-Sanchez, and R. J. Thomas, ”MAT-
POWER: Steady-State Operations, Planning, and Analysis Tools for
Power Systems Research and Education,” IEEE Trans. Power Syst., vol.
26, no. 1, pp. 1219, 2011.

You might also like