Documente Academic
Documente Profesional
Documente Cultură
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
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
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.
-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
5 '
-+-
line-flow measurement
bus-injection measurement
I \
zero-injection measurement
-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
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
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
145