898
2, May 1994
REACTIVE POWER OPTIMIZATION USING FUZZY LOAD REPRESENTATION
K. H. Abdul-Rahman
Department of Electrical and Computer Engineering
Illinois Institute of Technology
Chicago, Illinois 60616
Abstract- This paper presents a mathematical formulation for
the optimal voltage/reactive power control problem taking into
account linguistic declaration of system load values. The fuzzy
set theory which is based on the feasibility rather than the fre-
quency of occurrence of an outcome is considered, and possibility
distributions are assigned to load values and bus voltages. The
objective is to minimize power losses considering various load
conditions. The problem is decomposed into four subproblems
via the Dantzig-Wolfe decomposition for reducing the dimen-
sions of the problem. A second Dantzig-Wolfe decomposition
divides each subproblem into several areas leading to a consid-
erable reduction in the dimensions of subproblems. An illus-
trative example demonstrates the applicability of the approach.
The fuzzy approach provides a global solution for the system
behavior under various load conditions.
Keywords- Reactive Power Optimization, Fuzzy Sets, Dantzig-
Wolfe Decomposition
1. INTRODUCTION
Due to limited transmission capabilities for accommodat-
ing additional loads, reactive power allocation has received an
ever increasing attention from the electric utility industry in re-
cent years. Any changes in the system configuration or system
demand may result in higher or lower voltage profiles. In or-
der to maintain desired levels of voltage and reactive flow under'
various operating conditions and system configurations, power
system operators may utilize a number of control tools such as
switching var sources, changing generator voltages, and/or ad-
justing transformer tap settings. By an optimal adjustment of
these controls, the redistribution of the reactive power would
minimize transmission losses. Several methods had emerged in
the literature for finding the optimal settings of control variables
that will eliminate voltage violations and minimize real power
losses in a power system. In the past, several approaches pro-
posed the non-linear formulation as a solution to this problem.
Others suggested the possibility of linearizing the constraints
while maintaining a non-linear objective function (minimize real
power losses). Recently, the Linear Programming (LP) approach
has become dominant in the field for problems with separable
and non-separable objective functions, with satisfactory solu-
tions within a reasonable computation time [2-81. The LP for-
mulation is found to be much more reliable with a faster rate of
convergence than any alternative approaches. Reference [l]dis-
cusses the advantages and the drawbacks of most of the existing
techniques to the reactive power optimization for the operation
and planning of power systems.
A common drawback in previous approaches was based on
the assumption of fixed load values in the LP formulation. Un-
fortunately, this is not the case in real-life situations where un-
certainty in data are often encountered. If, for the data under
consideration, some values are known to occur more often than
93 SM 502-5 PWRS A paper recommended and approved
by the IEEE Power System Engineering Committee of
the IEEE Power Engineering Society for presentation
at the IEEE/PES 1993 Summer Meeting, Vancouver, B.C.,
Canada, July 18-22, 1993. Manuscript submitted Sept.
1, 1993; made available for printing April 14, 1993
PRINTED IN USA
0885-8950/94/$04.00 0 1993 IEEE
IEEE Transactions on Power Systems, Vol. 9, No.
S . M. Shahidehpour
others, then the nature of this uncertainty is obviously proba-
bilistic. Thus, a random variable can be assigned to each datum
and a probabilistic model would be used [9,10]. However, such
a knowledge is not always available, and the pertinent informa-
tion may be limited to some linguistic declarations about the
data (e.g., load at bus i is approximately 15 Mvar, load at bus
j is mainly an industrial type). This type of data is clearly
neither deterministic nor probabilistic. The situation is encoun-
tered most often in forecasting problems where the reflection of
data into the future is not a stationary phenomenon, and human
decisions are involved in an environment that is always fluctu-
ating. This type of data is said to be fuzzy and the nature of
the uncertainty is described as possibilistic.
The possibility theory that is based on fuzzy sets was first
introduced in [ll],where the feasibility rather than the fre-
quency of assessment of a given datum was considered. The
first attempt to apply fuzzy concepts to power systems decision
analysis was in [12]. Since that time, fuzzy sets have been ap-
plied to different fields of power systems [22]. Recently, fuzzy
sets were used to model the soft voltage constraints and the mul-
tiple objective functions of the voltage/reactive control problem,
and a standard LP was used to solve the new fuzzy based for-
mulation [13]. An expert system approach was proposed in [14]
where the approximate reasoning of voltage/reactive power con-
trol based on fuzzy sets was introduced. The heuristic controls
were introduced by a set of rules, and the adaptability of rules
was measured by membership functions. Reference [15] used
fuzzy sets as a tool to control reactive power flow via a heuris-
tic membership function for bus voltages. The objective was to
minimize real power losses, and the problem was modified into
a max-min format.
Although the above formulations used fuzzy sets to reflect
more realistic circumstances, they did not treat the load un-
certainty and assumed fixed values for loads. Fuzzy load flow
analysis incorporating load uncertainties was introduced in [16].
The outcome of the fuzzy load flow was a set of fuzzy voltages,
angles, active and reactive flows and losses as well as active and
reactive power generations. Reference [17] managed to use the
fuzzy DC load flow model, that was developed in [16],to model
real power demand uncertainties in optimizing the cost of real
power generation subject to generation and line flow limits. A
new approach for solving the ower flow problem with uncertain
load values was described in 181. The method was based on in-
terval arithmetic which was viewed as a special case of fuzzy
sets. The values inside the interval were assigned a member-
ship equal to 1, and zero membership was considered for values
outside the interval.
This paper presents a rigorous solution to the optimal volt-
age/reactive power control problem taking into account the un-
certainty associated with the reactive power demand. The ob-
jective is to minimize real power losses under various loading
conditions. There are two ways of solving this problem. The
first one would be to try several load values within a specific
range, optimize each case and prepare a set of guidelines for
the optimal control. This option would be expensive and time
consuming in which some essential features of the data would
not be taken into consideration. A more reasonable method of
analysis, which is considered in this paper, is the one that de-
pends on fuzzy set theory for analyzing different load scenarios.
It is concluded that the arbitrary reduction of fuzzy values to
ordinary closed intervals may result in misleading forecasts or
unclear risky decisions.

