This article was downloaded by: [130.132.123.
28]
On: 31 December 2014, At: 12:22
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK
International Journal of Control
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tcon20
Statistical linearization approach to power-system state
estimation
a a a
GULAB SINGH , K. K. BISWAS & A. K. MAHALANABIS
a
Department of Electrical Engineering , Indian Institute of Technology , Delhi, Hauz Khas,
New Delhi, 110029, India
Published online: 12 Mar 2007.
To cite this article: GULAB SINGH , K. K. BISWAS & A. K. MAHALANABIS (1978) Statistical linearization approach to power-
system state estimation , International Journal of Control, 28:3, 411-418, DOI: 10.1080/00207177808922467
To link to this article: http://dx.doi.org/10.1080/00207177808922467
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the
publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations
or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any
opinions and views expressed in this publication are the opinions and views of the authors, and are not the
views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be
independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses,
actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever
caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
INT. J. CONTROL, 1978, VOL. 28, No.3, 411-418
Statistical linearization approach to power-system
state estimation']
GULAB SINGRt, K. K. BISWASt
and A. K. MAHALANABISt
New algorithms are presented for power-system complex node voltage estimation,
based on statistical linearization of power-flow equations and a Jeeat-equeree criterion.
To illustrate the algorithms a 7-bus power net-work is simulated on a computer and
a comparison is made with estimates obtained through the conventional Taylor
series approach.
1. Introduction
Downloaded by [] at 12:22 31 December 2014
State estimation in a power system has been of considerable interest
within the power utility during the past few years. The state estimator
happens to be a data-processing algorithm for converting redundant meter
readings and other available information to yield an estimate of the state
vector (which can be either complex node voltages; tie-line power, supply
frequency or system loads). It is used for building up a reliable data base
giving information about the system structure and state. With suitable
display techniques the estimation algorithms effectively facilitate the task of
control-room supervision in a large power system. The state estimator can
also be used for security assessment, generation of suitable load frequency
control policies, economic load despatch and on line-stability analysis. In
this way state estimation becomes an integral part of the automatic power-
system control.
The static state estimation of complex node voltages in an interconnected
power system has been studied by several workers. Schweppe (1970) and
Schweppe and Rom (1970) have discussed the concepts underlying a static
state estimator and the corresponding detection and identification logics.
Larson et al. (1970) have presented a methodology for processing the measured
data from electric power networks in order to obtain the best possible estimate
of system variables. This involves linearizing the power-flow equations using
a Taylor series and applying a least-squares criterion. Debs and Larson
(1970) were the first to propose a dynamic estimator for a power system using
the well-known Kalman filtering technique. Masiello and Schweppe (1971)
have also suggested a tracking algorithm based on a weighted recursive least-
squares criterion. A complete review of such work has been presented by
Schweppe and Handschin (1974).
Received 19 Julv 1977.
t The work was supported by the National Science Foundation, Washington,
under grant No. OIP72-02921 AOI.
t Department of Electrical Engineering, Indian Institute of Technology, Delhi,
Hauz Khas, New Delhi-lI0029, India.
3F2
412 G. Singh et al.
Instead of linearizing the power-flow equations through a Taylor series
expansion, this paper employs a statistical linearization technique to develop
new algorithms for power-system state estimation (both static and dynamic).
In order to reduce the increased computational burden due to the introduction
of a system covariance matrix it is proposed that the voltage and node angle
bc estimated separately using decoupled estimators. The performance of
such estimators is compared with conventional algorithms on a simulated
power grid.
2. Problem formulation
The state vector corresponding to the complex node voltages can be
expressed either in terms of rectangular or polar coordinates. In an inter-
connected power system of n nodes let Pi and qi denote respectively the active
and reactive power injections and let viL(li represent the complex node
voltage at the ith node, The relationship between the various variables are
represented by the power-flow equations
n
Downloaded by [] at 12:22 31 December 2014
Pi= L
j",oj
vivjYij cos ((li-(lj+<Pij) ( 1)
qi= L" ViVjYij sin (Oi-(lj+<P;j), i=I, 2, ... , n (2)
j-I
where (Ii is the phase angle at the ith node with reference to the slack bus.
The parametcrs Yij and <Pij represent respectively the magnitude and angle
of admittance function of the branch connecting the ith and jth node and
are assumed known. The 2n -I-dimensional state vector X which com-
pletely specifies the system will have the structure
X=[(l2' ... , (In' VI' V2' ... ,vnJ' (3)
The power meter readings may be contaminated with telemetering noise or
biasing errors. Assuming these to be additive in nature, the following
measurement models are chosen:
Zli=Pi+'7li (4)
Z2i=qi+'72i' i=I,2, ... ,n (5)
where '7li and '72i represent errors of measurement. The measurement vector
Z is constructed as follows:
The dimension of Z will increase if measurement on VOltages and bus power
flows arc also made. The estimation problem is to determine the best value
of X based on the measurement Z,
3. Statistical linearization of power-flow equations
In terms of the state vector, the measurement eqns. (1), (2), (4) and (5)
can be put in a compact form as
Z=h(X)+'7 (7)
Power-system state estimation 413
The statistical linearization technique involves expressing the non-linear
function h(X) as (Mahalanabis and Farooq 1971)
h(X)=<X+B(X -X)+e (8)
where e represents the collection of error terms due to first-order approxima-
tion and X is the nominal value of the state X around which the non-linearity
is to be approximated. <X is a vector, and B a matrix of appropriate dimen-
sions, depending on the dimension of X and Z. These are to be determined,
such that the J 1 = E[ee' jZ] is minimized.
