Observability analysis for state estimation using linear

programming
I.O.Habiballah and M.R.lrving

Abstract: The linear programming technique is used to analyse the observability of power system
state estimation networks. This technique has the capability of identifying the observable islands of
the system network. It can also identify the necessary branches or nodes needed for pseudomeasurement placement to recover the overall system observability. The technique manipulates only
integer numbers. It uses the network connectivity matrix and the measurement Jacobian matrix as
equality constraints. Computational aspects of the proposed technique are presented and results of
several examples are reported.

Introduction

Observability analysis is one of the important functions

needed to run a state estimator. The state estimator for the
overall system cannot be executed unless it is observable. In
the case of unobservability, pseudo-measurements must be
added to the existing measurement set to restore observability. The problem of power system network observability
has been studied thoroughly by many researchers in the
past. There have been essentially three different approaches
presented so far [1-32]: graphical (or topological) methods,
numerical (or floating-point) methods, and symbolic methods.
The graphical methods [I-9, 17, 19-21, 251, which are of
a combinatorial nature, are based on constructing a spanning tree. The system is said to be observable if a spanning
tree of full rank can be constructed. The numerical methods [10-12, 14, 15, 23, 24, 2&31], which make use of either
the information (gain) matrix or the measurement Jacobian
matrix, are based on using the linearised real power (DC)
state estimation equations. The system is said to be algebraically observable if the matrix is of full rank. The symbolic
methods [13, 18, 291 are based on reduction of the DC
Jacobian matrix without the need for its numerical computation. Other papers [16, 22, 321 have used the combination
of graph and numerical computation methods.
In this paper, a new numerical method is introduced
based on the use of a linear programming technique. The
technique manipulates integer numbers only. It uses the
network and measurement Jacobian matrices as equality
constraints. This method has the capability of identifying
the observable islands of the system network. It can also
identify the necessary branches or nodes needed for
pseudo-measurement placement to recover the overall system observability.
0IEE, 2001
IEE Proceedhgs online no. 20010061
DO1 10.1O49/ipgtd:20010061
Paper fmt received 27th January and in revised form 13th September 2000
1.0. Habiballah is with King Fahd University of Petroleum and Minerals,
Dhahran, Saudi Arabia
M.R. Irving is with Brunel University, Uxbiidge, United Kingdom
142

Linear programming formulation

The identification of the observable islands and unobservable branches has been tested numerically using the factorisation of the gain matrix, or the orthogonal transformation
of the measurement Jacobian matrix [10-12, 14, 151. These
numerical methods require the removal of irrelevant injection measurements in an iterative manner.
An alternative numerical method using the linear programming technique is proposed. This does not require the
removal of any injection measurement, nor does it require
any iteration during the course of the solution (other than
the usual iterative solution of a linear programme).
Considering the linearised real power (or DC) model, the
error-free state estimation equations can be written as

H M X= z
where x is the (nb x 1) state vector of bus angles (including
reference bus), z is the (m x 1) real power measurement set
(comprising line-flows and bus-injections),HM the (m x nb)
measurement Jacobian matrix, nb the number of busses
and m the number of measurements.
Artificially setting each of the line parameters to one
(without loss of generality for the observability problem),
HM becomes a constant matrix. Its elements are integer
numbers that reflect the location of the measurement
meters in the system [12]. It should be noted that this formulation implies that the algorithm will analyse structural
(or topological) observability only. However, pathological
cases of networks that are structurally observable but
numerically unobservable, owing to the values of line
parameters, are not found in practice, provided that line
impedances of zero and infinity are prohibited.
Consider the network connectivity matrix HN, with nl
lines. The elements of the HN matrix are 1 and -1 for
each line connected from bus i to bus 7 , and 0 everywhere else. Assuming all the measurement values are equal
to zero (again without loss of generality), the proposed linear programming equation can be written as:
nl

min

p,
2=

+ n,

(2)

