Electrical Power and Energy Systems 32 (2010) 368374

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Improved particle swarm optimization applied to reactive power

reserve maximization
L.D. Arya a,*, L.S. Titare b,1, D.P. Kothari c
a

Department of Electrical Engg., SGSITS, 23-Park Road, Indore, MP 452 003, India
Department of Electrical Engg., Govt. Engineering College, Jabalpur, MP 456 010, India
c
Centre for Energy Studies, Indian Institute of Technology, Hauz Khas, New Delhi 110 016, India
b

a r t i c l e

i n f o

Article history:
Received 27 June 2006
Received in revised form 5 October 2009
Accepted 6 November 2009

Keywords:
Voltage stability
Particle swarm optimization (PSO)
Reactive power reserve
Schurs inequality
Proximity indicator

a b s t r a c t
This paper presents a new approach for scheduling of reactive power control variables for voltage stability enhancement using particle swarm optimization (PSO). Cost function selected is maximization of
reactive reserves of the system. To get desired stability margin a Schurs inequality based proximity indicator has been selected whose threshold value along with reactive power reserve maximization assures
desired static voltage stability margin. PSO has been selected because not only it gives global optimal
solution but also its mechanization is very simple and computationally efcient. Reactive generation participation factors have been used to decide weights for reactive power reserve for each of generating bus.
Developed algorithm has been implemented on 6-bus, 7-line and 25-bus 35-line standard test systems.
Results have been compared with those obtained using DevidonFletcherPowells (DFP) method.
2009 Elsevier Ltd. All rights reserved.

1. Introduction
Power system optimization problems including reactive power
optimization (RPO) have complex and non-linear characteristics
with large number of inequality constraints. Recently, as an alternative to the classical mathematical approaches, the non-traditional techniques [16] such as genetic algorithm, Tabu search,
simulated annealing, and PSO are considered practical and powerful solution schemes to obtain the global or quasi-global optimum
solution to engineering optimization problems. At times such
schemes are termed as heuristic optimization techniques [1]. PSO
has been selected as an optimization methodology in this paper
because its mechanization is extremely simple, robustness to control parameters and computational efciency when compared with
mathematical programs and other non-traditional algorithms.
Reactive power and voltage control [2], power system stabilizer
design [3], and dynamic security [4] studies are the areas to which
PSO has been successfully applied. Yoshida et al. [2] suggested a
modied PSO to control reactive power ow and alleviating voltage limit violations. The problem was a mixed-integer, non-linear
optimization problem with inequality constraints.

* Corresponding author. Tel.: +91 0731 2321016; fax: +91 07312432540.

E-mail addresses: ldarya@rediffmail.com (L.D. Arya), lstitare@yahoo.co.in
(L.S. Titare), dpk0710@yahoo.com (D.P. Kothari).
1
Tel.: +91 0734 2514246.
0142-0615/$ - see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijepes.2009.11.007

Availability of reactive power at sources and network transfer

capability are two important aspects, which should be considered
while rescheduling of reactive power control variables. Nedwick
et al. [5] have presented a reactive management program for a
practical power system. They have discussed a planning goal of
supplying system reactive demand by installation of adequately
sized and adequately located capacitor banks which will permit
the generating unit near to unity power factor. Vaahedi et al. [6]
developed a hierarchical optimization scheme, which optimized a
set of control of variables such that the solution satised a specied voltage stability margin. Menezes et al. [7] introduced a methodology for rescheduling reactive power generation of plants and
synchronous condenser for maintaining desired level of stability
margin. In Ref. [8] an algorithm for voltage stability enhancement
has been presented for rescheduling of reactive power control variables for voltage stability margin improvement using linearized
incremental model. Dong et al. [9] developed an optimized reactive
reserve management scheme using Benders decomposition
technique. Ant colony system algorithm is applied to the reactive
control problem in order to minimize real power losses, subject
to operating constraints over the whole planning period by
Vlachogiannis et al. [10]. Yang et al. [23] presented a technique
for reactive power planning based on chance constrained programming accounting uncertain factors. Generator outputs and load
demands modeled as specied probability distribution. Monte
Carlo simulation along with genetic algorithm has been used for
solving the optimization problem. Wu et al. [24] described an

