Electr Eng (2010) 92:227237

DOI 10.1007/s00202-010-0179-x

ORIGINAL PAPER

Allocation of multi-type FACTS devices using multi-objective

genetic algorithm approach for power system reinforcement
M. Gitizadeh

Received: 24 February 2010 / Accepted: 28 September 2010 / Published online: 15 October 2010
Springer-Verlag 2010

Abstract This paper presents a novel approach to find optimum locations and capacity of flexible alternating current
transmission systems (FACTS) devices in a power system
using a multi-objective optimization function. Thyristor controlled series compensator (TCSC) and static var compensator (SVC) are the utilized FACTS devices. Our objectives are:
active power loss reduction, new introduced FACTS devices
cost reduction, voltage deviation reduction, and increase on
the robustness of the security margin against voltage collapse. The operational and controlling constraints, as well
as load constraints, are considered in the optimum allocation. A multi-objective genetic algorithm (MOGA) is used
to approach the Pareto-optimal front (non-dominated) solutions. In addition, the estimated annual load profile has been
utilized in a sequential quadratic programming (SQP) optimization sub-problem to the optimum siting and sizing of
FACTS devices. IEEE 14-bus Network is selected to validate
the performance and effectiveness of the proposed method.
Keywords FACTS devices allocation
Multi-objective optimization Multi-objective genetic
algorithm

CMvar_SVCi
CMvar_TCSC j
dvi
DM
f1 , f2 , f3
JL
Ke
n facts , n plant
Plossi (x, u, z)
Ppeak (x, u, z)
TCSC , Q TCSC
Pinjf
injf
TCSC , Q TCSC
Pinjt
injt

P0 , Q 0
p f , q f , pt , q t

Symbols
A
Power plant installation cost in
($/kW)
B
Refundable investment rate in (percent)
BSVC j Susceptance of jth SVC in (pu)

M. Gitizadeh (B)
School of Electrical and Electronics Engineering,
Shiraz University of Technology, Shiraz, Islamic Republic of Iran
e-mail: gitizadeh@sutech.ac.ir

Pm
, S limit
S initial
j
j
SSVCi
STCSC j
u
vi
viideal
x
Xc
X TCSCi

Cost of one Mvar related to ith SVC in

($/Mvar)
Cost of one Mvar related to jth TCSC
in ($/Mvar)
Maximum voltage violation tolerance
(percent)
Decision Maker
Problem objective functions
A set contains all load buses
Active power cost in ($/kWh)
Life times of FACTS devices and power
plants, respectively in (year)
Active power loss of ith load level from
system annual load curve in (kW)
Peak point power generation in year of
study in (kW)
Injected active and reactive power at
bus f in (pu)
Injected active and reactive power at
bus t in (pu)
Prescribed real and reactive loads at
rated (normal) voltage in (pu)
Constants that reflect the load-voltage
characteristics at buses f and t
Mutation rate [0, 1]
Demands related to load bus j at initial
and limit (critical) states (MVA)
Complex power of ith SVC in (MVA)
Complex power of jth TCSC in (MVA)
Control variables vector
Voltage of bus i in (pu)
Ideal voltage of bus i in (pu)
State variables vector
Magnitude of X T C SC in (pu)
Reactance of ith TCSC in (pu)

123

228

Electr Eng (2010) 92:227237

Vector containing amount and type

of FACTS devices
Set of feasible solutions

1 Introduction
These days, high efficiency, maximum reliability, and security in the design and operation of power systems are more
important than ever before. The difficulties in constructing
new transmission lines due to limits in rights for their paths
make it necessary to utilize the maximum capacity of transmission lines. Therefore, it is difficult to provide voltage stability, even in normal conditions. The fact that the main duty
of generation units is based on the active power generation
requirements rather than the reactive power compensation
makes the problem more serious.
Flexible alternating current transmission systems
(FACTS) devices, modern active and reactive power compensators can be considered as viable and feasible options for
satisfying the voltage security constraints in power systems,
since their response to perturbations in urgent circumstances
is fast, their performance in normal conditions is flexible, and
their operation can fit the dynamic situations.
It is well documented in the literature that the effectiveness of FACTS controllers mainly depends on their locations
[1]. According to the characteristics of FACTS devices, various criteria have been considered in the allocation problem.
Some of the reported objectives in the literature are: static
voltage stability enhancement [25], violation diminution of
the line thermal constraints [6], network loadability enhancement [7,8], loss reduction [9], voltage profile improvement
[7], power plants fuel cost reduction using optimal power
flow [10], and economical approach which has minimized
the overall system cost function [11]. It should be noted that
each of the mentioned objectives improves the power system
network operation and reaching these objectives is desirable
in all power system networks. But improvement in one objective does not guarantee the same improvement in others. For
instance, satisfying the voltage magnitude constraint does
not result in the satisfaction of the voltage stability requirement [12]. Also, it is obvious that the minimum power loss
leads to power system lines optimum operation, whereas it
may exacerbate the static voltage stability limit. Therefore,
none of the mentioned technical objectives can be neglected
in FACTS devices allocation. On the other hand, allocation of
the unlimited FACTS devices due to one or more objectives
without considering the cost of the devices can not be justified
despite the assumption in [7]. Therefore, both technical and
economical objectives should be involved in the FACTS allocation problem. In previous efforts to approach these objectives, some simplifications have been made such as allocation
based on decoupled active and reactive components [5], the

