Applied Mathematical Sciences, Vol. 6, 2012, no.

54, 2649 - 2667

Utilization of Trust Region Algorithm in Solving
Reactive Power Compensation Problem
Bothina El-Sobky
1
, Yousria Abo-Elnaga
2
and Hanadi Zahed
Department of Mathematics, Faculty of Science
Taibah University, Al-Madinah Elmnowara, P.O. Box 344
Saudi Arabia
Abstract
In this paper, we introduce a trust region algorithm for solving a re-
active power compensation (RPC) problem in a multi-objective context.
The trust region algorithm has proven to be a very successful globaliza-
tion technique for nonlinear programming problems with equality and
inequality constraints. The proposed approach is suitable for (RPC)
problems where the objective function may be ill- dened, having a
nonconvex pareto-optimal front. Also, we identify the weight values
which reect the degree of satisfaction of each objective. The proposed
approach is carried out on the standard IEEE 30-bus 6-generator test
systems to conrm the eectiveness of the algorithm used to solve the
multi-objective RPC problem.
A Matlab implementation of our algorithm was used in solving one
case study and the results are reported.
Keywords: Reactive Power Compensation, Multiobjective Optimization,
Trust Region, Single Optimization
1 Introduction
In the study of the expansion, planning, and operation of power systems, one
of the most important problems is the issue of Reactive Power Compensa-
tion (RPC). Identifying the adequate size and the physical distribution of the
compensation devices in order to ensure a satisfactory voltage prole while
minimizing the cost of compensation, is the main goal of reactive power com-
pensation. In several previous studies of this problem (see [5] and [7]), it has
1
Permanent position: Department of Mathematics, Faculty of Science, Alexandria Uni-
versity, Egypt
2
Permanent position: Department of Mathematics, Faculty of Science, Alexandria Uni-
versity, Egypt, e-mail: yousria [email protected]
2650 B. El-Sobky, Y. Abo-Elnaga and H. Zahed
usually been regarded as a single objective optimization problem (SOP), where
only one objective is optimized. Multiobjective optimization algorithms are
preferable to solve the (RPC) problem over a single objective optimization al-
gorithm because most problems have more than one objective to be optimized
and usually the objectives are contradictory. For example, the (RPC) prob-
lem requires the optimization of power losses, voltage prole, and investment.
Many studies have been undertaken considering this situation, where multi-
objective optimization algorithms (MOA) were introduced to simultaneously
optimize several independent objectives, (see for example [1], [2], [3], [4], [14],
[17], [21], and [23]), demonstrates how traditional multiobjective optimization
algorithms usually provide a unique and optimal solution.
On the other hand, multiobjective optimization evolutionary algorithms
(MOEA) independently and simultaneously optimize several parameters turn-
ing most traditional constraints into new objective functions (see [1], [2], and
[3]). This seems more natural for real world problems where choosing a thresh-
old may seem arbitrary (see [17]). As a result, a wide set of optimal solutions
(pareto set) may be found. Therefore, an engineer may have a whole set of
optimal alternatives before deciding which solution is the best compromise
from the dierent features. In this paper we help the decision maker (DM) in
choosing the best comparative solution from all the nite sets of pareto optimal
solutions (see [15]). Our work is based on one simple way to combine multiple
objective functions into a scalar tness function which is the weighted sum.
Then we do some sort of parametric analysis to name the weight values which
reect the degree of satisfaction of each objective. In this paper, we intro-
duce a new trust region algorithm for solving our problem. In this algorithm,
an active set strategy is used together with multiplear method to convert the
computation of the trial step to an easy trust region subproblem similar to
the one for unconstrained case. The trust-region strategy for solving general
nonlinear programming problem with equality and inequality constraints has
proven to be very successful both theoretically and practically (see for exam-
ple [8], [10], [11], and [12]). Finally, the approach has been implemented to
select one solution which will satisfy the dierent objectives to some extent.
The standard IEEE 30-bus 6-generator test system has been used to verify
the validity of the proposed algorithm. The weighting approach is considered
as one of the most useful algorithms in treating multiobjective optimization
problems to generate a wide set of optimal solutions (Pareto set) (see [17]).
In this work, the eect of changing the weights on the compensation cost,
active power losses and voltage deviation was studied to show the degree of
satisfaction of each objective. In each case one weight is changed linearly and
the two other weights are generated.
Here, we introduces some notation for subscripted functions denote func-
tion values at particular points; for example, f
k
= f(x
k
),
x
f
k
=
x
f(x
k
),
Utilization of trust region algorithm 2651
and so on. The matrix H
k
denotes the Hessian of the objective function at the
point (x
k
) or an approximation to it. Finally, all norms are l
2
-norms.
In the following section, we present Reactive Power Compensation (RPC).
In section 3, we discuss the multiobjective formulation of the (RPC) problem
and in section 4, we present the outline of the trust region algorithm which is
used to solve (RPC) problem. We present the implementation of our approach
in section 5 and we discuss results and our approach in section 6. Finally,
section 7 contains concluding remarks.
2 Reactive Power Compensation (RPC)
In order to identify the adequate size and the physical distribution of the
compensation devices for obtaining a satisfactory voltage prole with minimal
compensation cost, we consider the following assumptions in the formulation
of the problem.
The power system is considered only at peak load. A shunt-capacitor bank
cost per MVAr is the same for all busbars of the power system
In this present work, we have identied three objective functions to be
minimized and a load ow equations. The three objective functions are f
1
,
f
2
, and f
3
, which are related to investment, transmission losses, and quality of
service respectively.
The detailed descriptions of these three objective functions is as follows:
First: f
1
is investment in reactive power compensation devices
minimize f
1
=

