930 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO.

3, AUGUST 2000

A Homogeneous Linear Programming Algorithm for

the Security Constrained Economic Dispatch Problem
Rabih A. Jabr, Alun H. Coonick, and Brian J. Cory

Abstract—This paper presents a study of the simplified homo- Recently, interior point (IP) methods have been applied to
geneous and self-dual (SHSD) linear programming (LP) interior the solution of the SCED problem. Lu et al.applied the primal
point algorithm applied to the security constrained economic dis- affine-scaling method to the relief of network overloads by ac-
patch (SCED) problem. Unlike other interior point SCED applica-
tions [1]–[3] that consider only the N security problem, this paper tive power controls [8]. Vargas et al. [1] used the dual affine
considers both (N-1) and (N-2) network security conditions. An im- scaling IP method to solve the SCED with N security and trans-
portant feature of the optimizing interior point LP algorithm is mission losses using the SLP. Yan and Quintana [2] studied the
that it can detect infeasibility of the SCED problem reliably. This primal–dual logarithmic-barrier IP method for the SLP solution
feature is particularly important in SCED applications since line of the SCED problem with N security. Later, the same authors
overloading following a contingency often results in an infeasible
schedule. The proposed method is demonstrated on the IEEE 24 showed that the predictor corrector interior point method (PCIP)
bus test system and a practical 175 bus network. A comparison is is superior to the method in [2] when used to solve the same
carried out with the predictor-corrector interior point algorithm problem set [3].
for the SCED problem presented in [3]. The proposed solution algorithm improves on the IP methods
Index Terms—Double outages, homogeneous interior point for SCED because of the following features. Firstly it deals with
method, infeasible problem, security constraints, single outages. network (N-1) and (N-2) security, rather than the simple N secu-
rity. Secondly, the SHSD optimization method [9] used is less
I. INTRODUCTION sensitive than the other methods to the selection of the initial
point. Numerical experience shows that the algorithm can al-

S ECURITY dispatch refers to the case when the system is

dispatched such that it is in the normal secure state relative
to a pre-specified contingency list. In other words, for any con-
ways be successfully started from a simple well-centered ini-
tial point. This eliminates the need for special heuristics to get
a good starting solution when using primal dual IP methods
tingency belonging to this list, the redistribution of power flows [2], [3] and big-M schemes in the affine-scaling algorithms [1].
and voltages in the network, following a contingency, will re- Thirdly, it offers an elegant and reliable approach to detect if
sult in a normal system state [4]. In the event of a failure, secu- the solution is infeasible or unbounded. Solution infeasibility
rity dispatch ensures continuity of supply within the specified may arise due to changes in network topology following a trans-
voltage levels and frequency, and without violating rated limits mission line contingency. The disadvantage of the proposed IP
on any power equipment. method is that the dimension of the Newton equation system
The problem can be simplified if the AC load flow equations solved by the SHSD algorithm is increased by 1.
are replaced by the DC approximation, which assumes that all This paper develops the SCED algorithm such that contin-
voltages remain at their nominal values (i.e. at 1 pu) and all an- gency constraints are added through iterative search. The op-
gular differences are small. Following this approximation, eco- timization at each stage is done through the SHSD algorithm.
nomic dispatching with N security refers to the case where no Comparisons with the PCIP [3] method indicates the superiority
line exceeds its pre-fault limit in the intact network case. The of the proposed algorithm.
ED with (N-1) and (N-2) security indicates that the line flows
will not violate their post-fault limits following any single or II. PROBLEM FORMULATION
double line outage. The SCED problem considered deals with minimizing the
Berry and Dunnet [5] proposed a SCED algorithm for trans- total generating cost subject to satisfying the power balance, the
mission planning, which caters for both single and double line pre-outage, and the post-outage network constraints.
outages. The solution was obtained using sequential linear pro-
gramming (SLP) and the sparse dual revised simplex algorithm. A. Pre-Outage Network
A network flow model solved with the out-of-kilter algorithm The power balance can be expressed as:
was used by Zhu and Xu [6] for the solution of the ED with N
and (N-1) security. Zhu et al. [7] proposed a nonlinear convex (1)
network flow-programming which improves on the model in [6]
where
by including transmission losses.
= power generated at bus [MW]
= demand at bus [MW]
Manuscript received March 25, 1999; revised July 16, 1999. = generator bus
The authors are with the Department of Electrical and Electronic Engineering,
Imperial College, London SW7 2BT, UK. = slack bus
Publisher Item Identifier S 0885-8950(00)07724-5. = any bus carrying load.
0885–8950/00$10.00 © 2000 IEEE
Authorized licensed use limited to: Universitas Indonesia. Downloaded on September 17,2023 at 12:22:36 UTC from IEEE Xplore. Restrictions apply.
JABR et al.: HOMOGENEOUS LINEAR PROGRAMMING ALGORITHM 931