123

definition of the cost function without including the interest rate, and active power loss price [11]. In [13], although
a multi-objective genetic algorithm (MOGA) approach has
been implemented for FACTS devices allocation, only two
objectives with different dimensions, including line overload
and voltage violation reduction, have been simplified and
augmented to constitute a single objective function. In addition, in economic objective function definition, the interest
rate has not been included. These assumptions cause some
problems such as the inability to use all achievable advantages of FACTS devices, impractical allocation results, and
inaccurate solution of the problem.
This investigation attempts to improve the previously
mentioned researches in the field of FACTS devices allocation in power systems. This is done by considering static voltage stability enhancement, power loss reduction, and
voltage profile improvement as the allocation objectives;
and that FACTS devices investment cost reduction considers interest rates, simultaneously. Therefore, multi-objective
optimization without simplification has been used in this
paper in an attempt to find a logical solution to the allocation problem. Despite previous works, and for approaching a
practical solution, estimated annual load profile has been considered for calculating power losses and voltage violation.
The utilized FACTS devices are thyristor controlled series
compensator (TCSC) and static var compensator (SVC).
Pareto [14] has proposed an optimal set, entitled a Paretooptimal set, in which every member of the set is considered to
be the best solution of the corresponding problem and other
design attributes cannot lead to a better solution. From the
perspective of optimality, it is sometimes interesting to find a
Pareto-optimal set of alternatives over the whole space of an
optimization problem. Note that without considering preference information to rank competing attributes, a solution in
a Pareto-optimal set is not believed to be superior compared
to others in the set.
Here, an approach based on multi-objective genetic
algorithm (MOGA) is used to compromise between contradictory objectives. The optimization procedure is capable of taking all the predetermined objective values defined
by the designer. In addition, in order to implement the
estimated annual load profile for accurately finding the optimum location and capacity of FACTS devices, sequential
quadratic programming (SQP) optimization sub-problem has
been used as a part of the overall optimization procedure
which is mainly based on MOGA.
The article is organized as follows: Sect. 2 is devoted to
the mathematical concept of the multi-objective allocation.
Section 3 describes the models of TCSC and SVC that have
been used for static security enhancement. The implemented
optimization procedure has briefly been described in Sect. 4.
The results achieved by applying the proposed method on
IEEE 14-bus Network are presented and analyzed in Sect. 5.

Electr Eng (2010) 92:227237

229

The locations and rating of the nominated devices that satisfy

the mentioned objectives are also determined in this section.
Finally, Sect. 6 reports the conclusions.

can be secured against voltage collapse. The security rate of

a system according to the critical state can be expressed as
follows [12],

SM =

2 Problem formulation and objective function

Three objective functions have been considered in this article. The first one is related to the active power loss, investment cost and peak point power generation. This objective
minimizes the active power loss cost, investment cost of proposing FACTS devices and peak point power generation. It
can be expressed as

(Plossi (x, u, z)Ti )
f 1 (x, u, z) = K e
i

+K i Cinvestment (z) + K p Ppeak (x, u, z)

(1)
where Cinvestment (z) is defined as follows,

Cinvestment (z) =
CMvar_SVCi SSVCi
i

CMvar_TCSC j STCSC j

(2)

where SSVCi and STCSC j are complex powers of ith SVC and
jth TCSC, respectively, and CMvar_SVCi and CMvar_TCSC j
are the cost of one Mvar related to ith SVC and jth TCSC,
respectively and are determined as [11],
2
CMvar_TCSC = 1.5STCSC
713STCSC + 153750
2
CMvar_SVC = 0.3SSVC
305SSVC + 127380

(4)

The next objective function is related to the security margin

of the system. This objective function depends on the static
voltage stability and investigates how the risk of voltage collapse is alleviated. Voltage collapse means a system is unable
to provide the load demand and this situation is considered to
be a critical state. By knowing this critical state, the system

S initial
j

jJL
limit
S
jJL j


jJL

S initial
j

jJL

Slimit
j

f 2 (x, u, z) = 1 S M = 

(5)

(6)