From (8)
./1 = E[{h(X) - <X- B(X - X)}'{h(X) - <X- B(X - X)}jZ]
It turns out that necessary and sufficient conditions which minimize the
above index are
<X = E[h(X)jZ] =k(X) (9)
and
B = E[(h(X) - h(X))(X - X)' jZ]P-l (10)
Downloaded by [] at 12:22 31 December 2014
where P is the error covariance matrix:
P = E[(X - X)'(X - X)jZ] (11 )
The evaluation of the matrix B is simplified on the assumption of gaussian
statistics for X. In other words, the conditional probability density function
of X is assumed to be
p(X) = (27T)-(2n-l)f2IPI-l/2 exp [ -l(X - X)' P-l(X -X)]
The elements of the matrix B ij are then obtainable as follows:
,. a". I
B··=-' (12)
'J ax. )
v_v
~\.j-~\.J
Expressing the active and reactive measurements through the vectors hi and
h2 respectively, (8) may be decomposed as
hli(X) = <Xli + jt! Biij(X j - Xj)) (13)
h2i(X)=<X 2i+ L" B2ij(Xj-Xj)
j=l
Representing the state X as X = X + X and carrying out the required expecta-
tion operation the components of <X are obtained as (Sharma 1973)
<Xli= L"
;=1
kYi;[COS ((Ji-Oj+.pij){ViVj+Pi/'''
- (P i/" - Pir)(Pi/V - Pir)} -sin ((Ji - OJ + .pij)
{Vi(P;/v- Pi/OJ +Vj(Pi/v - pirm (14)
<X2i= L"
j~l
kYij[sin (Oi-Vj+.pij){ViVj+P i;""
- (Pj/v - Pir)(Pi/v - Pjr)} + cos ((Ji - (Jj+ .pij)
{Vi(Pi/v - Pjr) + Vj(Pi/v - Pi/Oj}] (15)
414 G. Singh et al.
where k = exp [ - !(Pi/o + Pj/o - Pi/OJ
POO = [( Ii - 0)( Ii - 0)']
puo = rou = E[(v - v)( Ii - 0)']
puu = E[(v - v)(v - v)']
The matrices P'", Pv", pou and Pvv in fact form the partitioned submatrices
of the state covariance matrix P :
P OOi POU]
P= -j-...... = E[X, X']
[ r«, puu
The elements B l i j , B 2i j are obtained by carrying out the indicated differentia-
tions.
4. Estimation algorithms
4 .1. Static case
Downloaded by [] at 12:22 31 December 2014
Having computed IX and B based on nominal values of X and system
covariance, (7) can be expressed as
Z=IX+B(X -X)+'l')
Since IX represents the expected value of the measurement, a perturbation
measurement equation ml1Y be written as
I::.Z=Z-IX=BI::.X+71
where I::.X = X - X is the perturbation in state. A least-squares estimate of
I::.X based on I::.Z is obtained by minimizing the performance index
(16)
where Jl[ is a suitable weighting matrix. Minimization of (16) yields
I::.X = (B' Jl[ B)-I B' Jl[ I::.Z (17)
o
GENERATOR BUS
oLOAD BUS
- - 220 KY. LINE
--132 KY. LINE
BHAKRAIIIl-----------1
Figure 1. i-bus power network of Northern India grid.
Power-system state estimation 415
which gives rise to the following iterative algorithm:
gi+l=gi+PiBCg i, Pi)M[Z_IX(gi, Pi)] (18)
pi+l = [B'(gi, Pi)MB(gi, Pi)]-l (19)
Equations (18) and (19) are processed, starting with the initial values go
and po until gi converges. It is expected that inclusion of knowledge regard-
ing system covariance will improve the accuracy of estimation over the
conventional algorithm.
4.2. Dynamic case
For slowly varying loads, the dynamics of the power plant can be modelled
as
X(k + 1) =X(k) + ,B(k) (20)
where k represents the discrete time index. The term ,B is introduced to take
care of fluctuations in state from instant to instant. The measurement
equation corresponding to (20) is
Downloaded by [] at 12:22 31 December 2014
Z(k) =h[X(k)] + W(k) (21)
A tracking state estimator can be developed, assuming that the state variables
do not change appreciably between the time instants. Minimization of a
standard least-squares criterion yields the following algorithm:
g(k+ 1) = g(k) + P(k)B'(g(k), P(k))M[Z(k+ 1) -1X[g(kj, P(k)]] (22)
P(k+ 1) = [B'(x(k)1, P(k))MB(x(k), P(k))]-l (23)
To improve estimation accuracy each measurement may be iterated more
than once.
