Approach to automatic contingency selection by

reactive type performance index

J.Z. Zhu, PhD

Prof. G.Y. Xu, PhD

Indexing terms: Mathematical techniques, Power system security

of contingencies. One is the line MW overloads, and the

Abstract: The paper represents a novel approach other is the bus voltage limit violation. The former is
to studying the automatic contingency selection called ‘MW type PI’, and the latter is called ‘MVAR type
(ACS). The total curtailment of the reactive loads PI’. Various MW ranking methods, which rank the con-
is used as the performance index to evaluate the tingencies on the basis of the severity of line MW flow
severity of contingencies. The minimum cost limit violation, have been developed [1-81. The voltage
network flow model is proposed to determine the ranking method, which ranks the contingencies on the
optimal curtailment in the line outage case. The basis of the severity of bus voltage and/or bus MVAR
objective of the model is the determination of the generation limit violation, is a subject of much current
sum of branch voltage drops in terms of the line research [7-111.
reactive flows. The optimisation model is solved Most of the computation algorithms for computing PI
by the ‘out-of-kilter algorithm’. The numerical are based on AC, fast decoupled or DC power flow
results for several IEEE sample systems show that analysis. The branch outage is simulated by the sensi-
the proposed technique is feasible and effective. tivity method (the sensitivities of PI with respect to
changes in susceptance of the line under contingency), the
distribution factor or compensation method. These are
1 Introduction likely to be more complicated than the network flow
technique.
The security assessment of power systems has received a It is well known that network flow programming
great deal of attention fronl utilities in both operation (NFP) is an efficient method for solving the transporta-
and planning. The failure of one or more transmission tion problem with minimum cost. The out-of-kilter algo-
facilities and/or generation units leads to line overloads rithm (OKA) is a specialised form of NFP. This
and abnormal bus voltages in the systems. Such critical algorithm is characterised by simple manipulation and
cases should be quickly identified for possible corrective fast convergency. In addition, the method is simpler and
rescheduling strategy in operation, or for improving the more intuitive in simulating the single outage case of a
design in planning. Therefore the steady state security generator, a transmission line, and the secondary outage
analysis must be carried out. The traditional procedure of of a power system. The method can therefore be used in
the security analysis for examining all possible contin- contingency selection. In a companion paper [6], the
gencies sequentially using AC power flow is very time application of the OKA to line MW overload problems
consuming. Although the fast computational techniques, in ACS is discussed. The present paper suggests and
such as the fast decoupled or linearised AC power flow, focusses on an optimisation model of the reactive power
have well been developed, the computational burden for flow for ACS. The objective of this model is to obtain the
such analysis is still prohibitive for a large scale power sum of the branch voltage drops, expressed by reactive
system, especially in real-time security analysis. In addi- power flow instead of node voltage deviations. The total
tion, such analysis techniques would be inefficient, as reactive load curtailment is used as a performance index
most of the contingencies do not result in system prob- to measure the severity of contingency for filtering.
lems. Naturally, the development of a contingency selec-
tion and ranking technique is desirable, as this would 2 Performance index
screen the critical contingencies out of all probable con-
tingencies and rank them based on their relative severity The precontingency state of the system implies that all
without performing a full power flow for each contin- branch loadings and all bus voltages are within their
gency. The essential requirements of the technique are the limits at a given loading condition. If one or more system
accuracy of the ranking and the efficiency of the compu- components have failed, it may lead to abnormal bus
tation. Both the accuracy and efficiency are affected by voltage and line overload in the systems. The new state is
the definition of the performance index PI and the com- called postcontingency state.
putational technique for computing its parameters, In the real-time static security analysis of power
including outage simulation. systems, it is reasonable to find a simple, fast but approx-
From the viewpoint of the impact, there are two kinds imate algorithm that can filter a large amount of ‘slight’
of performance indices proposed to evaluate the severity contingencies. A few serious contingencies and their cor-
responding constraints to perform precise calculations
Paper 7568C (P9), first received 16th January and in revised form 13th are thus retained.
June 1990 The first problem of the automatic contingency selec-
The authors are with the Department of Electrical Engineering, Chong- tion and ranking is to define the performance index for
qing University, Chongqing, Sichuan, People’s Republic of China measuring the relative severity of contingencies correctly.
IEE PROCEEDINGS-C, Vol. 138, NO. I , J A N U A R Y 1991 65