The cost associated with the total generation is given by: system data [4]. Moreover, since the presented method is in-
(2) tended to be used in planning scenarios, high accuracy is not
required. Similar assumptions are currently adopted in the ES-
Based on the rules of power pool operation in England and CORT program [18], [5] used by the National Grid Company
Wales, the generator cost curve is assumed to be convex and (NGC) for solving cost-based problems in the power pool of
piecewise-linear. It can therefore be expressed using separable England and Wales. The above assumptions can be modeled as
programming [10] as: follows:
(8)
where is the contingent state vector and is the change in
the matrix due to the outage of lines ( is the
number of simultaneous outages). For network (N-1) and (N-2)
security, takes the value of 1 and 2, respectively. Given that
is a diagonal matrix of the line susceptances on outage, is
(3) obtained from [4]:
.. ..
where . .
= th break-point of generator on the genera-
tion level axis [MW].
= cost rate of generation of generator at the th
break point [sterling £/1/2 hr]. (9)
= cost rate of generation of generator at level
[sterling £/1/2 hr]. Using the Sherman–Morrison–Woodbury formula, and pro-
= interpolatory variable of generator at the th vided that ( ) remains nonsingular, the contingent state
break-point. vector can be obtained in terms of the intact state vector as
The analysis that follows is assumed to be in the pu system. ( represents the identity matrix):
The network constraints are modeled using a DC load flow ap- (10)
proximation [4]:
Similar to Eqs. (6) & (7), the post-outage line flows are con-
(4) strained to stay less than the post-outage line rating vector :
where
(11)
.. .. .. 1) Reduction of the Constraint Set: In many cases the im-
. . . pact of an outage on other lines is sufficiently weak so that if the
pre-outage flow on line – is less than , then the post-outage
= number of buses (die slack bus is number 1). flow can not exceed ;( ) for any permitted
Let denote the reduced line-bus incidence matrix. Also, flow on the outaged line [5]. Once this has been established,
let be the diagonal matrix of negative line susceptances, then these post-outage flows need not be considered any more. An
we can write (the superscript is for transpose): inequality test can be used to determine if a constraint corre-
(5) sponding to a particular post-outage flow can never become ac-
tive. Note that ( ) which is the ratio of the post-fault to
The vector of real power flows can be expressed as: pre-fault line rating should be specified as part of the system
(6) data. The security factor ( ) needs to be varied depending
Therefore, the pre-outage line limit constraints are: on previous network flows and experience. Temperature esti-
mation of the cables can also be used by control engineers to
(7)
estimate the capacity of the line in the normal state and the per-
where is a vector of pre-outage line ratings. mitted overload which can be sustained typically for a period of
half an hour.
B. Post-Outage Network From Eq. (11), the change in power on line –
For transmission line or transformer outages, it is assumed is given by ( is the line susceptance):
that all control variables ( ) and specified variables
( ) remain unchanged. Obviously, this
line-outage model is only approximate due to changes in voltage
and power flow conditions following the loss of a major trans-
mission line. Only with an adequate load model for each bus can (12)
one improve the accuracy of the contingency analysis. However,
the errors introduced by the above mentioned modeling assump- If the pre-outage flows are within the pre-outage flow limits,
tions are in most cases far less than those due to inaccuracies of we have:
Authorized licensed use limited to: Universitas Indonesia. Downloaded on September 17,2023 at 12:22:36 UTC from IEEE Xplore. Restrictions apply.
932 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 3, AUGUST 2000