n
i=1
B
i
subject to 0 f
1
f
1max
,
0 B
i
B
imax
,
(1)
where f
1max
is the maximum amount available for investment, B
i
is the com-
pensation at busbar i measured in MVAr and B
imax
is the maximum compen-
sation allowed at a particular bus of the system. The number n denoted to
the number of buses in the power system. To simplify, we take the price per
MVAr as unity.
Second: f
2
is the total active power losses in MW
minimize f
2
= P
g
P
l
subject to P
g
min
P
g
P
gmax
,
(2)
where P
g
is the total active power generated, P
l
is the total system load, P
g
min
is the minimum active power generated by the generator and P
gmax
is the
maximum active power generated by the generator.
2652 B. El-Sobky, Y. Abo-Elnaga and H. Zahed
Third: f
3
is the average voltage deviation per unit (pu)
minimize f
3
=
V
i
V

i

2
2
n
subject to V
i
min
V
i
V
imax
, i = 1, 2, ..., n
(3)
where V
i
and V

i
are the actual voltage and the desired voltage respectively
at busbar i per unity, while V
i
min
and V
imax
are the minimum and maximum
actual voltage respectively at busbar i per unity. For more details about f
1
,
f
2
, and f
3
(see [9] and[16]).
The load ow equations illustrated in [20] are the following equations:
P
p
= P
Gp
P
Cp
=
n

q=1
V
p
V
q
Y
pq
cos(
p

pq
)
Q
p
= Q
Gp
Q
Cp
=
n

q=1
V
p
V
q
Y
pq
sin(
p

pq
)
where P
Gp
and P
Cp
are the real power generations and demands at bus p re-
spectively, while Q
Gp
and Q
Cp
are the reactive power generations and demands
at bus p respectivly. V
p
and V
q
are the voltage magnitude at bus p and q, re-
spectively. Y
pq
and
pq
are the admittance magnitude and the admittance
angle respectively.
p
and
q
are the voltage angle at bus p and q, respectively;
where p = 1, 2, . . . , n, q = 1, 2, . . . , n.
The physics of the system as well as the desired voltage set points are
reected throughout the system by the load ow equations. The power ow
equations require that the sums of the net injection of real and reactive power
at each bus are zero, and thus enforce the physics of the power system.
The size of each reactive bank in the power system is represented by decision
vector B, see [6] for example:
B = [B
1
, B
2
, . . . , B
i
, . . . , B
n
] , B
i
R, |B
i
| B
imax
in order to represent the amount of reactive compensation to be allocated at
each busbar i.
(RPC) can therefore be seen as a complex, comprehensive optimization
problem dealing with multiple nonlinear functions with multiple local minima
which may be ill-dened and nonlinear constraints which lead to a nonconvex
Pareto-optimal front, (see [6] and [17]). The trust region concept could cover
these problems to get a nite set of Pareto optimals, where it is dicult for
the (DM) to choose the best compromise solution among these Pareto set.
In the following section we rewrite the reactive power compensation prob-
lem in mathematical form and using a weighting appoach to transform it to a
single objective optimization problem.
Utilization of trust region algorithm 2653
3 Multiobjective Formulation of the RPC Prob-
lem
The mathematical form of the reactive power compensation problem with n-
buses and m-generator is the following multi-objective optimization problem:
minimize f
1
=