Authorized licensed use limited to: PONTIFICIA UNIVERSIDADE CATOLICA DO RIO DE JANEIRO. Downloaded on November 27, 2009 at 11:03 from IEEE Xplore. Restrictions apply.
 As a contingency that causes a line MW overload M - 1 is a specialised linear programming model.
problem may not necessarily cause a bus voltage According to the dual theory, we can derive the condi-
problem, and vice versa, both problems should be studied tions for optimality. For convenience, we define the rela-
simultaneously. Two different rankings can be carried out tive cost Cij = C i j + ni - n j , where n is the dual variable
separately, based on the decoupling of the real and reac- (node parameter). Then, the complementary slackness
tive power in power systems. The existing methods for conditions for optimality [6] can be written as
solving these two problems are quite different. To form a
unified ranking model and algorithm with Reference, 6, = L i j for Cij > O (5)
we use the same form of the performance index to study Cij = O for Lij <fij < Uij (6)
the voltage ranking method.
Let Qg be the total reactive load in the precontingency f i j = U i j for Cij < O (7)
state, which is the sum of reactive power in all load Eqns. 5-7 correspond to three ‘in-kilter’ states of arc. In
branches. In the branch outage, if all the reactive loads addition, there exist six’ out-of-kilter’ states that do not
remain at precontingency values, the voltage at some satisfy eqns. 5-7. If all arcs are in kilter, then the optimal
buses and the reactive flow in some branches (which are solution is obtained; otherwise, we should vary the flows
associated with bus voltages) could violate their security or potentials (node parameter n) by a labelling technique
constraints. In other words, if the contraints are satisfied, to bring the out-of-kilter states into in-kilter states of arc.
the reactive power supply may be limited. Consequently, In comparison with a general algorithm of the
the reactive load may be curtailed or additional reactive minimum cost flow, the features of OKA are:
sources (local compensation) may need to be installed. (i) the lower bound of flows may be nonzero
The more severe the contingency is, the greater the (ii) the initial flow does not need to be a feasible or
amount the load needs to be curtailed. Hence we define a zero flow
new performance index using the total reactive load cur- (iii) non-negative constraints,hj 2 0, are not necessary
tailment to evaluate the relative severity of continency, (iv) it is easy to imitate the change of the network
i.e. topology, by changing the values of flow bound.
Qg - QDK(I)
PI&) = x 100% (1) 3.2 Network model of reactive power
Q: A power system is a particular network which consists of
where Q D K ( I ) is the reactive power in load branch k as generation, transmission and load arcs. The reactive
line I outage appears, the value of which is calculated power optimisation model for the contingency selection
using the reactive optimisation model (see Section 3.2). is proposed as follows:
To build a list of ACS, an appropriate threshold value
of P I , need be stipulated (e.g. 7%). Thus the slight con- minF,= IAV,l
jsNT
tingencies are filtered.
The form of eqn. 1 is the same as PI,(() (for real types) such that
i-w
Qci + j-w
QTj
in Reference 6.
+ C Q D k = O w e n , k ~ N D (9)
k-rw
3 Mathematical model
M - 2< -
Qci d QGi d Qci i E N G ( i # N L ) (10)
3.1 Brief statement of OKA model
For a network with n nodes and m arcs of a minimum -
QTj QTj QTj j E N T ( j # N L ) (11)
cost flow problem, we can transform the original network
into an OKA network by introducing a ‘return arc’ from AV, d AV,,,, (12)
sink node t to source node s, and the internal flows QDK
- d QDK d QDK k E N D (k # N L ) (13)
remain unchanged. The return arc flowf,, equals the orig-
inal network flow r. The OKA network model is shown
in Fig. 1, and its mathematical model [12, 131 can be
written as where, Q G i , QDk are the reactive power of generation
(concluding compensation capacitor), transmission and
min F = CijAj load branch, respectively; i, j , k are their corresponding
ijsm+l
subscript. The positive direction of reactive power flows
M -
{
1 such that c(fij
ien
-fji) =0 i E n (3) is specified when the power enters the node, while the
negative is specifed when the power leaves the node. N G ,
(Lijdfij d uij i j ~ m1+ (4) N T and N D represent the corresponding number of the
three arcs, i.e. the set of indices i , j , k . The total number of
nodes is n. Q and Q are the lower and upper bound of the
original network
(n nodes m arcs)
reactive power, respectively. The symbol i + w represents
that branch i is adjacent to node w . N L represents the set
of outage branches. I AV, I is the absolute value of voltage
=
,f
-r
A drop in the line j .
Fig. 1 O K A network model The objective function F , is the sum of branch voltage
drops that are associated with, and hence reflect, the
whereLj is the flow in arc ij, Lij and U i j is its lower and node voltage deviations. The voltage drop in branch j
upper bound, respectively; C i j is the arc cost per unit can be expressed by [141
flow; F is a cost function; n is the total number of nodes;
m is the total number of arcs (or branches) in the return
arc; i , j are the terminals of line i j , respectively.
66 I E E PROCEEDINGS-C, Vol. 138, NO. 1, J A N U A R Y I991