L.D. Arya et al. / Electrical Power and Energy Systems 32 (2010) 368374

OPF based approach for assessing the minimal reactive power support for generators in deregulated power systems. He et al. [25]
proposed a method to optimize reactive power ow (ORPF) with
respects to multiple objectives while maintaining voltage security.
Varadarajan and Swarup [26] applied differential evolutionary
algorithm for optimal reactive power dispatch and voltage control.
Zhang and Liu [27] developed an algorithm for reactive power and
voltage control using fuzzy adaptive particle swarm optimization
(FAPSO). Zhang et al. [28] developed a computational method for
reactive power market clearing.
Reactive power reserve available at source is an important and
necessary requirement for maintaining a desired level of voltage
stability margin. Power network may have the transfer capability
of reactive power but if reserve is not available and reactive power
limit violation occurs than the static voltage stability limit may be
inadequate. Further reactive reserves available at sources will not
be of much help in maintaining desired level of stability margin,
if network transfer capability is limited. This paper proposes a
methodology for voltage stability enhancement accounting network loading constraint as well as optimizing reactive power reserves at various sources in proportion to their participation
factors decided based on incremental load model. Voltage dependent reactive power model has been used for determining reactive
power reserves, which utilizes eld heating as well as armature
heating limit. Section 2 explains problem formulation. Section 3
presents an overview of PSO technique. Section 4 presents implementation of the algorithm for optimizing reactive reserves. Section 5 gives results and discussions. Section 6 gives conclusions
and highlights of main contributions of the paper.
2. Problem formulation
2.1. Reactive reserve
The voltage stability enhancement problem is formulated as an
optimal search problem whose objective is two fold: (i) maximize
the reactive reserves based on the participation of reactive sources
for increased loading condition and (ii) maintaining the desired
stability margin with respect to current operating point. Reactive
power reserve is the ability of the generators to support bus voltages under increased load or disturbance condition. Amount of
reactive power, which can be fed to network, depends on present
operating condition, location of the source, eld and armature
heating of the alternators. Nature of the change in load scenario
also has impact on reactive reserves. Availability of reactive power
reserve of a generator is calculated using capability curves. For a given real power output the reactive power generation is limited by
both armature and eld heating limit [11]. Maximum reactive
power output with respect to eld current limit is expressed as:

Q g;max V 2g =X d

q
V 2g  E2max =X 2d  P2g

Maximum reactive power output Qg,max of the generator is

determined by internal maximum voltage Emax corresponding to
the maximum eld current. Thus maximum reactive power output
is determined not only on real power output Pg but also on terminal voltage Vg. Maximum reactive power output due to armature
current limitation as follows:

q
V 2g  I2g;max  P2g

369

than Qg,max. However if Qg reaches its limit the reactive reserve is set
to zero. The bus is treated as variable voltage and internal voltage of
the generator behind synchronous reactance is assumed constant.
This way voltage dependent reactive power limits have been accounted in a realistic way. In such situation Qg moves on the capability curve governed by Eqs. (1) and (2). Hence if Qg reaches the
boundary the reactive reserve is set to zero and then Qg varies as
a function of terminal (bus) voltage. Such modeling (voltage dependent reactive power limit) has been adopted by many researchers
[9,1719].
2.2. Proximity indicator based on Schurs inequality
It is assumed that load ow Jacobian at current solution point is
known. Following relation can be written based on Schurs inequality
[12]:

Skmax 6

s
X
s2i;j

i;j

where, Skmax is greatest eigenvalue of sensitivity matrix given, as

the inverse of load ow Jacobian, si,j is ijth element of sensitivity
matrix [S], which is inverse of load ow Jacobian. It is observed from
matrix theory that minimum eigenvalue magnitude of load ow
Jacobian is reciprocal of greatest eigenvalue of sensitivity matrix
[S]. Hence following relation follows:

s
X
Jkmin P 1=
s2i;j

i;j

Jkmin is minimum eigenvalue of Jacobian.

Right hand side of the above expression is lower bound on the
minimum eigenvalue and termed in further application of this paper as proximity indicator (s).
Under low loading condition elements of sensitivity matrix
are smaller and value of proximity indicator is large. As the load
on the system increases the value of proximity indicator decreases since element of sensitivity matrix (si,j) increases in magnitude. In the vicinity of collapse point the value of proximity
indicator practically becomes zero. Hence magnitude of s has
been used for voltage stability assessment and control in this paper. For secure operation a threshold value of proximity indicator must be maintained. Variation of proximity indicator can
be co-related with load on the system with the help of power
ow run. A magnitude of s which provides adequate voltage
stability margin (distance to voltage collapse point from current
total load) is selected as threshold value. This value is system
dependent.
Computation of proximity indicator requires Jacobian inversion, which is available directly at the end of current load ow
solution. Further computational efciency is achieved using sparsity and inversion using LU factorization [21,22]. Developed algorithm is for base point setting of reactive power control
variables.
2.3. Mathematical formulation

Ig,maxis maximum armature current of the generator. The reactive power reserve of the gth generator is then represented as:

The reactive reserve optimization problem is formulated as

search problem whose objective is to maximize the effective reactive reserve subject to various operating and stability constraints.
Objective function is given as follows:

Q g;max;res Q g;max  Q g

J Max

Q g;max

where, Qg,max is the smaller of the two values obtained from Eqs. (1)
and (2). Reactive reserve is calculated using relation (3) if Qg is less

W i  Q i;res

Above objective function is optimized subject to following

constraints:

370

L.D. Arya et al. / Electrical Power and Energy Systems 32 (2010) 368374

Step-2 A current velocity for ith agent in Kth iteration is given

as:

(i) Power ow equations:

P f V; d

Q gV; d
(ii) Inequality constraints on load bus voltages:

V min
6 V i 6 V max
i
i

i 2 NL

sth is threshold value of s proximity indicator

(iv) Reactive power generation constraint:
Q min
6 Q p 6 Q max
p
p

p 1; 2; . . . ; NG

U min
6 U i 6 U max
i
i

i 2 NC

K1

10

11

NL and NC denotes set of load buses and number of control

variables.
In objective function (6) Qi,res denotes reactive power reserve
available at ith generation bus and Wi is the weighting factor for
ith generator bus. Weighting factor Wi is selected based on the relative participation of each generator to reactive load increase in
specic direction in load parameter space. The bus, which participates to a smaller extent, is given higher weight and the bus participating to a greater extent should be given lesser weight such
that reserve at such bus is reduced and increased respectively.
Such weighting factors can be obtained by incremental power ow
equations with some algebraic manipulation. Following sensitivity
relation is obtained using incremental power ow equation:

SQDi @Q i =@Q d

12

Qi and Qd represent reactive power injection at ith generator bus

and total reactive power demand of the system. This sensitivity
represents change in reactive power injection at ith bus with respect to change in total reactive power demand of the system. Then
weighting factor for ith generator bus is given as:

Wi

!,
SQDp

SQDi

13

The search optimization problem is solved using particle swarm

optimization (PSO). Next section presents an overview of PSO
technique.
3. Particle swarm optimization technique: an overview
The PSO is a population-based optimization algorithm. Its population is called swarm and each individual is called a particle [13].
Each particle ies through the solution space to search for global
optimal solution. The mechanization of the PSO procedure is explained in following steps:
Step-1 A current position is an n-dimensional search space,
which represents a potential solution (particle or agent):

U Ki uKi;1 ; uKi;2 ; . . . ; uKi;n

i denotes ith particle and K denotes the iteration.

14

Ui

It is stressed here that Q max

is voltages dependent and
p
obtained using Eqs. (1) and (2). It is further claried here that
based on minimum rotor current limiter, the purpose of
Q min
p
which is to avoid very small rotor current (these may cause
problem for excitation systems) are of interest for synchronous compensators not for synchronous generators [17].
(v) Inequality constraint on control variables:

Step-3 At each iteration, the particle is updated by the following

relation:

qK1
W  qKi c1 r1 P Kbesti  U Ki c2  r2 GKbest  U Ki
i

(iii) Voltage stability constraint:

s P sth

qK
qKi1 ; qKi2 ; . . . ; qKin
i

U Ki qK1
i

15

P Kbesti

where,
is the best previous position of ith particle,
GKbest is the global best position among all the participation in the swarm from objective function viewpoint,
r1() and r2() are random digit generated from uniform
distribution [0, 1], W is an inertia weight that is typically
chosen in the range [0, 1]. A large inertia weight facilitates global exploration and a smaller inertia weight
tends to facilitate local exploration to ne-tune the current search area. Therefore the inertia weight W is an
important parameter for the PSOs convergence behavior.
A suitable value for the inertia weight usually provides
balance between global and local exploration abilities
and consequently results in a better optimum solution.
In view of this it has been suggested that iteration wise
the weight W is varied according to following relation
[1]:

W W max  W max  W min =NIT  NIT max

where NITmax is the maximum number of iteration supplied and NIT denotes current number of iteration. Wmax
and Wmin denote maximum and minimum values of inertia weights. Thus as iteration increases inertia weight
varies from Wmax say 2.0 to Wmin say 0.5, c1 and c2 are
acceleration constant selected in the range 12.
4. Implementation of PSO for reactive reserve optimization
problem
In this section PSO implementation to reactive reserve optimization as formulated in Section 2 has been explained. The process
of optimization is summarized as follows:
Step-1 Determination of initial population:
Each particle in the population consists of NC component
as reactive power control variables. Each reactive power
control variables for an agent is selected from a uniform
distribution between Ui,min and Ui,max satisfying the
inequality constraint. M such set of initial population
may be selected. The following procedure is adopted for
initialization of an agent:
(a) Set j = 1
(b) Select a random digit from [0, 1].
(c) Using the random digits create the value of Uj, satisfying the inequality constraint (11).
(d) If j P NC (total number of control variables) then go
to step (e) otherwise j = j + 1 and repeat from (b)
(e) Stop the initialization.
Step-2 This step consists of initialization of velocities of each
particle selected in step-1 above. Initial velocity of each
particle is also created at random. The velocity of element j of individual is generated at random within the
boundary given as follows:

371

L.D. Arya et al. / Electrical Power and Energy Systems 32 (2010) 368374

U min
 U 0ij 6 q0ij 6 U max
 U 0ij
j
j

K1
Ui
;

if

K1
Ji

>

K
Ji

K
Pbesti ;

if

K1
Ji

Step-6 Further a search reduction strategy [1] has been

employed to accelerate the convergence. Space reduction strategy is introduced at a stage, when the enhancement in the objective function has not been taken place
during a pre-specied iteration period. In such situation
the search space is dynamically reduced according to
following relation:
K1

U j;max U j;max  U j;max  Gbestj

9
 D=

U j;min U j;min U j;min  Gbestj  D ;

K1

16

where, D is known as step size, which is pre-specied. In

fact magnitude of D will decide how search space is
reduced.
Step-7 The computational algorithm is stopped if maximum
numbers of iteration have been executed.
PSO is a heuristic search procedure and as such no analytical
convergence criterion exists, except that PSO is terminated when
a pre-specied maximum numbers of iterations are executed [1].
This maximum number of iterations may be specied based on
experience on the system. Further the convergence is seen by plotting a graph between objective function and number of iterations
(see Figs. 1 and 2).
5. Results and discussions
The developed algorithm has been implemented on 6-bus and
25-bus systems [14]. 6-bus system consists of two generator buses
and four load buses. This system has in all 6-reactive power control
variables. Two generators buses and shunt compensation at bus
Nos. 4 and 6. OLTCs provided at line numbers 4 and 7. Maximum
internal voltages and synchronous reactances were assumed as
Emax1 = 2.20 pu, Emax2 = 2.05 pu, Xd1 = 1.0 pu, and Xd2 = 1.15 pu.
The limits of PV-bus voltages have been assumed as 0.95 and
1.15 pu. Shunt compensation limits were assumed as between
0.00 pu and 0.055 pu. OLTC limits have been assumed between

4
3
2
1

20

40

60

No of iterations
Fig. 1. Plot of objective function with respect to number of iteration for 6-bus
system.

10
9

K
Ji

where Ji is the objective function (6) evaluated at the poK1

sition of individual i. Gbest at iteration is set as the best
K1
value in term of objective function among Pbesti .

Objective function (J)

K1
Pbesti

Objective function (J)

The initial P0besti of individual i is set as the initial position of individual i and the initial G0best is determined
as the position of an individual with maximum value
of objective function as obtained using relation (6).
Step-3 Velocity of each individual is updated using relation
(14).
Step-4 The position of each individual is modied using relation
(15). The resulting position of an individual does not
guarantee the satisfaction of inequality constraints.
Hence in this paper a modied constraint handling
method has been adopted [13]. In the proposed procedure the intuitive idea to maintain a feasible population
is for a particle to y-back to its previous position, when
it is outside the feasible region. This has been termed as
y-back mechanism. Since the population is initialized
in the feasible region ying back to a previous position
will guarantee the solution to be feasible. Flying back
to its previous position when a particle violates the constraint will allow a new search closer to boundary.
Step-5 This step consists of updating the Pbest(i) and Gbest. The
Pbest at (K + 1)th iteration is updated as follows:

8
7
6
5
4
3
2
1
0

20

40

60

Number of iterations
Fig. 2. Plot of objective function with respect to number of iteration for 25-bus
system.

0.90 and 1.10. Total base case real and reactive power load on
the system is 1.35 pu and 0.32 pu.Value of proximity indicator at
base case condition is 0.4251. Table 1 shows PV-bus voltage and
all other load bus voltages under base condition. Star marked buses
are violating the load bus voltage limit. The desired range of load
bus voltage is 0.951.05 pu. The static voltage stability limit is
2.199415 pu for base case setting of control variables. The static
voltage stability limit is total load in pu as obtained using continuation power ow up to nose point of PV curve with voltage settings of PV-buses as specied above [20]. A threshold value of s
as proximity indicator has been assumed as 0.55. Initially ve particles were selected satisfying all inequality constraints by procedure explained in Section 4 and are given in Table 2. The values
of c1 and c2 have been selected same as one. As in most power system problems Wmax = 1.0 and Wmin = 0.5 have been selected. A typical value of D selected as 0.12. Maximum numbers of iterations
were set equal to 50. Table 3 gives maximum value of objective
function J has been obtained after 19 iterations. Fig. 1 shows variation of objective function (J) with respect to number of iteration.
It is observed from this Fig. 1 that solution converged in 19 iterations. Maximum numbers of iterations were specied as 50. After
19th iteration no improvement is found in Gbest. Table 4 gives
optimized set of control variables. In this situation all load bus