List of Symbols:
n = number of system buses
I = number of load buses
m = number of generator buses
nl = number of transmission lines in the system
N A = number of areas
conductance of line k between buses i and j
gk
M = row vector relating real power loss increments to bus
voltage increments
J" = Jacobian matrix with added load effect and tap chang-
ing transformer factor
U = row vector containing the simplex multipliers (dual so-
lution)
PL = real power losses
Q L i = reactive load at bus i
Q . - initial value of reactive load at bus i
Lt -
Qi = injected reactive power at bus i
v;= voltage magnitude at bus i
v;= initial value of voltage magnitude at bus i
6i = angle of the voltage at bus i
G.. - real part of the ijth element of the Y-bus matrix
I1 -
Bij = imaginary part of the ijth element of the Y-bus matrix
P ( X ) = membership function of the x variable
max = maximum value of the variable
min = minimum value of the variable
Vectors and matrices are denoted by boldface letters.
2. DETERMINISTIC REACTIVE POWER
OPTIMIZATION
The purpose of the optimal reactive power control is to im-
prove the voltage profile and minimize system losses. This goal
is achieved by proper adjustments of reactive power variables
in a large power network. In the past, non-linear programming
approaches were proposed, which encountered the convergence
problem in evaluating system losses. In this paper, the hyper-
surface of the non-linear power loss function is approximated by
its tangent hyper-plane at the current operating point and the
LP approach to the reactive power control problem is adopted.
This linear approximation is found to be valid over a small de-
fined region which is formulated by imposing limits on the de-
viations of the control variables from their current values. Most
of the proposed L P approaches formulated the problem by a
sensitivity matrix resulting from the inversion of the Jacobian
matrix, which is essentially a time consuming process and re-
quires a large memory space. The approach implemented in
this study for the fuzzy set application is based on [6] which
incorporates all the variables in formulating a Jacobian matrix,
and hence does not require any matrix inv&ions.
The objective function in this case is represented by,
nl
PI,= c g k [ K 2f yz- 2VT/'jCOS(6;- 6j)]
k=l
Thus, the linearized objective function will be,
min AP, = [%
or,
min APL = M . AV (3)
The minimization problem is subjected to operating con-
899
straints which are the inequality constraints (i.e., limits on the
variables), and the equality constraints (i.e., reactive power de-
mands). These constraints are written as,
Q;'" L Qt L QTa" i = l , ...,m
I y n5 v, 5 l y a " j = 1,...,n (4)
Qk =-QL~ k = 1,...,1
+ +
There are rn n 1 constraints. The first m constraints are
for reactive power sources and tap changing transformer termi-
nals. We will refer to the matrix of reactive power injections at
these buses as Q1. The next n constraints are the bus voltage
constraints. The last 1 equality constraints are for loads and
junction buses that are not connected to transformer terminals.
We will refer to the matrix of reactive power injections at these
buses as Q2. So, the linearized form of the constraints is found
to be,
AQF'" 5 AQ1= JYAV 5 AQF" (5)
Avmin
5 A V 5 AVmau (7)
where JY and Ji are submatrices of J" which is a modified
Jacobian matrix. The load and the tap changing effects are
formulated in terms of voltage increments, thus their effects are
implicitly considered in J". Further information regarding the
derivation of this model is given in Appendix A.
The following section will discuss the formulation of the
reactive power optimization problem under uncertain load con-
ditions described by fuzzy sets.
3. REACTIVE POWER OPTIMIZATION WITH
UNCERTAIN LOAD CONDITIONS
The fuzzy set theory was introduced to various system en-
gineering problems in which uncertainties were represented as
intrinsic ambiguities. In Appendix B, some basic definitions of
fuzzy sets, which are of direct relevance to the paper, are re-
viewed.
3.1 Fuzzy Load Representation
In this study, linguistic declarations of variables are trans-
lated into possibility distributions by assigning a degree of mem-
bership to each possible value of the variable. Possibility distri-
bution refers to the mapping of a fuzzy variable on the [0,1]
interval. In the power system analysis, some loads and genera-
tions are determined precisely and others are described in terms
of "more or less" expressions. To model such fuzzy quantities,
we use trapezoidal or triangular possibility distribution where
the latter is a special case of the former. So, the fuzzy reactive
load at a certain bus that would never exceed QP, is always
higher than Q i ) .and typically falls between Q!) and QE), is
represented by Fig. 1. The possibility distribution will have a
value of 1 for the load values that are highly possible, and will
drop as possibility diminishes. A zero possibility is assigned to
-
the values that are rather impossible to occur, which are located
beyond the two extremes.
't
Fig. 1 possibility distribution of reactive power load
900