The minimization of this objective function causes voltage

collapse to be avoided.
The third objective function is in regards to the voltage
violation of the system. This voltage violation is defined for
each bus as follows [15],



 vi viideal  dvi
and
V Di =
vi
(7)

0 if x < 0
(x) =
x otherwise
Therefore, the third objective function is
f 3 (x, u, z) =

Ki =

S limit

j


S M takes a value between zero and one for a system with

normal operating condition. A negative value of S M means
the system cannot provide the initial load, and the voltage
will definitely collapse.
Since minimization is the aim of the optimization rather
than maximization, the objective function is rewritten as

(3)

It is noted that the comparison between power loss cost reduction and devices investment cost should be carried out in the
same year as the allocation study. Therefore, after the calculation of power loss according to the load curve of the
mentioned year, other costs such as the necessary investment
of new devices and benefits from peak point power generation reduction on the basis of interest rate, life time of new
devices and power plants are combined in a single objective
function. This is carried out using K p and K i factors with
the following definitions,
(1 + B)n facts B
(1 + B)n facts 1
(1 + B)n plant B
Kp = A
(1 + B)n plant 1

jJL


iJL




  vi v ideal  dvi
i
V Di =
vi
iJL

(8)
Minimization of this objective function forces the voltages
to remain in the specified range.
In the proposed multi-objective optimization, some constraints such as compromising between active and reactive
powers of load buses, permitted range of active and reactive generating powers of power plants, allowed tap range
of transformers, maximum power transmission of lines, and
permitted range of FACTS devices have been considered.
The goal of the problem is to find an optimum configuration,
*, among the feasible configurations, , through installing new devices or only on the basis of current devices in
such a way that all objective forces become optimum and the
defined nonlinear constraints are satisfied. The mathematical
description can be written as
min { f 1 (x, u, z), f 2 (x, u, z), f 3 (x, u, z)}

u, z

(9)

where is the set of feasible solutions.

123

230

Electr Eng (2010) 92:227237

Vf
If

Yft

Ift

TCSC
Pinjf
TCSC
Qinjf

Yf0

Vt

I tf

It

Yt0
Bus t

Bus f

TCSC
Pinjt
TCSC
Qinjt

Also, Z = R + j X is transmission line impedance, X c is

the magnitude of X T C SC and f t = f t = t f , Y f f =
Ytt = G f f + j B f f = Y f t , Y f t = Ytf = G f t + B f t .
According to Fig. 2, the drawn current by SVC can be
expressed as
ISVC = j BSVC Vk

Fig. 1 Injection power model of a TCSC

Fig. 2 Variable susceptance of
a SVC [17]

Vk
I SVC

(15)

Reactive power drawn by SVC that is the same as the injected

power to bus k is written as
Q SVC = Q k = BSVC Vk2

BSVC

(16)

Also, in using [12], the following extra constraints are considered for determining the security margin, while t and f
belong to JL ,
3 TCSC and SVC models: modification on voltage
security equations

pf

g f = P0 f V f

+ Pinjf +

n




V f V j Y f j cos f j f j

j=1

There are two possible characteristics for TCSCs, capacitive and inductive, to increase or decrease the transmission
line reactance. These devices can cause an increase in the
transmission power capacity of lines, static voltage security
margin enhancement, voltage profile improvement, and
decreasing power loss (power division between parallel
lines). SVCs also have capacitive and inductive characteristics and are predominantly utilized to improve and amend
voltage in static and dynamic conditions, reduce reactive network power loss, and enhance the static voltage security margin. In order to use TCSCs and SVCs to satisfy the mentioned
allocation criteria, the injection power model and variable
susceptance model shown, respectively, in Figs. 1 and 2 have
been considered. Figure 1 shows a lumped model of compensated linek between buses t and f . The injected active
and reactive powers to the mentioned buses are as follows
[16]:
TCSC
= G f f V f2 + (G f t cos f t + B f t sin f t )V f Vt
Pinjf

(10)
Q TCSC
injf

B f f

V f2

+ (G f t

sin f t

B f t

cos f t )V f Vt
(11)

TCSC
= G tt Vt2 + (G t f cos t f + Btf sin t f )V f Vt
Pinjt

=0

(17)
p

gt = P0t Vt t + Pinjt +

n




Vt V j Yt j cos t j t j

j=1

=0
q

h f = Q 0 f V f f + Q injf +

n


(18)


V f V j Y f j sin f j f j

j=1

=0
q

h t = Q 0t Vt t + Q injt +

n


Vt V j Yt j sin t j t j

(19)


j=1

= 0[t, f ] JL

(20)