n
i=1
B
i
minimize f
2
= P
g
j
P
l
j
minimize f
3
=
V
i
V

i

2
2
n
subject to 0 f
1
f
1max
,
0 B
i
B
imax
,
V
i
min
V
i
V
imax
,
P
g
j
min
P
g
j
P
g
j
max
,
P
p
= P
Gp
P
Cp
=

n
q=1
V
p
V
q
Y
pq
cos(
p

pq
),
Q
p
= Q
Gp
Q
Cp
=

n
q=1
V
p
V
q
Y
pq
sin(
p

q

pq
),
(1)
where i = 1, 2, ..., n and j = 1, ..., m.
Using a weighting approach to transform problem (1) to a single-objective
optimization problem which has the following form:
minimize f(x) = w
1
f
1
+ w
2
f
2
+ w
3
f
3
subject to 0 f
1
f
1max
,
0 B
i
B
imax
,
V
i
min
V
i
V
imax
,
P
g
j
min
P
g
j
P
g
j
max
,
P
p
= P
Gp
P
Cp
=

n
q=1
V
p
V
q
Y
pq
cos(
p

pq
),
Q
p
= Q
Gp
Q
Cp
=

n
q=1
V
p
V
q
Y
pq
sin(
p

q

pq
),
(2)
2654 B. El-Sobky, Y. Abo-Elnaga and H. Zahed
where x
T
= [B
1
, ..., B
n
, P
g
1
, ..., P
gm
, V
1
, ..., V
n
],

3
k=1
w
k
= 1, and w
k
0, k =
1, 2, 3.
The following section is devoted to presenting the detailed description of
our trust-region algorithm for solving problem (2).
4 Turst Region Algorithm Outline
In this section, we consider the following single objective optimization problem
minimize f(x)
subject to h(x) = 0,
g(x) 0,
(1)
where the functions f(x) :
2n+m
, h(x) :
2n+m

2n
, and g(x) :

2n+m

4n+2m+2
are twice continuously dierentiable. The Lagrangian
function associated with problem (1) is the function
l(x, , ) = f(x) +
T
h(x) +
T
g(x), (2)
where
2n
and
4n+2m+2
are the Lagrange multiplier vectors associ-
ated with equality and inequality constraints respectivly.
Following [8], we dene a 0-1 diagonal indicator matrix Z(x)
p p
, whose
diagonal entries are
z
i
(x) =

1 if g
i
(x) 0,
0 if g
i
(x) < 0,
(3)
where p = 2n + 4m + 2.
Using the above matrix, we transform problem (1) to the following equality
constrained optimization problem
minimize f(x)
subject to h(x) = 0,
Z(x)g(x) = 0.
(4)
Using a multiplier method, we transform the equality constrained optimiza-
tion problem (4) to the following unconstrained optimization problem
minimize (x, , ; ; r) = l(x, , ) +

2
Z(x)g(x)
2
2
+
r
2
h(x)
2
2
,
subject to x
2n+m
,
(5)
where is the positive parameter and r > 0 is a parameter usually called the
penalty parameter.
Utilization of trust region algorithm 2655
4.1 Computing a Trial Step
We compute the trial step s
k
by solving the following trust-region subproblem
minimize l
k
+l
T
k
s +
1
2
s
T
H
k
s +

k
2
Z
k
(g
k
+g
T
k
s)
2
+
r
k
2
(h
k
+h
T
k
s)
2
subject to s
k
,
(6)
where H
k
is the Hessian matrix of the Lagrangian function l(x
k
,
k
,
k
) or an
approximation to it. Since our convergence theory is based on the fraction of
Cauchy decrease condition, any method that computes the trial step in such a
way that the fractions of the Cauchy decrease can be used. Therefore, a dogleg
algorithm can be used to compute the trial step. More details can be found in
[8].
4.2 Testing the Step and Updating
k
Once the trial step is computed, it needs to be tested to determine whether it
will be accepted. To test the step, estimates for the two Lagrange multipliers

k+1
and
k+1
are needed. Our way of evaluating the two Lagrange multipliers

k+1
and
k+1
is presented in Step 5 of Algorithm (4.3) below.
Let
k+1
and
k+1
be estimates of the two Lagrange multiplier vectors. We
test whether the point (x
k
+ s
k
,
k+1
,
k+1
) will be taken as a next iterate.
The actual reduction in the merit function is dened as
Ared
k
= l(x
k
,
k
,
k
) l(x
k+1
,
k
,
k
)
T
k
h
k+1