3.2 Mathematical Model subject to :

3.2.1 O p e r a t i n g constraints
Based on the discussion in Appendix C concerning the re-
lation between incremental changes in loads and bus voltages,
the uncertainty in system voltages is depicted in Fig. 2 cor-
responding to reactive loads in Fig 1. The voltage possibility
distribution signifies four break-points, i.e., V(l), V(’) ,V(3),V(4)
and their possibilities. This representation of voltages concurs
with our experience in power systems indicating that as the sys-
tem demand decreases, the load bus voltages will increase and
smaller voltages would be required at generators to maintain where v(k) is the vector of bus voltages for break-point IC at the
specific load voltages. However, if the system demand increases, current operating state. The successive solution to this problem
the load bus voltages will decrease and higher generator voltages will determine break-points of the voltage possibility distribu-
will be needed to raise load bus voltages to the specified values. tion. A follow up power flow will determine the break-points for
The above discussion can be verified by a fuzzy load flow [16]. the possibility distribution of transmission losses, which corre-
The fuzzy load flow is an alternative to our previous argument sponds to the possibility distribution of loads. It is conceivable
regarding the correlation between bus voltages and reactive load that the number of the problem variables increase due to the
increments. fuzzy modeling. However, our specific problem formulation uti-
lizes the Dantzig-Wolfe decomposition as a mathematical tool
The relations between voltage magnitudes at four break to overcome this dimensionality problem.
points in Fig. 2 are subjected to their minimum and maximum
limits. Therefore, voltage constraints are,
4. D A N T Z I G - W O L F E D E C O M P O S I T I O N

In addition, the minimization problem is subjected to the 4.1 Decomposition of t h e Original P r o b l e m

operating conditions given by (5-6) for each break-point. In this It is seen in (13)-(15) that there are four different sets of
capacity, the Jacobian matrix will be evaluated for each break variables, each describing one of the break-points. Thus, we
point. decompose the equations into 4 subproblems which are linked
together through the linking constraints (13). After adding the
necessary slack variables, the fuzzy optimization problem is for-
mulated as,
min 2
k= 1
M(k)AV(k)

subject to :

Fig. 2 Possibility distribution of voltage

3.2.2 Objective function

The objective function for the deterministic case was given This angular structure of four diagonal blocks is solved by the
in (3), where AV was a crisp variable. In the fuzzy case, APL Dantzig-Wolfe (DW) decomposition [21 , which is a powerful
is a function of fuzzy voltages and is given as, f’
technique for reducing the dimension o the original problem.
DW decomposition coordinates the alternate solutions of sub-
AL
? = M-AV (9) problems via a master problem that is obtained by a linear trans-
formation. The dimension of the master problem is equal to the
The minimization of A L
? is imposed by taking the the removal number of linking constraints plus the number of subproblems,
of the fuzzy objective function as discussed in Appendix B. while each subproblem has a dimension equal to the number of
Therefore, the objective function is, constraints in the corresponding block. The solution of the mas-
ter problem generates new simplex multipliers (dual solutions)
that will adjust the cost function of the subproblems. The solu-
tion of the subproblems with the adjusted objective function will
provide the master problem with new columns primal solutions
which can be written as, d
to enter the master basis matrix. The revise simplex metho
is used to solve the master problem and either the simplex or
d
the revised simplex methods is used to solve the subproblems.
The interaction between the subproblems and the master prob-
lem is shown in Fig. 3. Each subproblem will have its objective
function and the respective constraints. For instance, the for-
3.2.3 P r o b l e m formulation mulation of the kfh subproblem is as follows,
In this study, the overall formulation of the loss minimiza-
tion is given as follows,
subject to