numerical problems because it results in the rapid growth of the

primal or dual objective.
(13) The solution to these problems is obtained by embedding the
original primal and dual pair in a homogeneous and self-dual
linear feasibility (HSDLF) model [9], [11], [12]. Starting simply
Therefore if , then and from , , , ( is a vector of ones of appropriate
this constraint need not be considered any more. size) the SHSD algorithm correctly solves or detects problem in-
feasibility or unboundedness. The disadvantage of this method
III. STANDARD FORM OF THE LP PROBLEM is that the dimension of the Newton equation system solved by
the SHSD algorithm is increased by 1. The SHSD algorithm can
The SCED problem can be summarized as follows: be implemented such that the same factorization as in the PCIP
Minimize the cost given by Eq. (2) algorithm is used. However, the factorization needs to be used
Subject to in one more solution of the Newton equations.
• Generation/demand balance [Eq. (1)].
• Interpolatory variable constraints in Eq. (3). A. The HSDLF Model
• DC load flow equations [Eq. (4)].
The Dual Problem of (14) is:
• Pre-outage network constraints [Eq. (7)].
• Post-outage network constraints [Eq. (11)]. subject to (16)
The standard form of the LP is:
where ,
subject to (14) Introducing a homogeneous variable , and combining the
where , , , primal problems (14) & (16) together gives the HSDLF model:
Two modifications of the SCED are required to put it in the
standard form of the LP problem:
1) The pre-outage state vector is not restricted in sign.
This problem is avoided by placing a redundant lower
bound constraint on the pre-outage network angles (17)
(where is
sufficiently large to ensure that the constraint cannot If we now consider the feasibility model (17) as an LP
become binding). This constraint is then transformed problem with a zero objective function and a zero right hand
to a positive constraint by performing the variable shift side, we see that it is obviously feasible with zero as its trivial
. Note that this is solution. Furthermore, it is self-dual and every feasible solution
equivalent to setting the reference angle at the slack node is a complimentary solution ( ) satisfying:
to .
2) The pre-outage and post-outage network constraints are (18)
transformed into equality constraints by the addition of where the capital letter above and in what follows indicates a
slack variables: diagonal matrix with the elements of the corresponding vector
(small letters) on the diagonal. Since the HSDLF model is fea-
sible, LP theory tells us that it has a strict complimentary solu-
tion. Such a solution must satisfy:

(19)
(15)
Let ( ) be a strict complimentary solution for
the HSDLF model, then the following holds [9], [11], [12]:
IV. THE SHSD ALGORITHM • Equation (14) has a solution if and only if .
Two important aspects in the PCIP described in [3] have not In this case, is an optimal solution for (14) and
been solved satisfactorily from a practical point of view. is an optimal solution for (16).
The first aspect is the choice of the initial solution. Although • If , then , which implies that
the heuristic method presented in [3] works well in practice, , i.e. at least one of and is strictly less than
there is no guarantee that it presents a well-centered point. zero. If , then (16) is infeasible; if ,
The second aspect is the lack of a reliable method to detect so- then (14) is infeasible; and if both and
lution infeasibility or unboundedness of the LP problem. When , then both (14) and (16) are infeasible.
presented with an infeasible problem, the PCIP algorithm is not Since the SHSD algorithm generates a strictly complimentary
able to reduce the residuals below a certain nonzero level. There- solution for the HSDLF model [9], we can use this solution to
fore, the iterates diverge and their norm approaches infinity. Al- obtain a strictly complimentary solution for (14) and (16) or
though this can be used to detect infeasibility, it may lead to detect the infeasibility of at least one of them.
Authorized licensed use limited to: Universitas Indonesia. Downloaded on September 17,2023 at 12:22:36 UTC from IEEE Xplore. Restrictions apply.
JABR et al.: HOMOGENEOUS LINEAR PROGRAMMING ALGORITHM 933

B. Algorithm Description 1) Stopping Criteria: The algorithm continues until one of

The SHSD algorithm basically solves, at each iteration, the the following stopping criteria is satisfied:
following equations for direction ( , , , , ): • The LP is (near) infeasible or unbounded if:

(29)
(20)
• An optimal (approximate) solution ( , , )
is obtained if:
where ( , , , ) is an infeasible point (optimality)
for the HSDLF, are positive scalar components, and
(primal infeasibility)
(21)

(22) (dual infeasibility) (30)