These constraints are related to the power balance in load

buses in locations where injection power exists. P0 V p and
Q 0 V q represent the voltage dependency of loads and p, q
{0, 1, 2}.
Note that the minimum and maximum constraints of
TCSC and SVC values should be imposed to determine the
security margin,
min
max
X TCSCi X TCSCi
i = 1, . . . , n TCSC
X TCSCi
min
BSVC
j

BSVC j

max
BSVC
j

j = 1, . . . , n SVC

(21)
(22)

(12)
Q TCSC
injt

Btt Vt2

+ (G t f

sin t f

Btf

cos t f )V f Vt

4 Multi-objective genetic algorithm approach

(13)
4.1 Multi-objective optimization

where G f t and B f t are defined as

G f t
B f t

X c R(2X + X c )
=
2
(R + X 2 )(R 2 + (X + X c )2 )
X c (R 2 X (X + X c ))
=
(R 2 + X 2 )(R 2 + (X + X c )2 )

123

A general multi-objective design problem is expressed by

(23). It is assumed that there exist k objective functions we
want to minimize [18,19].
(14)

Min F(X ) = ( f 1 (X ), f 2 (X ), . . . , f k (X ))T

X S

(23)

Electr Eng (2010) 92:227237

231

f2
33

3
2

1
1

Fig. 3 Parameter/solution and Pareto surface for a two-dimensional

problem with two objectives [18]

A general design problem, F is non-linear and multi-modal,

and S might be defined by non-linear constraints containing
both continuous and discrete member variables.
Definition 1 Dominating: Vector w is said to dominate vector y, if:
i (1, 2, . . . k) : f i (w) f i (y) and
j (1, 2, . . . k) : f j (w) < f j (y)

(24)

Definition 2 Pareto Optimality: Vector X S is a Paretooptimal solution, if and only if there is no vector Y S,
for which F(Y ) = ( f 1 (Y ), f 2 (Y ), . . . , f k (Y )) dominates
F(X ) = ( f 1 (X ), f 2 (X ), . . . , f k (X )).
The space in R k is well-known as the Pareto-optimal frontier
, and is formed by the objective vectors of Pareto-optimal
solutions. Obviously any solution needs to be an element of
the Pareto-optimal set if possible. Other terminologies such
as non-dominated or efficient solutions are also used in literature instead of Pareto-optimal solutions. A graphical representation of this concept is depicted in Fig. 3.
In Fig. 3, f 1 , f 2 , . . . , f k will be used to denote the individual minima of each respective objective function, and the
utopian solution is defined as F = ( f 1 , f 2 , . . . , f k )T . As
F simultaneously minimizes all objectives, it is an ideal
solution that is rarely feasible.
4.2 Optimization approach
Pareto-optimal front solutions, for compromising between
conflicting objectives, can be obtained using multiobjective genetic algorithm. A genetic algorithm is a search
technique based on a specific class of evolutionary
algorithms. It is capable of solving various kinds of constrained/unconstrained optimization problems in which the
objective function is discontinuous, non-differentiable,
stochastic, or highly nonlinear. Standard optimization
algorithms such as gradient based methods are not appropriate for such problems. GAs use operators inspired by evolutionary biology such as mutation, natural selection, and
crossover [20].

2
1
f1

Fig. 4 Ranking based on non-dominated sorting

GA-based multi-objective optimization approaches can

generally be categorized as non-Pareto and Pareto-based
methods. Although non-Pareto techniques usually converge
to a subset of the Pareto-optimal frontier, a significant portion of the Pareto set is left unexplored. The so-called Paretobased techniques are generally categorized as non-dominated
and Pareto-storing approaches.
The concept of the non-dominated sorting for ranking a
search population based on Pareto optimality has been presented in [20]. In this approach, the non-dominated individuals in the populations are identified, given the rank 1
and purged from the population. At the next stage of the
algorithm, the same procedure is repeated for the reduced
population, which means the non-dominated individuals in
the reduced population are found, given the rank 2 and eliminated from the population. This process, as shown in Fig. 4,
is repeated until all the individuals are ranked.
On the other hand, Foseca and Fleming [21] have proposed
the so-called Pareto-optimal storing procedure in which ranking is carried out based on the degree of dominance of each
individual. In other words, the number of population members that dominate an individual defines the ranking of the
individual (see Fig. 5). The rank 1 is given to those nondominated individuals. In order to score individuals in the
population, the rankings need to be scaled. To reduce the
solution set to those that satisfy certain attribute values, goal
levels can also be utilized.
Advantages of the Pareto-storing procedure related to the
non-dominated sorting approach are the fast selection and
convergence. Therefore, Pareto-storing procedure in [21] has
been implemented in this paper.
Here, two-point crossover and roulette wheel selection
[20] have been utilized to generate the next generation. Additionally, In order to prevent fast convergence of the population to a specific value and getting stuck in a local optimum,
adaptive mutation rate Pm has been used.