Authorized licensed use limited to: PONTIFICIA UNIVERSIDADE CATOLICA DO RIO DE JANEIRO. Downloaded on November 27, 2009 at 11:03 from IEEE Xplore. Restrictions apply.
 where R j , X j , P T j and Q T j is the resistance, reactance, revised ; using the labelling technique, when an incremen-
real a d reactive power flow in branch j , respectively; VTj tal loop of flow exists the arc flow fij in this loop is
is the rated voltage of line j and can be set equal to 1 in varied; if there is no incremental loop, the values of n at
P.u.; then eqn. 15 can be written as unlabelled nodes, are varied, and cij
is changed; once the
arc state has been revised, go back to step (2).
A Y = P T jR j + Q T j X j (16)
for reactive power optimisation, the real power P , and 4.2 Analogue of branch outage
parameter R j are kept constant. Thus, to minimise The outage simulation is an important problem in ACS.
-
the objective xjcNT I A 5 1 is equivalent to minimising In general, the outages of the generating units or loss of
loads need a modification of the bus injection vector. For
L j e N T IQTjXjl.
Eqn. 9 is the reactive power balance equation for each the line outage case, the base-case admittance matrix has
node: Eqns. 1&13 are the inequality constraints of the to be modified. In network flow programming, it is pos-
corresponding variables. The inequality constraint of the sible to imitate the contingencies in the same manner for
reactive power flow in the return arc in an O K A network various outages, depending on whether a load or a gener-
is given in eqn. 14. Eqn. 14 is redundant, as it is incorpo- ating unit is regarded as a branch or a transmission line.
If an element is in outage, the flow in the corresponding
rated in eqn. 13.
On deleting the P T jR j component in the voltage drop, branch is zero. It is obtained by setting the upper and
eqn. 16, the inequalities of eqn. 11 and 12 can be com- lower bound of the flow in that line to zero. The imita-
bined. Hence we have tion of the multiple branch outage is treated in the same
way. Hence, this is a more straightforward and easier
method than the general methods.
4.3 Selection of factor k,
It should be pointed out that the voltage drops in It is very important to select the factor kj as it effects the
branches differ in value significantly for different configu- proper filtering and ranking of contingencies.
rations and R / X ratios of the networks as well as the The value of kj is obtained by the following experi-
performance of the power systems. It is quite difficult to mental formula:
specify a value of AY,,, for every branch which can
properly reflect the node voltage deviations. The values
of AY,,, determine the feasible domain of the opti-
kj = J(3)
misation problem directly. If the value of AF,,, is too
large, it could lead to the filtering out of some critical
-
Here, AY,,, can be specified to be 4 5%. If AV: z 0,
we assume AV: = E (a small positive value) to calculate
contingency, caused by a wide band of the reactive power factor k j .
flow restriction. O n the other hand, if AV,,,, is set too As an exception, in the radial network with homoge-
small, then some 'no problem' contingencies could not be neous line ( R / X = constant) and almost the same power
filtered out. For the reasons mentioned above, instead of
AY,,,, we adopt the voltage drop AV: in branch j ,
based on the power flow of the normal state multiplied
-
factor (cos $), we use AV,,,, directly, and set its value
equal to 1 3% in an approximate calculation for con-
tingency selection.
by a factor kj i.e.
4.4 Numerical examples
AV,,,, = kjAVg (18) The proposed ACS technique, including the network flow
The optimisation model M - 2 can be rewritten as model and its algorithm, are examined with IEEE 5-bus,
14-bus, 30-bus, 57-bus and 118-bus test systems on
M-340. The value of AV,,,, for calculating factor kj is
5%. The threshold value of the performance index P I ,
for filtering is 7%. The numerical results solved by the
O K A are given in Tables 1-3.