The affine-scaling direction ( ) where is the number of digits accuracy in the solution,
is determined by solving (20) with the right hand side as ( ; and is the norm. Note that by dividing the numer-
; ; ; ) [13]. The infeasible IP methods usually ator and denominator of the fractions in (30) by , we find
take different primal and dual step sizes [3]. It is not obvious that these stopping criteria are equivalent to the ones given
how to implement this feature in the SHSD algorithm, since the in [3], and are therefore independent of .
HSDLF model causes the primal and dual variables to be cou- 2) Starting Point: To initiate the algorithm, we choose the
pled. Xu et al. [9] showed how different step lengths can be used starting point simply as:
for primal ( ) and dual variables ( ). In our implementation,
we have adopted the approach in [9] for the choice of the step
and
length.
To measure the efficacy of the affine-scaling direction, we (31)
define as the hypothetical value of resulting from a full Starting from this point and using the stopping criteria with
step to the boundary and then choose the centring parameter as: , the algorithm solves or detects problem infeasibility for all of
if the test problems.
otherwise. C. Solving the Newton Equations
(23)
and let The most computationally involved task of the primal–dual
interior point methods is the solution of the Newton equations
(24) to find the affine scaling and centering-corrector directions. In
The actual step ( , , , , ) can be obtained as the our implementation [13], Cholesky factorization of the normal
sum of the affine-scaling and centering-corrector step ( , system is used because of its advantages relative to other ap-
), where the centering-corrector step is obtained from proaches [11], [12]. For the SHSD algorithm, the normal system
the, solution of (20) with right hand side set to ( ; ; of equations has the general form:
; ; ) [13]. (32)
Then for the primal update: The direction is therefore obtained using the Sherman–Mor-
(25) rison–Woodbury formula:
and for the dual update:
(26)
(33)
where and are the primal and dual step lengths, respec- where the quantities and are
tively. Since the feasibility model introduces one variable for solved using Cholesky factorization after minimum degree
both and , we choose: ordering is performed [14].

if , V. ITERATIVE CONSTRAINT SEARCH

(27)
if . Although the constraint set is reduced as indicated in Sec-
and the update iterate is: tion II-B-1, the number of constraints will still be large and only
a subset of them will be active. Therefore only critical cases
need to be included in the IP algorithm so as not to overburden
it by constraints. The method of iterative constraint search [15],
(28) [5] is used where only post-outage flows greater than 95% of
Authorized licensed use limited to: Universitas Indonesia. Downloaded on September 17,2023 at 12:22:36 UTC from IEEE Xplore. Restrictions apply.
934 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 3, AUGUST 2000

Fig. 2. Number of constraints checked for system security—24 bus.

Fig. 3. Number of constraints added to the optimization problem—24 bus.

A. IEEE 24 Bus Network

The 24 bus system was dispatched at 100% load. The
pre-fault line rating is given in [16]. The cost rate curves
are given in the Appendix. For (N-1) and (N-2) network
security, all single and double line outages, which do not
split the system, are considered. This amounts to 37 single
line outages and 659 double line outages. The total number
Fig. 1. SCED algorithm flowchart.
of constraints is therefore 25 093. Although the ratio of the
post-fault to pre-fault line rating ( ) is specified as part of
the post-fault line rating are included in the constraints. Using the system data, in this study it is varied in order to view its
this method, only a small number of constraints are added to impact on the dispatch cost. Fig. 2 shows the number of the
the monitored set and the SCED algorithm (see Fig. 1) is usu- total number of constraints that need to be checked for system
ally terminated in one iteration. In our implementation, the ED security. These constraints are obtained following the analysis
and SCED schedules as referred to in Fig. 1 are computed using in Section II-B-1. For all cases considered where ,
the SHSD algorithm. For comparison purposes, the same sched- one iteration of the SCED algorithm (see Fig. 1) is required.
ules (in case they are feasible) can be obtained using the PCIP The number of constraints that are added to the monitored set
algorithm. and therefore considered by the optimization algorithm are
shown in Fig. 3. The ED schedule with the pre-outage network
constraints only, and the SCED schedule for increasing values
VI. RESULTS AND DISCUSSION of ( ) was solved using the SHSD and PCIP methods. the
computational performance with an exit tolerance of is
The SCED algorithm is tested on the IEEE 24-bus system shown in Tables I and II. The results indicate that the SHSD
[16] and a practical network with 175 nodes, 54 generators and algorithm is superior to the PCIP in both iteration count and
303 lines. The algorithm was programmed in Matlab™ [14] and computational time. Moreover, for (first data row,
runs on a Pentium 11 400 MHz PC. The computational results Table II), the SHSD can detect solution infeasibility whilst
are presented below. PCIP fails. For the PCIP, the starting solution heuristic is kept
Authorized licensed use limited to: Universitas Indonesia. Downloaded on September 17,2023 at 12:22:36 UTC from IEEE Xplore. Restrictions apply.
JABR et al.: HOMOGENEOUS LINEAR PROGRAMMING ALGORITHM 935