Master Problem
Subproblem 1
Subproblem 2
> - linkingo f constraints
Fig. 3 Interaction between the master problem and
the subproblems
The last set of constraints are added for accuracy as we are
dealing with a linearized form of non-linear equations. So, we
restrict the changes in the control variables to a small range
in each iteration, otherwise the LP solution will greatly deviate
from that of load flow. The load flow solution after each iteration
finds another operating point according to the revised settings
of control variables obtained from the LP.
To minimize the total number of iterations for the final
solution, more than one candidate columns are introduced into
the basis of the master problem in each iteration. This technique
is referred to as the multiple column strategy.
4.2 Decomposition of Subproblems
The objective of each subproblem is to minimize the cor-
responding power losses. The problem formulation of the kth
subproblem was described in (17). However, each subproblem
in our analysis may consist of multi areas that are linked to-
gether through transmission lines. In this situation, the reactive
power control devices may be located locally as each area tries
its own localized control.
We can decompose multi-area systems in each subproblem
by applying a second order DW decomposition. The model con-
sists of a master problem for linking constraints associated with
buses that link these areas, and N A independent blocks repre-
senting areas. The objective function for each area is to mini-
mize area power losses. This decomposition of a large problem
into subproblems, and the further decomposition of these sub-
problem into several small areas reduces the dimension of the
problem to the level where personal computers can be used. Fig.
4 shows the first and second order decompositions, and the over-
all structure of the problem.
4.2.1 Enhancing the system security
The basic function of an electric power system is to provide
an adequate supply of electrical energy to all its customers as
economically as possible with a reasonable level of quality and
continuity. In our study, we have added another feature as the
final solution of bus voltages are to be within security limits.
Let us assume that a certain operating state of a power system
is found to be most secured according to the worst case sce-
nario analysis, and denote the voltage at bus i corresponding
to this secured operating state as vi.. In order to direct the
voltage profile, we specify the degree of satisfaction with dif-
ferent variables for being closer to the secured operating point
within their operating ranges. A fuzzy LP implemented in [15
is used to modify the objective function of each area into severa
objective functions and assign a membership function to each
i 3 contains buses 21 to 30. Buses 4,6,8,9 and 10 link area 1 to
objective. The limits on variables are fuzzified as well in favor
of a more reasonable solution with minimum losses. The feasible
901
Master problem
linking Constraints for
the subproblems
1
I
- Objective function (
areas
t
Suboroblem 4
linking constraints
4-97 I
Area
.. .
.NA
I
Fig. 4 Overall structure of the problem
region is the one defined by the intersection of the constraints
and the objectives. This intersection is simulated as the mini-
mum membership for the constraints and the objectives. The
problem is solved by the standard LP to maximize the intersec-
tion, which will direct the final solution towards the values of
variables with higher memberships. In this study, the proposed
method is used to enhance the security of the system by offering
a tighter control on variables within their operating ranges.
5. ILLUSTRATIVE EXAMPLE
In this section, the application of the proposed approach
to a 30-bus system is presented. The system is shown in Fig.
5, and line and bus data are given in [6,15]. The permissible
voltage range is 0.9 to 1.1 pu. The possibility distributions of
the reactive loads are given in Table 1 which are used to find
the possibility distributions of bus voltages and system losses.
The solution starts by computing the elements of J" and
identifying the 30-bus voltage constraints which represent the
linking constraints of the subproblems. Each bus voltage is
represented by inequality constraints (13). For this example,
we have 20 inequalities representing (14) and 10 equalities for
(15) at each value of k . The DW decomposition is applied, and
the problem is decomposed into 4 subproblems and one master
problem. The master problem provides the subproblems with
the dual solutions, and the subproblems will feed the primal so-
lutions to the master problem. The second DW decomposition
is applied and each subproblem is divided into 3 areas. Area 1
contains buses 1 to 10, area 2 contains buses 11 to 20 and area
other areas, buses 11,12,15,17 and 20 link area 2 to other areas
and buses 21,22,23 and 28 link area 3 to other areas. So, each
subproblem will consist of linking constraints corresponding to
 For the sake of comparison, we have also included the opti-
mized values, using the conventional methods, for minimum and
maximum loads (i.e., &(I) and Qf' in Table 1). It is seen in Ta-
bles 2 and 3 that the fuzzy solution provides a smaller range
of voltages and generations than that of the fixed interval cor-
responding to the two-extreme values, indicating that the fixed
load interval leads to an overestimate of the system behavior in
an uncertain environment.