T
k
g
k+1
+

k
2
[g
T
k
Z
k
g
k
g
T
k+1
Z
k+1
g
k+1
] +
r
k
2
[h
k

2
h
k+1

2
], (7)
where
k
= (
k+1

k
) and
k
= (
k+1

k
).
The predicted reduction in the merit function is dened to be
Pred
k
= q
k
(0) q
k
(s
k
)
T
k
(h
k
+h
T
k
s
k
)
T
k
Z
k
g
k
+
r
k
2
[h
k

2
h
k
+h
T
k
s
k

2
], (8)
where
q
k
(s) = l
k
+
x
l
T
k
s +
1
2
s
T
H
k
s +

k
2
Z
k
(g
k
+g
T
k
s)
2
. (9)
After computing a trial step and updating the Lagrange multipliers, the
penalty parameter is updated to ensure that Pred
k
0. To update r
k
, we use
a scheme that has the avor of the scheme proposed by [10]. This scheme is
described in Step 6 of Algorithm 4.3 below. After that, the step is tested to
know whether it is accepted. This is done by comparing Pred
k
against Ared
k
.
2656 B. El-Sobky, Y. Abo-Elnaga and H. Zahed
If
Ared
k
Pred
k
<
1
where
1
(0, 1) is a small xed constant, then the step is
rejected. In this case, the radius of the trust region
k
is decreased by setting

k
=
1
s
k
, where
1
(0, 1), and another trial step is computed using the
new trust-region radius.
If
Ared
k
Pred
k

1
, then the step is accepted. Our theory requires that at the
beginning of the next iteration,
k+1
must be greater than or equal to
min
,
where
min
is a positive constant chosen at the beginning of the algorithm.
That is,
k
can be reduced below
min
while nding an acceptable step. But,

k+1

min
is required at the beginning of the next iteration after accepting
the step s
k
.
Our way of evaluating the trial steps and updating the trust-region radius
is presented in Step 7 of Algorithm (4.3) below. After accepting the step, we
update the parameter
k
and the Hessian matrix H
k
. To update
k
, we use a
scheme suggested by [24]. This scheme is described in Step 8 of Algorithm 4.3
below.
Finally, the algorithm is terminated when either s
k

1
or
x
l
k
+
g
k
Z
k
g
k
+h
k

2
, for some
1
,
2
> 0,
4.3 Main Algorithm
A formal description of our trust-region algorithm for solving problem (2) is
presented in the following algorithm.
Algorithm (The Main Algorithm)
Step 0. (Initialization)
Given x
1

2n+m
. Compute Z
1
. Evaluate
1
and
1
(see Step 5
with k = 0 and
0
= (0, 0, ..., 0)
T
). Set
1
= 1, r
0
= 1,
1
= 1, and
= 0.1. Choose
1
,
2
,
1
,
2
,
1
, and
2
such that
1
> 0,
2
> 0,
0 <
1
< 1 <
2
, and 0 <
1
<
2
< 1. Choose
min
,
max
, and
1
such that
min

1

max
. Set k = 1.
Step 1. (Test for convergence)
If
x
l
k
+g
k
Z
k
g
k
+h
k

2
, then terminate the algorithm.
Step 2. (Compute a trial step)
a)Compute the step s
k
by solving (6)
b) Set x
k+1
= x
k
+ s
k
.
Step 3. (Test for termination)
If s
k

1
, then terminate the algorithm.
Utilization of trust region algorithm 2657
Step 4. (Update the active set)
Compute Z
k+1
.
Step 5. (Compute the Lagrange multipliers
k+1
and
k+1
)
a) Compute
k+1
by solving
minimize (f
k+1
+h
k+1

k
+g
k+1
Z
k+1
)
2
subject to Z
k+1
0,
(10)
and set the rest of the components of
k+1
to zero.
b) If f
k+1
+h
k+1

k
+g
k+1
Z
k+1

k+1
,
then set
k+1
=
k
.
Else, compute
k+1
by solving
minimize f
k+1
+g
k+1

k+1
+h
k+1

2
. (11)
End if.
Step 6. (Update the penalty parameter r
k
)
a) Set r
k
= max(r
k1
,
2
k
).
b) If Pred
k

r
k
4
[h
k

2
h
k
+h
T
k
s
k

2
],then set
r
k
=
4[q
k
(s
k
) q
k
(0) +
T
k
(h
k
+h
T
k
s
k
) +
T
k
Z
k
g
k
]
h
k