TABLE I TABLE III

COMPUTATIONAL PERFORMANCE OF THE ED OPTIMIZATION—24 BUS SCHEDULE [pu ] PRODUCED BY THE SCED ALGORITHM—24 BUS

TABLE II
COMPUTATIONAL PERFORMANCE OF THE SCED OPTIMIZATION—24 BUS

TABLE IV
CONSTRAINTS ADDED/COST—175 BUS

TABLE V
COMPUTATIONAL PERFORMANCE OF THE ED OPTIMIZATION—175 BUS

Fig. 4. Secure dispatch cost—24 bus.

as in [17] since the modifications proposed in [3] deteriorated

the algorithm performance. ( ). For example, when only the previous 11 con-
Fig. 4 shows a plot of the secure dispatch cost. It can be seen straint limits ( ) that were violated need to be updated
that with an increasing value of ( ), the secure dispatch cost with the security factor 1.3. It is also seen that for , only
decreases until it reaches the value of £360.2732/1/2 hr at one line is overloaded in the ED solution and therefore needs
which corresponds to the point where the ED schedule for to be added in the SCED optimization. The table also includes
the pre-outage network is secure. The corresponding generation the ratio of secure to insecure cost for varying values of . It is
schedule is given in Table III. possible as before to add a specific set of double line outages.
For (N-1) network security, the ED schedule produces a se- Tables V and VI give the computat7ional performance for the
cure dispatch even if . In this case the SCED optimization ED and SCED optimization. The same conclusions as for the
does not need to be invoked. The schedule in this case is given 24 bus test system are reached, mainly that the SHSD algorithm
by the last column in Table III. outperforms the PCIP and can detect the infeasible problem
given in the first data row of Table VI.
B. Practical 175 Bus Network
Similar tests were run for the 175 bus network. In this case, VII. CONCLUSIONS
only (N-1) security is considered. This amounts to 91 506 con- This paper proposes a SCED algorithm for preventive sched-
tingency constraints. Fortunately, only a small proportion of uling used in scenario planning. The optimization is carried out
these constraints can ever become binding, and also a much using the SHSD interior point method. As compared to the PCIP
smaller number needs to be added to the IP optimization. These algorithm, the SHSD algorithm was shown to have better com-
are summarized in Table V under the headings reduced set and putational performance as well as superior computational prop-
monitored set, respectively. Note that as increases, essentially erties: i) it can always successfully start from a simple point and
only the lines which were overloaded in the previous ED solu- ii) it can reliably detect solution infeasibility and unbounded-
tion need to have their rating updated with the security factor ness. The SCED algorithm is used to dispatch the IEEE 24 bus
Authorized licensed use limited to: Universitas Indonesia. Downloaded on September 17,2023 at 12:22:36 UTC from IEEE Xplore. Restrictions apply.
936 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 15, NO. 3, AUGUST 2000