(eqns. 9, 10, 13, 14

Table 1 : 5-bus system contingency ranking list
The linear model M - 3 corresponds to the O K A
network model M - 1. Hence, the optimal reactive Line i-JAQ,, PI, ( Y o ) Rank Rank [8]
power flow model can be solved by the out-of-kilter algo- 4 3-2 0.1335 22.25 1 1
rithm. 2 4-1 0.0800 13.33 2 2
5 2-5 0.0735 12.25 3 3
3 5-1 0.0700 11.67 4 4
4 Implementation and numerical example 1 1-3 0.0535 8.90 5 5
6 3-4 0.0525 8.75 6 6
4.1 Major procedures of the OKA
The essence of the O K A is to revise the out-of-kilter Table 2: 14-bus system ACS list
states of arcs into in-kilter states, according to comple-
mentary slackness conditions for optimality. The major Branch AQD PI, (%) Rank AC Rank
procedures are summarised as follows: 8-7 0.3091 38.0 1 1
(1) set the initial value of arc flows; the initial flow has 6-5 0.1548 19.0 2 2
to satisfy eqn. 3 only, regardness of eqn. 4 1-5 0.1504 18.5 3 3
6-13 0.1152 14.2 4 4
(2) check the state of arcs; if all arcs are in-kilter, the 9-14 0.0690 8.5 5 5
optimal solution has been found, thus terminate the iter- 1-2 0.0670 8.2 6 6
ation process; otherwise, go to step (3) 9-10 0.0657 8.1 7 7
(3) revise the state of arcs; arbitrarily choose an arc 6-11 0.0635 7.8 8 8
3-4 0.0617 7.6 9 9
from the arc set which is in the out-of-kilter state to be
IEE PROCEEDINGS-C, Vol. 138, N o . I , J A N U A R Y 1991 61

Authorized licensed use limited to: PONTIFICIA UNIVERSIDADE CATOLICA DO RIO DE JANEIRO. Downloaded on November 27, 2009 at 11:03 from IEEE Xplore. Restrictions apply.
 Table 3: Computational time of several systems China for the financial assistance in support of this
project.
~~