123

232

Electr Eng (2010) 92:227237

f2

objective function f 3 , is calculated through the sum of each

load level optimum voltage deviation and peak load voltage deviation. Computing all objectives, one can find the fitness values for each chromosome in the population based on
Pareto-optimal ranking. Next generation based on GA operators will be executed until stop criteria is satisfied.

10

4
1
2

4
3

1
1

5 Results
1

f1
Fig. 5 Ranking based on Pareto-optimal storing

FACTS_NUM

FACTS Type

FACTS_NUM

FACTS Value

FACTS_NUM

FACTS Location

Fig. 6 Formation of one chromosome

Each chromosome has been formed from the type of

FACTS devices, their value (reactance of TCSC candidate
lines and the susceptance of SVC candidate buses), and their
location as shown in Fig. 6. With this structure, specified
number of FACTS devices (equal to FACTS_NUM) including TCSC and SVC can be optimally allocated. This is possible for the devices to be selected from specific lines/buses
candidate locations.
The GA terminates when the maximum number of generations is reached. If the quality of the best member of the
population according to the problem objectives is not acceptable, the GA will be restarted or a fresh search initiated.
Figure 7 illustrates the MOGA optimization procedure. As
is clear from Fig. 7, after initialization and randomly generating the first population, the optimization proceeds to find
objective functions for each chromosome in the population.
In this stage, different load levels are taken into account to
consider the estimated annual load profile. It can be helpful to find accurate solutions when the optimization process
runs on a practical network. To find the investment cost of
TCSC and SVC, their capacities have to be known according
to Eq. (2). The capacity of TCSC and SVC is determined
through a sequential quadratic programming approach [22]
to have the optimum loss and voltage deviation in each load
level. The maximum TCSC and SVC capacity of all load
levels, in addition to each level of optimized cost of loss,
determines the f 1 objective function. With the updated TCSC
and SVC values, the security margin objective function f 2
computes just for peak load duration. The voltage deviation

123

The standard IEEE 14-bus test system has been used to

show the validation and effectiveness of the MOGA method.
Figure 8 shows the single line diagram of the test system. The
information related to lines, transformers, generators, synchronous condensers, network peak load, initial compensators, and lines nominal powers of the test system are available
in appendix Appendix A. The participation factors of generators are chosen according to their initial MW. In (17)(20),
loads are assumed to be independent of bus voltages
(pf = pt = q f = qt = 0) and increased uniformly to determine the stability limit. Table 1 lists the necessary information for economic study. Estimated annual load profile has
been determined in this Table to find the allocation results,
accurately.
In this study, all branches (except transformers) have been
nominated for TCSC installation; and all load buses have
been considered for SVC installation. TCSC compensation
degree constraints have been assumed to be 80% for TCSC in
capacitive mode and 20% in inductive mode [7]. Real rating
of TCSC can be determined using TCSC rating and sizing
method described in [23]. In addition, considering the voltage of 1 pu for all buses, the susceptance of SVC can be
changed between 2 and 2 pu in power base of 100 MVA.
The voltage magnitude of the buses should vary in the band of
0.95 and 1.05 pu. DM preferred goals for objective functions
f 1 , f 2 and f 3 have been initialized as f goal = [0.85,0.5,0].
The parameters of MOGA i.e. number of generations, size of
population and mutation rate are set to 70, 60 and 0.2, respectively. It must be mentioned that the mutation rate is increased
adaptively when the possibility of convergence into a local
Pareto surface is increased. In addition, two-point crossover
(crossover fraction is 0.8) and roulette wheel selection [20]
have been utilized to generate the next generation.
The results of allocation of two FACTS devices have
been denoted in Table 2. These results have been extracted
from Pareto-optimal solutions in the objective space. The
Pareto-optimal solutions in this Table are consisting of
FACTS devices type, their size and their locations for different load levels. Figure 9 shows Pareto-optimal solutions
in the objective space.
Among the Pareto-optimal solutions in Table 2, the solution number five has the least distance to our goals which
are f goal = [0.85, 0.5, 0]. In Table 3, optimum results in

Electr Eng (2010) 92:227237

233

Begin

Read information of population

size, mutation rate, crossover
rate, elite count, DM goals, and
maximum generation

Generate the first

population

Determine the vectors of TCSCs values and

corresponding locations: S1(1k) , M1(1k)
Determine the vectors of SVCs values and
corresponding locations: S2(1m), M2(1m)

Use SQP optimization method to compute

optimum loss cost (LCL1,LCL2,) and voltage
deviation(VDL1,VDL2,) by finding optimum
values for S1 & S2 in each load level except peak
period. Let S1L1, S1L2, are optimal values for
TCSC in M1 locations and S2L1, S2L2, are
optimal values for SVC in M2 locations
S1 , S1L1,

Update current chromosome with

C1 in locations of M1 for TCSC &
C2 in locations of M2 for SVC

Compute TCSC & SVC

investment cost and augment
it to LCL1,LCL2, to find f1.