2
h
k
+h
T
k
s
k

2
+.
End if.
Step 7. (Test the step and update the trust-region radius)
If
Ared
k
Pred
k
<
1
.
Reduce the trust-region radius by setting
k
=
1
s
k

and go to step 2.
Else if
1

Ared
k
Pred
k
<
2
, then
Accept the step: x
k+1
= x
k
+ s
k
.
Set the trust-region radius:
k+1
= max(
k
,
min
).
Else
Accept the step: x
k+1
= x
k
+ s
k
.
Set the trust-region radius:
k+1
= min{
max
, max{
min
,
2

k
}}.
2658 B. El-Sobky, Y. Abo-Elnaga and H. Zahed
End if.
Step 8. (Update the parameters
k
and
k
)
a) Set
k+1
=
k
and
k+1
=
k
.
b) If
Tpred
k

T
k
(h
k
+h
T
k
s
k
)
T
k
Z
k
g
k
<
k
g
k
Z
k
g
k
min{g
k
Z
k
g
k
,
k
},
then set
k+1
= 2
k
and
k+1
=
1
2

k
.
End if.
Step 9. Set k = k + 1 and go to Step 1.
5 Implementation of the Proposed Approach
To investigate the eectiveness of the proposed approach the described method-
ology is applied to the standard IEEE 30-bus 6-generator test system, see gure
(1). In [25] we can see a detailed presentation of the data for this system. In
this work, our program was written in MATLAB and run under MATLAB
Version 7 with machine epsilon about 10
16
.
Given a starting point x
1
, we chose the initial trust-region radius to be

1
= max(s
cp
1
,
min
), where
min
was taken to be
min
= 10
3
. We chose the
maximum trust-region radius to be
max
= 10
5

1
.
The values of the constants that are needed in Step 0 of Algorithm (4.3)
were set to be
1
= 10
4
,
2
= 0.5,
1
= 0.5,
2
= 2,
1
= 10
8
,
2
=
10
8
, and = 0.1. For computing the two components of the trial steps, we
used the dogleg algorithm. Successful termination with respect to our trust-
region algorithm means that the termination condition of the algorithm is met
with
2
= 10
8
. On the other hand, unsuccessful termination means that the
number of iterations is greater than 300, the number of function evaluations
is greater than 500, or the length of the trial step is less than
1
= 10
8
.
6 Results and Discussions
Here we discuss the eects of changing weights on active power losses, the
compensation cost, as well as the average voltage deviation. As one weight is
changed linearly in each case, the other two weights are generated randomly
thus that: w
1
+w
2
+w
3
= 1. To obtain the best compromise for the operating
point we observed the corresponding values of the weights for the values of
f
1
(.), f
2
(.) and f
3
(.).
The three tables (1), (2) and (3) show the values of the weights for the
three cases which were studied, where in each case, one weight is changed
Utilization of trust region algorithm 2659
Figure 1: IEEE 30-bus Test System.
2660 B. El-Sobky, Y. Abo-Elnaga and H. Zahed
linearly taking six values, while the other two weights are obtained using the
relation w
1
+w
2
+w
3
= 1. Figures (2), (3) and (4) show the objective functions
obtained from the six solutions corresponding to the six weights compared to
the weights for the three cases. Thus, we observe that:
1- We obtained the highest cost at the lowest w
1
, while the highest value of
w
1
gave us the lowest cost.
2- The change of weight in w
2
and w
3
has less eect on active power losses and
average voltage division, respectively, than the change of the weight w
1
which
has a stronger eect on compensation cost.
3- In the light of these results we recommended to choose w
1
around 0.6 because
the change of the cost corresponding to values of w
1
higher than 0.6 is not
signicant.
7 Conclusions
In this paper we use a trust region method to solve the reactive power com-
pensation problem formulated as multiobjective optimization problem with
competing amounts of reactive compensation devices, active power losses and
voltage deviation. We have arrived at the conclusion that this approach is a
new technique for the numerical, parametric study of RPCs, when the prob-
lem is complex and has many real world applications. We have solved the
RPC, by applying the proposed technique while considering three objectives
simultaneously. Non-convex multiobjective optimization problems have been
eciently solved using the proposed approach. Our approach is interactive
because it allows the decision maker to specify the relative weights of criterion
importance which show the degree to which the objectives have been satised.
8 Acknowledgements
First and foremost, we give thanks to God for the grace and mercy he has
shown us. We depend on Him and we could not have accomplished this work
without Him.
Words cannot express our deep gratitude to The Deanship of Scientic
Research, Taibah University, KSA for supporting this project by grant No.
1432/79.
Utilization of trust region algorithm 2661
Run w
1
w
2
w
3
1 0 0.5721 0.4279
2 0.2000 0.6705 0.1295
3 0.4000 0.2118 0.3882
4 0.6000 0.2636 0.1364
5 0.8000 0.0759 0.1241
6 1.0000 0 0
Table 1: Dierent weights (w
1
is changed linearly).
Run w
1
w
2
w
3
1 0.6028 0 0.3972
2 0.5676 0.2 0.2324
3 0.4573 0.4 0.1427
4 0.2218 0.6 0.1782
5 0.1468 0.8 0.0532
6 0 1.0000 0
Table 2: Dierent weights (w
2
is changed linearly).
Run w
1
w
2
w
3
1 0.7477 0.2523 0
2 0.2994 0.5006 0.2
3 0.1576 0.4424 0.4
4 0.1354 0.2646 0.6
5 0.1976 0.0024 0.8
6 0 0 1
Table 3: Dierent weights (w
3
is changed linearly).
2662 B. El-Sobky, Y. Abo-Elnaga and H. Zahed
1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Run
w
e
i
g
h
t
e
s
(
w
1
,
w
2
,
w
3
)