subject to
IEE Pioc -Gene1 Tiansm D I S I I , Vol
~ , 148 No 2 Mcrrcli 2001

H M X= 0
H,yx + p - n = i
The first constraint will tend to force observable islands to
have equal phase angles xi.The second constraint is
needed to identify the unobservable branches that exist
between the observable islands. The pi and ni variables are
positiveand negativeerror components, whose difference
will be equal to i, if the corresponding branch in the H N
matrix has the same phase angles, otherwise the branch is
unobservable.
Therefore, the two constraints assume initially that there
are lineflows everywhere and that the phase angles are not
the same. The technique then starts to isolate every group
of busses that have their phase angles similar to its reference bus together in one island. This island is defined as an
observable island. An island could be a single bus only. In
this case its phase angle value will be different from all
other values. The lineflows within any observable island are
equal to zero because the phase angles have the same values as their reference bus. The lineflows between the different islands are not equal to zero because every island has a
reference value different from the other islands.
The technique processes all the lineflow measurements
and almost all the injection measurements successfully and
allocates the associated busses in their correct island. There
are, however, certain injection measurements whose associated busses cannot be allocated correctly in their proper
island. The property of such an undetermined injection
measurement is that it can be connected to an observable
island through two or more unobservable branches but is
unable to make a decision as to which unobservable branch
it should use.
3

ble islands (usually an island with the largest number of

busses in it) around it and include the bus associated with it
in the selected island. Also, enforce the value of the unobservable branches associated with this selection to zero to
make them observable
8) Count the number of observable islands. If only one
observable island is identified, then the system is observable. Otherwise, continue to step 9.
9) Identify the remaining unobservable branches
10) Use the results of steps 6 to 9 to suggest the required
pseudo-measurements.
Identification of undetermined injection measurements in
step 7 can be performed as follows. Trace the unobservable
branches identified in step 6. If any is connected to a bus
with an injection measurement, then this measurement will
be classified as an undetermined injection measurement.
4

Test results

The proposed technique has been tested successfully on several examples available in the literature. Three illustrative
examples of a 7-bus system [12], a 6-bus system [14] and a
14-bus system [15] are presented here in detail. The 6-bus
system is unobservable with one critical injection measurement. The 7-bus system is unobservable with a critical
injection measurement, one zero-injection measurement,
and one irrelevant (redundant) injection measurement. The
14-bus system is initially observable with a combination of
different types of measurements, and then it will be made
unobservable by removing the injection measurement of
bus 5. The unobservable IEEE 30-bus [14] is also reported
in this section.

Algorithm for observability analysis

A network is considered to be observable, if when all the

measurement values are set to zero, all state variables (or
the phase angles) will have the same value and all branches
(or the lineflows) will equal zero [lo]. If, however, the network is not observable, then setting the measurement set
equal to zero will create different observable islands. Within
each island, the state variables will have the same value and
the branch flows will equal zero. The branches (lineflows)
between the different islands will have nonzero values and
are known as unobservable branches.
It is possible that an island may have only one node. In
this case its state variable value will be different from any
other state variables. Counting the number of islands gives
an indication of how many pseudo-measurements will be
needed to make the network observable. The unobservable
branches can be used to select the possible types and locations of the pseudo-measurements to be added.
The following algorithm describes the necessary steps for
determining the observable islands and the unobservable
branches of a power system state estimation network.
1) Establish the network connectivity matrix HN
2) Establish the measurement Jacobian matrix HM
3) Consider all the measurement values equal to zero, z = 0
4) Use eqn. 2 to solve the observability problem
5) Classify each group of state variables having the same
value as an observable island
6) Identify the unobservable branches that exist between
the observable islands
7) Identify if any undetermined injection measurements
exist. If so, select for every one of them one of the observaIEE Proc -Gene,. Transm. Distrib., Vol. 148, No. 2, March 2001