Compute f2 and voltage deviation

(VDP) using the load of peak period.
f3 is [(VDL1,VDL2,), VDP]

Store objective values for each

choromosome

Compute the fitness values for each

chromosome in the population based on
Pareto-optimal ranking
(lower rank receives higher fitness value)

S1L2,

Compare each columns of

to find the maximum TCSCs values and

constitute vector C1(1k).
Compare each columns of S2 , S2L1, S2L2,
to find the maximum SVCs values and
constitute vector C2(1m).

Update each chromosome in the population

and compute their fitness function

No

Yes

Select parents for mating

(Roulette wheel mechanism)

Apply crossover, mutation and elitism

(nearer to DM goals) operators to
create offspring

Insert offspring into population

and constitute new generation

Stop criteria
satisfied?

Reduce the Pareto-optimal

solutions to those that meet
the DM goals

Show results

End

Fig. 7 MOGA optimization procedure

13

Table 1 Information for economic study

14

12
11

10

9
7

Parameters

Values

Factor and duration of load level 1

0.81, 2,136 h

Factor and duration of load level 2

1.00, 2,832 h

Factor and duration of load level 3

0.90, 4,392 h

Interest rate

15%

Ke

0.1 $/kW h

Cost of power plant installation

150 $/kW

Life time of FACTS devices and power plants

30 years

Fig. 8 Standard IEEE 14-Bus test system

each load level based on solution number five have been presented. These results confirm that better system performance
could be achieved in all load levels after installing allocated
devices. It can be seen that in Table 4, after optimum allocation and installation of two FACTS devices based on fifth
Pareto-optimal solution, the cost of installation will be optimized to 1.0078 pu. In other words, reducing the cost of
the system performance (0.733% loss reduction and 0.033%

power generating reduction in the load peak), resulted in only

0.78% additional investment cost. Furthermore, the security margin and voltage profile have, simultaneously, been
improved compared to initial condition.
Table 4 shows little improvement in the security margin
of the system, however further improvement is not necessary
because of the initial proper robustness of the IEEE 14-bus
network against the voltage collapse (SM = 0.3286). This
conclusion is not valid for voltage violation which is 0.1381

123

234

Electr Eng (2010) 92:227237

Table 2 Pareto-optimal
solutions and their
corresponding objective values

Number

Initial bus Final bus Type

Mvar

Total cost 1SM VDi Loss (MW)

L1

L2

L3

Pareto-optimal solutions with MOGA

TCSC

3.925

SVC

8.96

SVC

112.2

10

SVC

7.97

2
3

1.0108

0.6486 0.2304 8.559 13.253 10.635

1.1802

0.5729 0.4829 9.953 14.338 11.906

TCSC

0.1632 1.0009

0.6711 0.1380 8.622 13.394 10.732

14

10

11

TCSC

0.3961

TCSC

4.196

TCSC

3.929

TCSC

3.833

4
5
6
7

SVC

5.833

SVC

134.36

SVC

4.407

SVC

6.67

TCSC

0.420

9
8

10

1.0286

0.6566 0.1340 8.850 13.753 11.019

1.0078

0.6679 0.0865 8.549 13.295 10.649

1.2365

0.5512 0.8734 10.818 15.112 12.719

1.0832

0.6036 0.3048 8.956 13.511 10.958

1.0000

0.6714 0.1381 8.622 13.393 10.732

Initial solution

1.0000

0.6714 0.1381 8.622 13.393 10.732

Base cost = 15.767 M$

0.2

2.5

Third Objective: Z

Third Objective

2
1.5
1
0.5

0.15

0.1

0.05
X: 1.008
Y: 0.668
Z: 0.0865

0
0.7

0
0.7
0.65

1.6
0.6

1.4
0.55

Second Objective

0.69

1.06

1.02

0.67

1.2
0.5

1.04

0.68

First Objective

Second Objective: Y

1
0.66

0.98

First Objective: X

Fig. 9 Pareto-optimal solutions in the objective space

initially. The fifth Pareto-optimal solution has, properly, been

reduced the voltage deviation to 0.0865 (37.36%).
Figure 10 shows the voltage profile during the peak period
after using the mentioned solution. Table 3 confirms that the
voltage deviation has been reduced in all load levels. This
is also true for the loss reduction. Therefore, optimization

123

procedure is able to locate the devices in such a way that

the system performance will be improved in all load levels.
Although GAs are considered to be time consuming methods, due to the off-line characteristic of planning problems
this deficiency has no negative effect on the optimization