Table 2 Final possibility distributions for the 30 bus-system

-
Bus
voltages (pu) corresponding to Fig. 2

v (1) v (2) v (3) v (4) Results of minimum

- and maximum loads
G1 1.097 1.097 1.098 1.100 1.095 1.100
G2 1.093 1.093 1.094 1.094 1.093 1.094
3 1.062 1.083 1.091 1.100 1.100 1.058
4 1.061 1.070 1.083 1.100 1.100 1.057
G5 1.092 1.095 1.097 1.100 1.090 1.100
6 1.068 1.074 1.087 1.098 1.100 1.065
7 1.066 1.078 1.081 1.092 1.094 1.062
G8 1.088 1.090 1.092 1.093 1.087 1.094
9 1.083 1.085 1.089 1.100 1.100 1.082
10 0.987 0.991 1.012 1.026 1.028 0.984
G11 1.097 1.098 1.098 1.098 1.096 1.098
12 0.974 0.979 0.991 1.040 1.043 0.971
G13 1.075 1.080 1.099 1.100 1.073 1.100
14 0.985 0.991 1.010 1.050 1.052 0.984
15 0.986 0.995 1.021 1.060 1.063 0.981
16 0.979 0.985 0.999 1.050 1.051 0.977
17 0.994 1.001 1.015 1.038 1.041 0.992
18 0.982 0.989 0.995 1.041 1.045 0.979
19 0.981 0.987 0.994 1.021 1.024 0.978
20 0.982 0.987 0.994 1.022 1.026 0.977
21 0.991 1.011 1.031 1.078 1.081 0.986
22 0.968 0.974 0.985 1.020 1.023 0.965
23 0.974 0.988 1.001 1.036 1.039 0.970
24 0.972 0.985 0.998 1.023 1.027 0.968
25 0.982 0.991 1.010 1.041 1.045 0.981
26 0.976 0.983 0.997 1.024 1.027 0.974
27 0.991 0.999 1.021 1.061 1.061 0.990
loads (pu) corresponding to Fig. 1 28 1.011 1.027 1.050 1.082 1.085 1.009
29 0.981 0.989 0.997 1.033 1.037 0.976
-
30 0.971 0.985 0.992 1.025 1.029 0.967

Table 3 Final possibility distributions of reactive power

3 0.000 0.010 0.015 0.020 generation for the 30 bus-wstem
4 0.010 0.020 0.025 0.030
7 0.050 0.100 0.110 0.150 Bus Q:') Q,'" Q,(3) Q,'" Results O f minimum
and maximum loads
10 0.010 0.020 0.027 0.030
12 0.050 0.070 0.080 0.089 1 -0.1943 -0.1548 -0.1314 -0.1211 -0.2000 -0.1208
2 0.3994 0.4312 0.4987 0.5434 0.3991 0.5434
14 0.010 0.015 0.017 0.020 5 0.2801 0.2998 0.3114 0.3402 0.2799 0.3402
15 0.010 0.025 0.030 0.035 8 0.2654 0.2898 0.3102 0.3334 0.2652 0.3335
16 0.005 0.010 0.020 0.025 11 0.1542 0.1699 0.1823 0.2112 0.1540 0.2114
17 0.030 0.050 0.070 0.080
18 0.005 0.010 0.015 0.020
19 0.020 0.030 0.050 0.060 The fuzzy real power loss, which corresponds to fuzzy loads,
20 0.000 0.005 0.010 0.015 and voltages obtained in (12)-(15), has a possibility distribution
21 0.050 0.100 0.110 0.150 as shown in Fig. 6. The optimal values obtained from the min-
23 0.010 0.015 0.020 0.025 ima and maxima of the load values are included in Fig. 6 to
24 0.040 0.050 0.070 0.090 provide a comparison between the results. In this case the fived
26 0.005 0.010 0.020 0.025 load interval provides a wider range of system losses than that
29 0.000 0.005 0.010 0.020 of the fuzzy model, indicating that fixed load intervals result in
30 0.005 0.015 0.025 0.035 overestimated forecasts and may lead to a higher system opera-
tion cost.
the reactive power constraints for linking buses, and three sets
of independent equations for areas 1,2 and 3. The first set con-. TPP,
tains the reactive power constraints for buses 1,2,3,5 and 7, the
second set of constraints represents buses 13,14,16,18 and 19,
and the third set contains constraints for buses 24,25,26,27,29
and 30. In this example, we use the restricted step sizes as
AVatep= .02pu ,ATstep= .05pu ,AQc = .05pu. After each iter-
ation of the first DW decomposition, new settings for the control
variables are computed. To account for the non-linearity of the 4.86 4.89 5.1 1 6.61 7.32 7.39
loss equation, a new operating point is computed based on the
updated variables. The final solution is reached after 11 itera- t w o e x t r e m e cases
tions, and the final possibility distributions for bus voltages and
reactive power generation are given in Tables 2 and 3, respec- Fig. 6 Possibility distribution of real power losses for
tively. the 30 bus-system