-1 -1
0
0
1 0 - 1 0
1 0
0
0
0 1 - 1 0
0
0
HN= 0 1
0 0
1 - 1
0 0
0
1
0 0
0
1
-0 0
0
0

0
0
0 0
0
0
- 1 0
0
0
0
0
-1
0
0
0
0
0
- 1 0
0
0
0 - 1
1 -1
0 -

cates five unobservable branches: (1, 2), (1, 5), (2, 3), (3, 4),
and (4, 5). The solution discards the injection measurement
at bus one.
To check the existence of any undetermined measurement, trace the unobservable branches and see whether any
is connected to an injection measurement. It can be
observed that branches (1, 2) and (1, 5) are connected to an
injection measurement at bus 1 (measurement # 5). This
means that this measurement is an undetermined injection
measurement, as it could not connect one of its unobservable branches to another observable island island (2, 5, 6).
If the unobservable branch (1, 2) is selected to connect bus
1 with the observable island (2, 5, 6), the two islands { 1, 3)
and (2, 5, 6 ) will be integrated in an island { 1, 2, 3, 5, 6)
and branches (1, 2), (1, 5) and (2, 3) become observable.
This is because their associated busses belong to the same
island, and therefore their values, according to step 6, must
be forced to zero.
Investigating the rest of the unobservable branches, (3, 4)
and (4, 5), no more undetermined injection measurements
exist. The final solution will be two observable islands: { 1,
2, 3, 5, 6}, and (4, 7}, and two unobservable branches (3,
4), and (4, 5). Therefore, we need only one pseudo-measurement: an injection at node 3, 4, or 5, or a lineflow on
branch (3, 4) or (4, 5).

at bus 2.
To check the existence of any undetermined measurement, trace the unobservable branches and see if any is
connected to an injection measurement. It can be observed
that branches (2, 3) and (2, 5) are connected to an injection
measurement at bus 2. This means that this measurement is
an undetermined injection measurement, as it could not
connect one of its unobservable branches to one of the
observable islands around it: island (5) or island (1, 3, 4,
6). According to step 6, the unobservable branch (2, 3) is
selected to include bus 2 with the observable island (1, 3, 4,
6 ) and form island (1, 2, 3, 4, 6). Branch (2, 3) becomes
observable, and its value only must be forced to zero.
Investigating the rest of the unobservable branches, (2, 5)
and (5, 6), no more undetermined injection measurements
exist.
The final solution will be two observable islands: { 1, 2, 3,
4, 6}, and {5}, and two unobservable branches: (2, 5) and
(5, 6). Therefore, we need only one pseudo-measurement:
an injection at bus 5 or 6, or a lineflow on branch (2, 5) or
(5, 6). It can be observed that the unobservable branch (2,
5) is now connected to a normal injection measurement at
bus 2 which has been already used to connect another
branch (2, 3).

4.3

14-bus example

The network and measurement set of the 14-bus network is

shown in Fig. 3.

5 '

-+-

line-flow measurement

bus-injection measurement

I \

zero-injection measurement

Fig.2 6-his network exumple

-1 0
1 0
0 1 HN= 0 1
0 0
0 0
-0 0

0
0
1

-1

0
0

0
0
1 - 1
0
1
0
0

0
0
0
-1
0
0
1

0 -1
0
0
0
-1

-1-

0 0 1 - 1 0
2 0 0 -1
0 -1
O I
0 2 -1
0 -1
-1 0 -1
3
0 -1
The solution of the optimisation problem before treating
the undetermined injection measurements gives three
observable islands, (1, 3, 4, 6}, {2), and {5}, and indicates
three unobservable branches, (2, 3), (2, 5), and (5, 6). The
solution considered the redundant injection measurements
at busses 1 and 2 and discarded the injection measurement

H,=

144

Fig.3 14-busnetwork exumple

The solution of the optimisation problem before removing the injection measurement on bus 5 confirms that the
system is observable, i.e. one observable island and no
unobservable branches.
If we remove the injection measurement on bus 5, the
system becomes unobservable. The solution of the optimisation problem gives seven observable islands: { 1-5, 7-9},
{6), {lo}, {ll}, {12}, (13) and {14}, and indicates nine
unobservable branches: (5, 6), (6, lo), (6, 12), (6, 13), (9,
ll), (9, 14), (10, ll), (12, 13) and (13, 14).
Tracing the unobservable branches, measurements on
busses 6, 9, 11, 12, and 13 are identified as undetermined.
Selecting the unobservable branches (6, 12), (9, 1l), (9, 14),
(12, 13) and (13, 14) will reduce the number of observable
islands from seven to only two: {IO}, and {remaining
busses} and the number of unobservable islands from nine
to only two: (6, lo), and (10, 11). Therefore, we need only
one pseudo-measurement: an injection at bus 10, or a lineflow on branch (6, 10) or (10, 11).
IEE Proc -Gener Transm Distrib , Vol 148, No 2, Mnrrh 2001