TABLE VI REFERENCES
COMPUTATIONAL PERFORMANCE OF THE SCED OPTIMIZATION—175 BUS
[1] L. Vargas, V. Quintana, and A. Vannelli, “A tutorial description of an
interior point method and its applications to security constrained eco-
nomic dispatch,” IEEE Transactions on Power Systems, vol. 8, no. 3,
pp. 1315–1324, August 1993.
[2] X. Yan and V. H. Quintana, “An infeasible interior point algorithm for
optimal power-flow problems,” Electric Power Systems Research, vol.
39, no. 1, pp. 39–46, October 1996.
[3] , “An efficient predictor-coffector interior point algorithm for secu-
rity-constrained economic dispatch,” IEEE Transactions on Power Sys-
tems, vol. 12, no. 2, pp. 803–810, May 1997.
[4] A. S. Debs, Modern Power Systems Control And Operation. Boston:
Kluwer Academic Publishers, 1988.
[5] P. E. Beny and R. M. Dunnett, “Contingency constrained economic dis-
patch algorithm for transmission planning,” IEE proceedings—Part C:
Generation, Transmission and Distribution, vol. 1, no. J, pp. 238–224,
1989.
[6] J. Z. Zhu and G. Y. Xu, “A new economic power dispatch method with
security,” Electric Power Systems Research, vol. 25, pp. 9–15, 1992.
[7] J. Z. Zhu, M. R. Irving, and G. Y. Xu, “A new approach to secure eco-
system with (N-1) and (N-2) security, and a practical 175 bus nomic power dispatch,” Electrical Power & Energy Systems, vol. 20, no.
network with (N-1) security. However, the method is general 8, pp. 533–538, 1998.
enough to deal with any combination of outages. An optimiza- [8] C. N. Lu and M. R. Unum, “Network constrained security control using
an interior point algorithm,” IEEE Transactions on Power Systems, vol.
tion method that incorporates the system losses into the SCED 8, no. 3, pp. 1068–1076, August 1993.
algorithm has also been developed [13] and is currently being [9] X. Xu, P.-F. Hung, and Y. Ye, “A simplified homogeneous and self-dual
tested. linear programming algorithm and its implementation,” Annals of Op-
erations Research, vol. 62, pp. 151–171, 1996.
APPENDIX [10] M. S. Bazaraa and C. M. Shetty, Nonlinear Programming—Theory and
Algorithms. New York, NY: J. Wiley & Sons, 1979.
[11] S. J. Wright, Primal Dual Interior Point Methods. Philadelphia:
TABLE VII SIAM, 1997.
GENERATOR BREAK POINTS [MW] FOR THE IEEE 24 BUS SYSTEM [12] T. Terlaky, Interior Point Methods of Mathematical Programming, The
Netherlands: Kluwer Academic Publishers, 1996.
[13] R. A. Jabr and A. H. Coonick, “A computational study of interior point
methods for the economic operation of power systems,” Department of
Electrical and Electronic Engineering, Imperial College, London, UK,
Technical Report RAJ-01, January 1999.
[14] Matlab User’s Guide. Natick, MA: The MathWorks, Inc., 1996.
[15] B. Stott and J. L. Marinbo, “Linear programming for power system net-
work security applications,” IEEE Transactions on Power Apparatus
and Systems, vol. PAS-98, no. 3, pp. 837–848, May/June 1979.
[16] E. D. Farmer, B. L. P. P. Perera, and B. J. Cory, “Optimal pricing of trans-
mission service: Application to large power systems,” IEE Proceedings,
Generation, Transmission and Distribution, vol. 142, no. 3, pp. 263–268,
May 1995.
[17] I. J. Lustig, R. E. Marsten, and D. F. Shanno, “On implementing
TABLE VIII Mehrotra’s predictor-corrector interior-point method for linear pro-
COST BREAK-POINTS [STERLING £/1/2 HR] FOR THE IEEE 24 BUS SYSTEM AT gramming,” SIAM Journal on Optimization, vol. 2, no. 3, pp. 435–449,
THE CORRESPONDING BREAK-POINT IN TABLE VII.
1992.
[18] J. Dyer, “Escort: DC optimization for production cost estimation,” IEE
Colloquium (Digest), no. 102, pp. 8/1–8/4, May 13, 1997.

Rabih A. Jabr received his B.E. in electrical engineering (with high distinction)
from the American University of Beirut in 1997. He is Postgraduate Research
Student in the Department of Electrical and Electronic Engineering at Imperial
College. His research interest is in power system optimization.

Alun H. Coonick graduated from the University of Southampton in 1980 and

obtained his Ph.D. in 1991. He is a Lecturer in the Department of Electrical and
Electronic Engineering, Imperial College. His research interests include power
system stability and control using FACTS devices, artificial intelligence and
optimization techniques applied to power systems.

ACKNOWLEDGMENT Brian J. Cory lectured at Imperial College, London from 1956 until his re-
R. A. Jabr would like to thank the Committee of Vice-Chan- tirement in 1993. He is now a Visiting Professor and Senior Research Fellow.
His main research interests are in all aspects of electrical energy supply. These
cellors and Principals of the Universities of the LJK (CVCP) for include planning; pricing; operating modern systems; and coping with deregu-
the Overseas Research Student (ORS) Award. lation, privatization and competition.

Authorized licensed use limited to: Universitas Indonesia. Downloaded on September 17,2023 at 12:22:36 UTC from IEEE Xplore. Restrictions apply.