Electr Eng (2010) 92:227237

235

Table 3 Optimum results in each load level based on the fifth pareto-optimal solution
Load levels

Before allocation
L1

L2

After allocation
L3

L1

Reduction in %

L2

L3

L1

L2

L3

Lossa (LSi )

0.0862

0.1339

0.1073

0.0855

0.1330

0.1065

0.812

0.672

0.746

Generated powera (PGi )

2.1841

2.7239

2.4383

2.1834

2.7230

2.4375

0.032

0.033

0.033

Voltage deviation

0.0621

0.0292

0.0468

0.0429

0.0158

0.0277

30.92

45.89

40.81

Base power: 100 MVA

Table 4 The results before and

after installation of two facts
devices based on the fifth
pareto-optimal solution

Base power: 100 MVA.

b Base cost: 15.767 M$

Before allocation

Objective values

After allocation

Reduction in %

Lossa

0.3274

0.3250

0.733

Peak point power generationa

2.7239

2.7230

0.033

Costb

0.85

1.0078

1SM

0.6714

0.5

0.6679

0.5213

VDi

0.1381

0.0865

37.36

Voltage Magnitude

1.06
1.055
1.05
1.045
1.04
1.035

Before Allocation

1.03

After Allocation

1.025
1.02
1.015
3

10 11 12 13 14 15

Bus Number

Fig. 10 Voltage magnitude of load buses in the peak period

procedure. Finding the optimum solution to simultaneously

reduce all the objectives in a FACTS devices allocation problem is really vital for the prospective system, and therefore, it
is worth spending more time on such an important decision.
A comparative study between the proposed method and previous studies in [3,4,6,7] reveals that, in order to carry out a
comprehensive study of FACTS devices allocation, it is feasible to satisfy all the objectives simultaneously. On the other
hand, unlimited FACTS devices to reach the maximum loadability of a network [7] cannot be practical, and it is possible
to use a limited number of devices according to economic
considerations.

considerations. One of these considerations is using the estimated annual load curve, which makes the allocation more
accurate. In contrast to some previous researches, the cost
objective function is considered along with other objectives to reach a precise and practical solution. In addition, the MOGA method has been utilized to find the
Pareto-optimal solutions which help the designer to select
the best compromise between conflicting objectives based
on his/her preferred goals, even if the problem is nonconvex. According to the obtained results on the IEEE
14-bus network, the mentioned method results in the
satisfaction of such allocation objectives as power loss reduction, investment cost reduction, security margin improvement, and voltage violation alleviation. It is also concluded
that the investment of the FACTS devices is reduced to
0.78% of initial system cost due to providing better performance for the system. Besides, the cost reduction of peak
point power generation in this study implies that power
plant expansion, providing the demand load, can be postponed.
Acknowledgments The author would like to acknowledge the financial support from Fars Regional Electric Company (FREC is one of the
sixteen major RECs in Iran). The author also appreciates Dr. A. Salehi,
A. Sadrzadeh, and J. Zeraatpisheh for their guidelines

Appendix A
6 Conclusion
In this paper a novel approach has been proposed to determine the optimum type, amounts and locations of TCSCs
and SVCs based on a multi-objective function. In this
method the allocation problem is investigated with practical

Initial normal and week operating conditions and the

branches data of the IEEE 14-bus test system have been presented in Tables 5, 6 and 7 based on the data mentioned in
[24]. In addition, the assumption for lines nominal power
ratings has been added into the branches data in Table 6.

123

236

Electr Eng (2010) 92:227237

Table 5 Estimation of initial

operating conditions of 14-bus
network

Bus number

Initial conditions

Initial MVar

Voltage

Bus power

Mag (pu)

Final
bus

R
(pu)

X
(pu)

P (MW)

Q(MVar)

1.06

1.045

4.983

21.7

12.7

1.01

12.725

94.2

19

1.018

10.313

47.8

3.9

1.02

8.774

7.6

1.6

1.07

14.221

11.2

7.5

1.062

13.36

1.09

13.36

1.056

14.939

29.5

16.6

19

10

1.051

15.097

5.8

11

1.057

14.791

3.5

1.8

12

1.055

15.076

6.1

1.6

13

1.05

15.156

13.5

5.8

14

1.036

16.034

14.9

Table 6 Branches data of 14-bus network

Initial
bus

Ang (pu)

B
(pu)

Table 7 Regulated bus data

Nominal
power
(MW)

TAP
position

Bus

MVar limits

MW limits

Min.

Min.

Max.

Specified voltage

Max.