6. CONCLUSION
Uncertainty in data is often encountered as we deal with
real-life situations. In majority of cases, as the available data
are insufficient for the solution of the problem, the linguistic dec-
laration may be used to describe the validity of the data. The
uncertainty in this kind of modeling is based on intrinsic ambi-
guity rather than the frequency of assessment of the data under
consideration. Such uncertainty is suitably modelled via fuzzy
sets. Forecasting future demands is one area where this kind
of uncertainty can be encountered. The proposed formulation
of the optimal voltage/reactive power control problem indicated
that the fuzzy modeling of loads would enable power system op-
erators to operate the system more economically in an uncertain
environment.
It should be emphasized that the fuzzy set modeling is not a
replacement for the probabilistic approach. Each model is used
to describe a different type of uncertainty, and it may even be
possible to incorporate both models in one approach.
The ability of fuzzy sets to discriminate between different
values of variables inside a given range via the variable mem-
bership is an important feature of the proposed approach that
cannot be obtained in the standard LP. This feature can be best
utilized for controlling the behavior of power system variables
in the optimization process. Initial steps towards this direction
were reported in [13-151. Finally, fuzzy set theory has merits for
incorporating heuristics and powerful numerical methods under
one framework for large scale applications.
REFERENCES
S.M. Shahidehpour and N. Deeb, L‘AnOverview of the Re-
active Power Allocation in Electric Power Systems,” Elec-
tric Machines and Power Systems, Vol. 18, pp. 495-518,
1990.
K. Mamandur and R. Chenoweth, “Optimal Control of Re-
active Power Flow for Improvements in Voltage Profiles and
for Real power Loss Minimization,” IEEE Trans. on Power
Apparatus & Systems, Vol. PAS-100, No. 7, pp. 3185-
3194, July 1981.
J. Qiu and S.M. Shahidehpour, “A New Approach for Mini-
mizing Power Losses and Improving Voltage Profile,” IEEE
Trans. on Power Systems, Vol. 2, No. 2, pp. 287-295, May
1987.
0. Alsac, J. Bright, M. Prais and B. Stott, Vurther Devel-
opments in LP-Based Optimal Power Flow,” IEEE Trans.
on Power Systems, Vol. 5, No. 3, pp. 697-711, Aug. 1990.
Walter L. Snyder, Jr., ‘LLinearProgramming Adapted for
Optimal Power Flow,” IEEE Tutorial # 90EH0328-5-PWR,
pp. 20-36, 1990.
N. Deeb and S.M.Shahidehpour, “An Efficient Technique
for Reactive Power Dispatch Using a Revised Linear Pro-
gramming Approach,” Electric Power Systems Research
Journal, Vol. 15, pp. 121-134,1988.
N. Deeb and S.M. Shahidehpour, L‘LinearReactive Power
Optimization in a Large Power Network Using The Decom-
position Approach,” IEEE Trans. on Power Systems, Vol.
5 , No. 2, pp. 428438, May 1990.
N. Deeb and S.M. Shahidehpour, “A Decomposition Ap-
proach for Minimizing Real Power Losses in Power SYS-
tems,” IEE Proceedings, Part C, Vol. 138, No. 1, pp.
27-38, Jan. 1991.
J. F. Dopazo, 0. A. Klitin and A. M. Sasson, “Stochastic
Load Flow,” IEEE Trans. on Power Apparatus & Systems,
Vol. PAS-94, No. 2, pp. 299-309, Mar./April 1975.
[lo] A. P. Meliopoulos, G. J. Cokkinides and X. Y. Chao, ‘LA
New Probabilistic Power Flow Analysis Method,” IEEE
Trans. on Power Systems, Vol. 5, No. 1, pp. 182-190,
Feb. 1990.
903
L.A Zadeh, “Fuzzy Sets,” Information and Control, Vol. 8,
pp. 338-353, 1965.
S. B. Dhar, ‘LPowerSystem Long-Range Decision Analysis
Under Fuzzy Environment,” IEEE Trans. on Power Appa-
ratus & Systems, Vol. PAS-98, No. 2, pp. 585-596, Mar.
1979.
1131 K. Tomsovic, “A Fuzzy Linear Programming Approach to
the Reactive Power/Voltage Control Problem,” IEEE
Trans. on Power Systems, Vol. 7, No. 1, pp. 287-293,
Feb. 1992.
R. Yokoyama, T. Niimura and Y. Nakanishi, “ A Coordi-
nated Control of Voltage and Reactive Power by Heuristic
Modeling and Approximate Reasoning,’’ Paper # 92 WM
114-9 PWRS, Presented at the IEEE/PES 1992 Winter
Meeting, New York, New York, Jan. 1992.
K. H. Abdul-Rahman and S. M. Shahidehpour, “A Fuzzy -
Based Optimal Reactive Power Control,” Paper # 92 SM
402-8 PWRS, Presented at the IEEE/PES 1992 Summer
Meeting, Seattle, WA, July 1992.
V. Miranda, M. Matos and J. T. Saraiva, L‘Fuzzy Load
Flow - New Algorithms Incorporating Uncertain Genera-
tion and Load Representation”, Proceedings of the 10th
PSCC, Graz, Austria, pp. 621-627, Aug. 1990.
1171 V. Miranda and J.T. Saraiva, “Fuzzy Modeling of Power
System Optimal Load Flow,” IEEE Trans. on Power Sys-
tems, Vol. 7, No. 2, pp. 843-849, May 1992.
Z. Wang and F. L. Alvarado, “Interval Arithmetic in Power
Flow Analysis,” IEEE Trans. on Power Systems, Vol. 7,
NO. 3, pp. 1341-1349, Aug. 1992.
H. J. Zimmermann, “Fuzzy Sets, Decision Making, and Ex-