4.4 I 30-bus system

The measurement set of the IEEE 30-bus system is given in
[ 141. The proposed method for observability analysis has
been tested extensively. Three observable islands have been
obtained: {13}, {18, 19, 201, and {remaining busses}. The
unobservable branches (not presented in [14]) are (12, 13),
(15, l8), and (10, 20). This system has no undetermined
injection measurements (because none of the busses associated with the unobservable branches has an injection measurement). The results confirm those of [14]. Therefore, we
need two pseudo-measurements.
The first pseudo-measurement is to connect bus 13 to the
rest of the system. This can be accomplished either by adding an injection measurement at bus 12 or 13, or a line-flow
measurement on the branch (12, 13). The second pseudomeasurement is to connect busses 18, 19 and 20 to the rest
of the system. This can be accomplished either by adding
an injection measurement at bus 10, 15, 18, or 20, or a lineflow measurement on the branch (15, 18) or (10, 20).
5

Conclusions

The observability of power system state estimation networks has been analysed using a linear programming technique. This technique has been shown to have the
capability of identifying the observable islands and unobservable branches of a power system network. It can therefore automatically identify the necessary branches or nodes
needed for pseudo-measurements placement to recover the
overall system observability. It uses the network connectivity matrix and the measurement Jacobian matrix as equality constraints. The proposed observability method has
been tested successfully on several systems.
6

Acknowledgments

The authors acknowledge the support of King Fahd University of Petroleum & Minerals in Saudi Arabia, the British Council Summer Program in Saudi Arabia, and Brunel
University in the United Kingdom.
7

References

1 SASSON, A.M., EHFFNANN, S.T., LYNCH, P., and VANSLYCK, L.S.: Automatic power system network topology determination,
IEEE Trans. Power Appar. Syst., 1973, PAS-92, pp. 61M18
2 FETZER, E.E., and ANDERSON, P.M.: Obseiyability in the state
estimation of power systems, IEEE Truns. Power Appur Syst., 1975,
PAS-94, pp. 1981-1988
3 CLEMENTS, K.A., and WOLLENBERG, B.F.: An algorithm for
observability determination in power system state estimation. Proceedings of IEEE PES Summer Meeting, 1975, Paper A75, pp. 447443
4 KRUMPHOLZ, G.R., CLEMENTS, K.A., and DAVIS, P.W.:
Power system observability: a practical algorithm using network
topology, IEEE Truns. Power Appar. Syst., 1980, PAS-99, (4), pp.
1534-1542
5 FLEMENTS, K.A., KRUMPHOLZ, G.R., and DAVIS, P.W.:
Power system state estimation residual analysis: an algorithm using
network topology, IEEE Truns. Powe? Appar. Syst., 1981, 100, (4),
pp. 1779-1787
6 QUINTANA, V.H., SIMOES-COSTA, A., and MANDEL, A.:
Power system topological observability using a direct graph theoretic
approach, IEEE Truns. Power Appur. Syst., 1982, PAS-101, (3), pp.
617426
7 FLEMENTS, K.A., KRUMPHOLZ, G.R., and DAVIS, P.W.:
Power system state estimation with measurement deficiency: an algorithm that determines the maximal observable sub-iietwork, IEEE
Trans. Power Appar. Syst., 1982, 101, (9), pp. 30443052