2a

0.01938

0.05917

0.0528

300

40

50

140

0.05403

0.22304

0.0492

150

40

1.01

0.04699

0.19797

0.0438

150

24

1.07

0.05811

0.17632

0.034

150

24

1.09

0.05695

0.17388

0.0346

150

0.06701

0.17103

0.0128

150

0.01335

0.04211

150

0.20912

60

0.978

0.55618

36

0.969

0.25202

80

0.932

11

0.09498

0.1989

60

12

0.12291

0.25581

45

13

0.06615

0.13027

45

0.17615

45

0.11001

60

10

0.03181

0.0845

45

14

0.12711

0.27038

45

10

11

0.08205

0.19207

45

12

13

0.22092

0.19988

45

13

14

0.17093

0.34802

45

2.

3.

4.

5.
6.

7.

In this table, in the case of Branch 12, the presented data are for the
combination of the two parallel lines
8.

References
1. Okamoto H, Kurita A, Sekine Y (1995) A method for identification
of effective locations of variable impedance apparatus on enhance-

123

9.

1.045

ment of steady-state stability in large-scale power systems. IEEE

Trans Power Syst 10:14011407
Sharma NK, Ghosh A, Varma RK (2003) A novel placement
strategy for FACTS controllers. IEEE Trans Power Delivery 18:
982987
Yorino N, EL-Araby EE, Sasaki H, Harada SH (2003) A new formulation for FACTS allocation for security enhancement against
voltage collapse. IEEE Trans Power Syst 18:310
Chang CS, Huang JS (1998) Optimal multi-objective SVC planning for voltage stability enhancement. IEE Proc Generation
Transm Distrib 145:203209
Song SH (2004) Installation and operation of FACTS devices for
enhancing steady-state security. Electr Power Syst Res 70:715
Lu Y, Abur A (2002) Static security enhancement via optimal utilization of thyristor controlled series capacitors. IEEE Trans Power
Syst 17:324329
Gerbex S, Cherkaoui R, Germond AJ (2001) Optimal location of
multi-type FACTS devices in a power system by means of genetic
algorithms. IEEE Trans Power Syst 16:537544
Jurado F, Rodriguez JA (1999) Optimal location of SVC based on
systems loadability and contingency analysis. In: 7th IEEE International Conference on Emerging Technologies and Factory Automation
Singh SN, David AK (2000) Congestion management by optimal FACTS device location. In: IEEE International Conference

Electr Eng (2010) 92:227237

10.

11.

12.

13.

14.
15.

16.

on Electric Utility Deregulation, Restructuring and Power Technologies

Ongsakul W, Bhasaputra P (2002) Optimal power flow with
FACTS devices by hybrid TS/SA approach. Electr Power Energy
Syst 24:851857
Cai LJ, Erlich I, Stamtsis G (2004) Optimal choice and allocation
of FACTS devices in deregulated electricity market using genetic
algorithms. Power Syst Conf Expos IEEE-PES 1:201207
Obadina OO, Berg GJ (1988) Determination of voltage stability
limit in multi-machine power systems. IEEE Trans Power Syst
3:15451554
Radu D, Bsanger Y (2006) A multi-objective genetic algorithm
approach to optimal allocation of multi-type FACTS devices for
power systems security. In: IEEE Power Engineering Society General Meeting
Pareto V (1986) Cours dconomie Politique. Lausanne, Rouge
Chen YL, Liu CC (1994) Multi-objective VAR planning using
the goal-attainment method. IEE Proc Generation Transm Distrib
141:227232
Narayana PP, Abdel Moamen MA (2005) Power flow control and
solutions with multiple and multi-type FACTS devices. Electr
Power Syst Res 74:341351

237
17. Ambriz-Perez H, Acha E, Fuerte-Esquivel CR (2000) Advanced
SVC model for Newton-Raphson load flow and Newton optimal
power flow studies. IEEE Trans Power Syst 15:129136
18. Andersson J (2001) Multi-objective optimization in engineering
design applications to fluid power systems. Linkping Studies in
Science and Technology, Dissertations No. 675
19. Popov A (2005) Users manual of genetic algorithms for optimization programs for MATLAB
20. Goldberg DE (1989) Genetic Algorithms in search, optimization,
and machine learning. Addison-Wesley, Reading
21. Fonseca CM, Fleming PJ (1998) Multi-objective optimization and
multiple constraint handling with evolutionary algorithms - Part I:
a unified formulation. IEEE Trans Syst Man Cybern Part A Syst
Humans 28:2637
22. Gill PE, Murray W, Wright M (1981) Practical optimization. Academic Press, London
23. Kosterev DN, Mittelstadt WA, Mohler RR, Kolodziej WJ
(1996) An application study for sizing and rating controlled and
conventional series compensation. IEEE Trans Power Delivery
11:11051111
24. Freris LL, Sasson AM (1968) Investigation of the load flow problem. IEE Proc 115:14591469

123