w1
w2
w3
(a) Dierent weights (W
1
is changed linearly)
1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
0
0.005
0.01
0.015
0.02
0.025
f
2
,
f
3
Runs
Labeling plotyy


1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
0.1
0.15
0.2
0.25
0.3
0.35
f
1


f2
f3
f1
(b) Plot showing (f
1
, f
2
and f
3
)
Figure 2: Plot showing best compromise solution for dierent weights in 6 runs
of table 1.
Utilization of trust region algorithm 2663
1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Run
w
e
i
g
h
t
e
s
(
w
1
,
w
2
,
w
3
)


w1
w2
w3
(a) Dierent weights (W
2
is changed linearly)
1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
0
0.005
0.01
0.015
0.02
0.025
f
2
,
f
3
Runs
Labeling plotyy


1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
0.1
0.15
0.2
0.25
0.3
0.35
f
1


f2
f3
f1
(b) Plot showing (f
1
, f
2
and f
3
)
Figure 3: Plot showing best compromise solution for dierent weights in 6 runs
of table 2.
2664 B. El-Sobky, Y. Abo-Elnaga and H. Zahed
1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Run
w
e
i
g
h
t
e
s
(
w
1
,
w
2
,
w
3
)


w1
w2
w3
(a) Dierent weights (W
3
is changed linearly)
1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
0
0.01
0.02
0.03
0.04
f
2
,
f
3
Runs
Labeling plotyy