System 5-bus 14-bus 30-bus 57-bus 118-bus

Total branches 6 20 41 78 179
CPU time(s) 0.4 0.88 2.54 5.47 65.72 7 References
Computer type M-340
1 EJEBE, G.C., and WOLLENBERG, B.F.: ‘Automatic contingency
selection’, IEEE Trans., 1979, PAS-98, pp: 77-109
For comparison, the ranking result for the 5-bus 2 VEMURI, S., and USHER, R.E.: ‘On-line automatic contingency
system in Reference 8, which is a more accurate method selection algorithm’, IEEE Trans., 1983, PAS-102, pp. 3 4 6 3 5 4
[ll], is listed in Table 1. The result of the contingency 3 JONDEA, K., GAMAL, M., and WONG, V.: ‘Contingency ranking
selection and ranking for the 14-bus system, based on and simulation for on-line use’, IEEE Trans., 1985, PAS-104, pp.
exact AC power flow, is shown in Table 2. It is seen that 2401-2407
4 FARRAG, M.A., and ATTIA, H.A.: ‘Fast and exact method for line
the results are coincident with each other. Therefore, the outage studies in power systems’, Electric Power Systems Research,
accuracy of the proposed method is feasible and satisfac- 1988,15, pp. 181-187
tory. 5 LO, K.L., BISMIL, M.A., McCOLL, R.D., and MOFFATT, A.M.:
‘A comparison of MW ranking methods’, Electric Power Systems
Research, 1988,15, pp. 157-171
5 Conclusion
6 ZHU, J., and XU, G.: ‘Application of out-of-kilter algorithm to
A new fast optimal reactive power flow model, namely a automatic contingency selection and ranking’. Proceeding of Inter-
network flow model is proposed, in which the objective is national Symposium on Engineering Mathmatics and Applications,
ISEMA-88, July 1988, China, pp. 301-305
the calculation of the total voltage drops in the branches 7 ALBUYEH, F., BOSE, A., and HEATH, B.: ‘Reactive power con-
expressed by the reactive power flow. The network flow siderations in automatic contingency selection’, IEEE Trans., 1982,
model is solved by the out-of-kilter algorithm. The total PAS-101, pp. 107-1 12
reactive power load curtailment (i.e. compensation) is 8 WASLEY, R.G., and DANESHDOOST, M.: ‘Identification and
ranking of critical contingencies in dependent variable space’, IEEE
selected as a performance index to measure the severity Trans., 1983, PAS-102, pp. 881-892
of contingencies. The numerical results of several sample 9 NARA, K., TANAKA, K., SHOULTS, R.R., CHEN, M.S., and
systems show that the proposed ACS technique, includ- OLINDA, P.V.: ‘On-line contingency selection algorithm for
ing the model and its algorithm, is feasible and effective. voltage security analysis’, IEEE Trans., 1985, PAS-104, pp. 847-855
It can be used in real-time security analysis. In addition, 10 DABBAGHCHI, I., and IRISARRI, G.: ‘AEP automatic contin-
gency selector; branch outage impacts on load bus voltage profile’,
the performance index and optimisation model has the IEEE Trans., 1986, PWRS-1, pp. 3 7 4 5
same form as the optimal real power model used in ACS 1 1 LO, K.L., BISMIL, M.A., McCOLL, R.D., and MOFFATT, A.M.:
[SI. By taking advantage of this feature, it is possible to ‘A comparison of voltage ranking methods’, Electric Power Systems
form a unified model and algorithm for the P- and Q- Research, 1989,16, pp. 127-140
12 PHILLIPS, D.T., JENSEN, P.A. : ‘The out-of-kilter algorithm’,
subproblem in ACS. Industrial Engineering, 1974,6, pp. 36-44
13 JENSEN, P.A., and BARNES, J.W.: ‘Network flow programming’
6 Acknowledgment (John Wiley & Sons, Inc. 1980)
14 MAGRATH, I.J., and KOTHARI, D.P.: ‘Modern power system
The authors thank The Scientific and Technical Manage- analysis’ (Tata McGraw-Hill Publishing Company Limited, New
ment Center of the National Education Committee of Delhi, 1980)

68 IEE PROCEEDINGS-C, Vol. 138, N O . I , J A N U A R Y 1991

Authorized licensed use limited to: PONTIFICIA UNIVERSIDADE CATOLICA DO RIO DE JANEIRO. Downloaded on November 27, 2009 at 11:03 from IEEE Xplore. Restrictions apply.