IEE Proc -Genet. Trcmsm. Disrrib.. Vol. 148. No. 2,Mrrrch 2001

8 FLEMENTS, K.A., KRUMPHOLZ, G.R., and DAVIS, P.W.:

Power system state estimation with measurement deficiency: an
observabilityimeasurementplacement algorithm, IEEE Truns. Power
A p p ~ rSyst.,
.
1983, PAS-102, pp. 2012-2020
9 VAN CUTSEM. T.. and GAILY. P.J.: A simole algorithm for oower
system observadility analysis and related fuictionsu. Proceediigs of
CIGRE/lFAC, Florence, Sep. 1983, pp. 2628
10 MONTICELLI, A., and WU, F.F.: Network observability: identification of observability islands and measurement placement, IEEE
Trans. Power Appar. Syst., 1985,104, (3, pp. 1035-1041
11 MONTICELLI, A., and WU, F.F.: Network observability: theory,
IEEE Trans. Power Appur. Syst., 1985,104, (S), pp. 1042-1048
12 MONTICELLI, A., and WU, F.F.: Observability analysis for
orthogonal transformation based state estimation, IEEE Trans. Power
Syst., 1986, 1, (I), pp. 201-208
13 SLUTSKER, I.W., and SCUDDER, J.M.: Network observability
analysis through measurement Jacobian matrix reduction, IEEE
Truns. Power Syst., 1987, 2, (2), pp. 331-337
14 WU, F.F., LIU, W.E., and LUN, S.: Observability analysis and bad
data processing for state estimation with equality constraints, IEEE
Trans. Power Syst., 1988, 3, (2), pp. 541-548
15 WU, F.F., LIU, W.E., HOLTEN, L., GJELSVIK, A., and AAM, S.:
Observability analysis and bad data processing for state estimation
using Hachtels augmented matrix method, IEEE Trans. Power Syst.,
1988, 3, (2), pp. 604-61 1
16 CONTAXIS,G.C., and KORRES,G.N.: A reduced model for
power system observability analysis and restoration, IEEE Truns.
Power Syst., 1988, 3, (4), pp. 1411-1417
17 CLEWER, B.C., IRVING, M.R., and STERLING, M.J.H.: Topologically independent state estimation. Proceedings of the IFAC symposium on Power systems, modeling and control cpplicutions, 5-8 Sep.
1988, (Brussels),pp. 17.1.1-17.1.13
18 CHEN, R.: A fast integer algorithm for observability analysis using
network topology, IEEE Trans. Power Syst., 1990, 5, (3), pp. 10011009
19 CRAINIC, E.D.,
HORISBERGER, F.L.P., DO, X.D., and
MUKHEDKAR, D.: Power network observability: the assessment of
the measurement system strength, IEEE Truns. Power Syst., 1990, 5,
(4), pp, 1267-1285
20 NUCERA, R.R., and GILLES, M.L.: Observability analysis: a new
topological algorithm, ZEEE Truns. Power Syst., 1991, 6, (2), pp. 466475
21 MORI, H., and TSUZULD, S.: A fast method for topological
observability analysis using a minimum spanning tree technique,
IEEE Truns. Power Syst., 1991,6, (2), pp. 491499
22 KORRES. G.N.. and CONTAXIS. G.C.: Identification and uodatiiig of mihmally dependent sets of measurements in state estimation,
IEEE Trans. Power Syst.. 1991, 6, (3), pp. 999-1005
23 MONTICELLI, A., GARCIA, A.V., and SLUTSKER, I.W.: Handling discardable measurements in power system state estimation,
IEEE Trans. Power Syst., 1992,7, (3), pp. 1333-1339
24 FALCAO, D., and ARIAS, M.: State estimation and observability
analysis based on echelon forms of the linearized measurement models, IEEE Truns. Power Syst., 1994, 9, (2), pp. 979-987
25 SLUTSKER, I., and MOKHTARI, S.: Comprehensive estimation in
oower svstems: state. toooloev. and oarameter estimation. Proceedlngs of <he American Pdwer ucbnference, Chicago, Illinois, April 1995,
Vol. 57-1. nn. 149-155
26 BRETAS,G.G.: Network obseiyability: theory and algorithms based
on triangular factorization and path graph concepts, ZEE Proc.,
Gener. Tmnsm. Distrib., 1996, 143, (I), pp. 123-128
27 BRETAS, N.G.: Network observability. a theory based on triangular
factorizatioii and path graph concepts - part 1. Proceedings of international conference on Autonzatic control, Cidade do Porto, Portugal,
September 1996
28 BRETAS, N.G., DEO, E.C., and MORI, F.N.: Network observability: algorithms based on triangular factorization and path graph concepts - part 11. Proceedings of intemational conference on Automutic
control, Cidade do Porto, Portugal, September 1996
29 BRETAS, N.G., and LONDON, J.B.A.: Network observability: the
critical measurement identification using the symbolic Jacobian
matrix. Proceedings of intemational conference on Power system technology, POWERCON98, Beijing, China, 18-21 August 1998, pp.
1222-1226
30 EXPOSITO, A.G., and ABUR, A.: Generalized observability analysis and measurement classification, IEEE Truns. Power Syst., 1998,
13, (3), pp. 1090-1095
31 MORI, H., and SONE, Y.: Tabu search based meter placement for
topoligical observability in power system state estimation. Proceedings
of IEEE T&D99, April 1999, Vol. 2, pp. 603408 (1)
32 BARGIELA, A., IRVING, M.R., and STERLING, M.J.H.:
Observability determination in power system state estimation using a
network flow technique, IEEE Trans. Power Syst., 1986, 1, (l), pp.
108-1 14

145