pert Systems,” Kluwer Academic Publishers, Boston, 1987.
A. Kaufmann, M. M. Gupta, “Fuzzy Mathematical Models
in Engineering and Management Science,” North-Holland
Publishing Company, Amsterdam, 1988.
G. B. Dantzig and P. Wolfe, L‘TheDecomposition Algorithm
for Linear Programs,” Econornetrica, Vol. 29, No. 4, pp.
767-778, Oct. 1961.
[22] S.M. Shahidehpour, “A Fuzzy Set Approach to Heuristic
Power Generation Scheduling with Uncertain Data,” Pro-
ceedings of 1991 NSF/EEI Workshop, Norman, OK., July
1991.
APPENDIX A
For buses i and j and a tap changing transformer in be-
tween, the Jacobian (J) elements are given by (A.1). These
elements are derived assuming that the bus phase angles are
fixed in each optimization iteration, which disregard the cou-
pling between variables. The elements of J are modified to take
into account the effect of voltage increments on reactive loads.
Assuming that the load is at bus i, the modification is given in
(A.2). The effect of tap changing transformers is added in (A.3)
I
and AA), to form the modified Jacobian (J“) elements. The
detai s of formulation are given in [6].
J,!j=J,j i , j = 1 , ..., n, i # j )
904
1'
APPENDIX B
The idea of fuzzifying a variable is to replace the concept
that a variable has a precise value by the fuzzy concept indicat-
ing that a variable has a degree of membership assigned to each
possible value of the variable. In this paper the term possibil-
ity distribution, [19], refers to the mapping of a set X to [0,1].
We will represent the possibility distribution by its break-point
values (i.e., Z ~ , Z ~ , Q . , Zas~ shown
), in Fig. B1.
The minimization of a fuzzy variable X , given in Fig. B1,
is translated into the minimization of its removal from the px
axis. The concept of removal in fuzzy sets, can be interpreted as
the distance of the fuzzy variable from the pz axis. It is proved
in [20] that the distance is given as,
where A1 and A2 are given in Fig. B1. For the fuzzy variable
X in Fig. B1, the distance given by (B.l) is computed as,
21 + 322 + +
z3 24
(B.2)
4
where (B.2) is the same as (9) for the minimization of losses. It
should be emphasized that no matter how many points we used
to describe the possibility distribution in Fig. B1, the distance
from the pz axis will always be given as (B.2).
T px
Fig. B1 Trapezoidal possibility distribution of a fuzzy variable
APPENDIX C
Consider a power system with m generation buses and 1
load buses summing up to a total of 12 buses. The reactive
power injection at bus i is given as,
n
Qi = c y [G;jsin(Si - Si) - Bijcos(6i - Sj)] (C.l)
j=1
The reactive power demands are, in general, voltage dependent.
Thus, the reactive load at bus i is given as,
where qi is 0 for constant power load, 1 for constant current
load and 2 for constant impedance load. If the injected power
at bus i is changed by AQ,, due to a capacitor switching or
load changes, the corresponding load bus voltage increments will
be found by linearizing (C.l) and (C.2) with respect to load
voltages, as follows,
or in the matrix form,
i, j = rn + 1,..., n
where,
aQ-
D=-+ ~ Q L
avL avL
n
39.
2 = -2XBii
av; + [Gij sin(& - Sj) -
j=1
and
A quick procedure to check the signs of equations (C.7)-(C.9)
is to assume unity voltage magnitudes and a small difference
between their angles. If we assume all Gij are equal to zero, we
can write the following,
aQi = -Bij
-
av, 50 (C.10)
Thus, D has positive diagonal elements that are equal to the
negative s u m of its off-diagonal elements. So, matrix D is a M-
matrix and by definition D-' is a non-negative matrix, which
suggests that if the incremental reactive power injected into a
load bus (AQ,,) is positive due to aload reduction, then voltages
at load buses will increase and AVL will be positive. On the
other hand, if the injected reactive power decreases due to a load
increase, then the load bus voltages will decrease. For generator
buses, we can easily verify that an increase in the injected power
will cause the generator voltages to decrease and vice versa.
BIOGRAPHIES
K.H. Abdul-Rahman was born on June 2nd, 1964 in Jor-
dan. He received his BS and MS in Electrical Engineering from
Kuwait University in 1986 and 1990, respectively. He is cur-
rently a Ph.D. student in the ECE Dept. at IIT. His research
interests include optimization and control of power system.
S.M. Shahidehpour received his Ph.D. in electrical engineer-
ing from the University of Missouri-Columbiain 1981. He has
been with the IIT since 1983, where he is currently a profes-
sor in the ECE Department. He has published over 120 papers
on power systems planning and operation. He was the vice-
chairman for the 1992 IEEE International Conference on Sys-
tems, Man and Cybernetics, and serves as the associate director
of the American Power Conference.