1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
0.1
0.2
0.3
0.4
0.5
f
1


f2
f3
f1
(b) Plot showing (f
1
, f
2
and f
3
)
Figure 4: Plot showing best compromise solution for dierent weights in 6 runs
of table 3.
Utilization of trust region algorithm 2665
References
[1] M. A. Abido. Multiobjective evolutionary algorithms for electric power
dispatch problem. IEEE Trans. On Evolutionary Computation, Vol. 10,
No.3, ( 2006).
[2] B. Baran, J. Vallejos and R. Ramos. Multi-objective reactive power com-
pensation with voltage security. Proc. IEEE Transmission and Distribu-
tion Conference and Exposition, Latin America, Sao Paulo, Brazil, (2004).
[3] B. Baran, J. Vallejos, R. Ramos and U. Fernandez. Reactive power com-
pensation using multi-objective evolutionary algorithm. Proc. IEEE Porto
Power Tech 2001, Portugal, (2001).
[4] B. Baran, J. Vallejos, R. Ramos and U. Fernandez. Multi-objective re-
active power compensation. Proc. IEEE Transmission and Distribution
Conference and Exposition, Atlanta, USA, (2001).
[5] J. Carlisle, A. El-Keib, D. Boyd, and K. Nolan. A review of capacitor
placement techniques on distribution feeders. in Proc. IEEE 29 South-
eastern Symposium on system Theory (SSST97), (1997).
[6] K. Deb. Multi-objective optimization using evolutionary algorithms. Wi-
ley, NY, USA, (2001).
[7] M. Delfanti, G. Granelli, P. Marannino, and M. Montagna. Optimal
capacitor placement using deterministic and genetic algorithms. IEEE
Trans. Power Systems, Vol.15, No.3, pp. 1041-1046,( 2000).
[8] J. Dennis, M. El-Alem, and K. Williamson. A trust-region ap-
proach to nonlinear systems of equalities and inequalities. SIAM J
Optimization,9:291-315, (1999).
[9] H. Dommel, and W. Tinney. Optimal Power Flow Solutions. IEEE Trans.
Power Apparatus and Systems, vol. PAS-87, No.10, pp.1866-1876, (1968).
[10] M. El-Alem. A global convergence theory for a class of trust-region algo-
rithms for constrained optimization. PhD thesis, Department of Mathe-
matical Sciences, Rice University, Houston, Texas, (1988).
[11] B. El-Sobky. Arobust trust-region algorithm for general nonlinear con-
strained optimization problems. PhD thesis, Department of Mathematics,
Alexandria University, Alexandria, Egypt, (1998).
[12] B. El-Sobky. A global convergence theory for an active trust region al-
gorithm for solving the general nonlinear programming problem. Applied
Mathematics and computation archive,144 No.1:127-157,(2003).
2666 B. El-Sobky, Y. Abo-Elnaga and H. Zahed
[13] R. Fletcher. Practical methods of optimization. John Wiley and
Sons,Chichester, (1987),2nd ed.
[14] L. Furon, J. D. Pilgrim, C. Dabeedin, A. Chebbo, and R. K. Aggarwal.
Genetic algorithms for optimal reactivepower compensation on national
grid system. IEEE Trans. Power Systems, Vol. 20, No.1, pp. 493-500,
(2005).
[15] Y. Haimesy. Multiple-Criteria Decisionmaking Aretrospective analysis,
IEEE Transactions on systems, Man and Gyberrnetics Vol. 15, No.3, pp.
313-315, (1985).
[16] P. Kundur. Power System Stability and Control. New York: Mc Graw-
Hill, NY, USA, (1993).
[17] K. Miettinen. Non-linear multiobjective optimization. Dordrecht: Kluwer
Academic Publisher, (2002).
[18] K. Miu, H. Chiang and G. Darling. Capacitor placement, replacement
and control in large scale distribution systems by A-based two-stage al-
gorithms. IEEE Trans. Power Systems, Vol. 12, pp. 1160-1166, (1997).
[19] J. More and D. Sorensen. Computing a trust-region step. Siam.J on
scientic and statistical computing, 4:553-572,(1983).
[20] M. S. Osman, M. A. Abo-Sinna and A. A. Mousa. A solution to the
optimal power ow using genetic algorithms. Appl. Math. Comput, Vol.
155, pp. 391-405, (2004).
[21] D. F. Pires, C. H. Antunes and A. G. Martins. A tabu search multiobjec-
tive approach to capacitor allocation in radial distribution systems. MIC
2001, Porto, Portugal, pp. 169-174, (2001).
[22] D. Van Veldhuizen. Multi-objective evolutionary algorithms classica-
tions, analysis and new innovations. PhD. Dissertation, Faculty of Grad-
uate School of Engineering, Air Force Institute of Technology, (1997)
[23] B. Venkatesh, G. Sadasivam and Khan Abdullah. A new optimal reac-
tive power scheduling method for loss minimization and voltage stability
margin maximization using successive multi-objective fuzzy LP technique.
IEEE Trans. Power Systems, Vol. 15, (2000).
[24] Y. Yuan. On the convergence of a new trust region algorithm. Numer.
Math. 70:515-539, (1995).
[25] R. Zimmerman, D. Gan, MATPOWER: A Matlab power system simula-
tion package. Available: http://www.pserc.cornell.edu/ matpower/.
Utilization of trust region algorithm 2667
Received: December, 2011