STATIC ESTIMATION
<;>-- --<> LUMPE0 ESTIMATOR
-----..< DECOlJPLED ESTIMATOR
~ ~ ~ ~ ~ ~ ~ ~ ~ ~
STATES -
Figure 2. Static estimation results for decoupled estimator.
416 G. Singh et al.
4.3. Decoupled estimator
It is apparent that while the estimator structures obtained are similar to
the conventional ones, the computation of '" and B involves increased effort.
To reduce the computational burden it is proposed that (J and v be estimated
using two separate estimators. Since the variation in (J is more dependent
on variation in active power p and less on reactive power q and since the
variations in v are largely attributed to changes in q, the estimator proposed
earlier can be decoupled as follows.
Assuming a nominal value of v, the phase angle (J can be iteratively
cstimated using only hI from (13). Using this value of (J, v can be estimated,
based on the remaining measurements (h 2 ) . Thus the task of handling
2n x 2n-dimensional matrix multiplications and inversions is reduced to
handling twicc n x n-dimensional matrices. A rough idea of saving in com-
putation time in decoupled estimation can be obtained by counting the total
number of multiplications and inversions involved in either case. For an
n-node system this works out to be (40n 3-32n2+12n) for the conventional
algorithm and (lln 3 - 8 n 2+3n) for the proposed algorithm at each iteration.
Downloaded by [] at 12:22 31 December 2014
DYNAMIC ESTIMATION
_ _ _ 7'AYLo/l SERI£S
_ _ _ _ _ STATISTICAl. LIAlEAIlIz,t.TION
~
'I
, I
I I
I 1
I 1
I I
I :
I I
,,
I \
I
\
I \ P
I
I
\ ,
/
5
TIME.
7
INDEX
9
"
Figure :1. Absolute value of estimate error plot for 83 using dynamic estimator
5. Results of simulation studies
In order to test the proposed algorithm a 7-node power network of
Northern India was chosen (Fig. 1). Using some nominal values of (J and
v, p,. and q,. were computed using power-flow equations. The corrupted
measurements were simulated by adding random noise to Pi and qi. A random-
number generator subroutine was used which supplied zero mean, gaussian
Power-system state estimation 417
distributed noise sequences with a standard deviation of 0·13. The measure-
ments were fed both to decoupled estimators and conventional lumped
estimators. The results have been compared in Fig. 2. To test the dynamic
estimation algorithm (3(k) was generated with a standard deviation of 0·03.
e
Using some initial values of and v, a corrupted measurement sequence was
generated. This was processed by the proposed estimator (22) and (23) as
well as the conventional one. The results are presented in Figs. 3 and 4. It
is apparent from Figs. 2-4 that in both dynamic and static cases the statistical
linearization algorithms yield better estimates.
-- TAYLOR SUU"S
- - - - - STATIST/CAL lfNEI\RIZI\T/ON
Downloaded by [] at 12:22 31 December 2014
, 2 8 /0
'NOEX.
Figure 4. Absolute value of estimate error plot for v. using dynamic estimator.
6. Conclusions
New algorithms have been obtained for a power system state estimation
by replacing the Taylor series linearization with statistical linearization. To
overcome the increased computational effort P - e, Q - V decoupling has been
proposed. Computer simulations on a 7-node power network illustrate the
various algorithms. Comparison has been carried out with the conventional
algorithms. A rough idea of computational requirements has also been
presented.
REFERENCES
DEBS, A. K, and LARSON, R. K, 1970, I.E.E.E. Tra·ns. pwr Appar. Syst., 89, 1970.
LARSON, R. K, TINNEY, 'V. F., HAJDU, L. P., and PIERCY, D. S., 1970, I.E.E.E.
Trans. pwr Appar. Syst., 89, 353.
LARSON, R. K, TINNY, W. F., and PESCHON, J., 1970, I.E.E.E. Trans. pwr Appar.
Syst., 89, 345.
418 Power-system state estimation
MAHALANABIS, A. K., and :FAROOQ, M;, 1971, Int. J. Control, 14,631.
MASIELLO, R. D., and SCHWEPPE, F. C., 1971, I.E.E.E. Trans. pwr Appar. Syst.,
90,1025.
SCHWEPPE, F. C., 1970, I.E.E.E. Trans. pwr Appar. Syst., 89, 130.
SCHWEPPE, F. C., and DOUGLAS, B. ROM, 1970, I.E.E.E. Trans. pwr Appar. Syst.,
89, 125.
SCHWEPPE, F. C., and MASIELLO, R. D., 1970, I.E.E.E. Trans. pwr Appar. 8yst., 89,
120.
SCHWEPPE, F. C., and HANDSCHIN, E. J., 1974, Proc.l.E.E.E., 62, 72.
SHARMA, K. L. S., 1973, Ph.D. Thesis, Indian Institute of Technology, New Delhi.
Downloaded by [] at 12:22 31 December 2014