Discussion
T. Niimura (University of British Columbia, Vancouver, Canada):
The authors are to be commended for an interesting paper. I would
like to supplement the work by raising two possible improvements
for hrther development of firzzy set’s applications:
(1) One of the advantages to apply f i z z y sets for voltage-reactive
power control modeling is the explanation of state by defining
linguistic terms such as “high voltage” corresponding to numerical
values [Dl]. For example, in Fig. 1 the authors define the possible
reactive power by “always higher than QL(’)’’ and so on. However,
the possibility distribution of voltage (Fig. 2) is just given by four
“break-points”. To facilitate the comprehension they could be
correspondingly referred to such that voltage V will “never be
lower than V(1) but always lower than V(4) “ and “most possibly
between V(z)and V(3).”
(2) Another advantage of fizzy sets’ application is $he flexibility to
conform with changing power system states. However, it is a sort
of perpetual concern how the re-definition of the fizzy sets ~ Q L
and pv or “the break-points” affects the overall solution. Due to
the authors’ modeling, based on linear membership finctions and
LP frame work, it seems relatively straight-forward to introduce
sensitivity or marginal cost of the adjustment in parameters for the
change of solution.
I would appreciate the authors’ comments or insight for the above
discussions.
[Dl] R. Yokoyama, T. Niimura, and Y. Nakanishi, “A Coordinated
Control of Voltage and Reactive Power by Heuristic Modeling
and Approximate Reasoning,” IEEE Trans. PWRS, Vol. 8,
“0.2, pp. 636-645, May 1993.
Manuscript received August 10, 1993.
905
K.H.Abdul-Rahman a n d S.M.Shahidehpour- We would
like to express our appreciation to Dr. Niimura for his interest
in our paper and his valuable comments. In the following, we
will provide our response to his comments.
(1) The voltage possibility distribution as seen in Fig. 2 indi-
cates that voltage is always higher than V(’)and lower than
V(4) and most likely to be between V ( 2 )and V ( 3 ) .This in-
terpretation of voltage possibility distribution is similar to
that of the load given in Fig. 1. However, the break points
of the voltage possibility distributions are determined ac-
cording to the given load possibility distribution. Hence
the range of voltage values or the break points in Fig. 2
may vary between V”’” and V“””. The importance of the
fuzzy set approach is that it brings forth the notion of im-
precision that is overlooked in conventional approaches.
(2) We agree with the discusser that artificial costs of changes in
the solution for variations in parameters may be introduced
to find the flexibility of the approach. Alternatively, the
optimization problem can be formulated as a parametric
LP to study the effect of varying the break points.
As more complex and additional constraints are imposed
on the system operation, there will be a need for enhancing
the existing technology to reduce the computation time, meet
the increasing complexity and provide proper options for the
operation of power systems. In [C2], advances in AI technology
are utilized for the solution of the var control problem. The
approach is based on modeling load uncertainty by fuzzy sets,
which is given as input to artificial neural network to identify
the closest control solution corresponding to uncertain loads.
Expert system is used to assure the feasibility of the solution.
[CZ]K.H. AbduI-Rahman, S.M.Shahidehpour, “Application
of artificial intelligence to optimal var control in electric
power systems,” in Proceedings of Expert System Applica-
tions for the Electric Power Industry: International Con-
ference and Exhibition, Phoenix, AZ, Dec. 1993
Manuscript received October 11, 1993.