372

L.D. Arya et al. / Electrical Power and Energy Systems 32 (2010) 368374

Table 1
Load ow solution for 6-bus test system under base case condition.
Sr. No.

Control variables

Control variables magnitudes (pu)

Load bus voltages

Load bus voltages magnitudes (pu)

1
2
3
4
5
6

V1
V2
BSH4
BSH6
TAP4
TAP7

1.0000
0.9500
0.0500
0.0550
1.0000
1.0000

V3
V4
V5
V6

0.8303*
0.8588*
0.7901*
0.8420*

Total load; Pd = 1.35 pu, Qd = 0.32pu.

Proximity indicator s = 0.4251.
*
is the buses are violating the bus voltage limit.

Table 2
Initial solutions for 6-bus test system for PSO.
Sr. No.

V1 (pu)

V2 (pu)

BSH4 (pu)

BSH6 (pu)

TAP4

TAP7

1
2
3
4
5

1.1480
1.1448
1.1445
1.1421
1.1478

1.0933
1.1249
1.1279
1.1405
1.1228

0.0232
0.0397
0.0450
0.0466
0.0497

0.0234
0.0491
0.0362
0.0007
0.0075

0.9481
1.0299
1.0495
0.9142
1.0682

0.9554
1.0676
1.0408
0.9842
1.0289

2.33852
2.41468
2.48728
2.58620
1.44843

Table 3
Various parameters in PSO for 6-bus test system.
Case

Wmax

Wmin

C1

C2

No. of initial particles

Max J

No. of iterations for convergence

1.0

0.5

1.0

1.0

05

6.0529

19

voltages are also given in the same table. Total reactive reserve
available 1.42294 pu. Continuation power ow was carried with
optimized set of reactive power control variable and static voltage
stability limit was obtained as 2.31484 pu. Best initial solution
(particle) selected as V1 = 1.1421 pu, V2 = 1.1405 pu, BSH4 =
0.0466 pu, BSH6 = 0.0007 pu, TAP4 = 0.9142, TAP7 = 0.9842. Reactive
reserves at bus Nos. 1 and 2 with 6 best initial solution were
0.2937 pu and 0.7839 pu. Where as with optimized solution these
reactive reserves are obtained as 0.50874 pu and 0.9142 pu.
Weighting factors are 3.1875 and 4.8472 obtained by sensitivity
analysis. Magnitude of proximity indicator with optimized solution
is s = 0.5523.
Similar results have been obtained for 25-bus test system. This
system consists of ve generator buses and these are reactive power
control variables. Remaining 20 buses are load buses. Maximum
internal voltages and synchronous reactances were assumed as
Emax1= 2.60 pu, Emax2 = 2.15 pu, Emax3 = 2.10 pu, Emax4 = 2.30 pu,
Xd1 = 1.00 pu,
Xd2 = 1.15 pu,
Xd3 = 1.05 pu,
Emax5 = 2.15 pu,
Xd4 = 1.20 pu, and Xd5 = 1.15 pu. Total base case real and reactive
power load on the system is 12.41 pu and 3.876 pu.Value of proximity indicator at base case condition is 0.30. Table 5 shows PV-bus
voltage and all other load bus voltages under base condition. Star

marked buses are violating the bus voltage limit. The desired range
of load bus voltage is 0.95 pu to 1.05 pu.The static voltage stability
limit is 17.229396 pu in base case setting of control variables. A
threshold value of s as proximity indicator has been assumed as
0.39. Initially ve particles were selected satisfying all inequality
constraints by procedure explained in Section 4 and are given in
Table 6. The values of c1 and c2 have been selected same as 1.00.
As in most power system problems Wmax = 1.0 and Wmin = 0.5 have
been selected. A typical value of D selected as 0.12. Maximum
numbers of iterations were set equal to 50. Table 7 gives maximum
value of objective function J has been obtained after 27 iterations.
Fig. 2 shows variation of objective function (J) with respect to number of iteration. Solution gets converged in 27 iterations. Maximum
numbers of iterations specied were 50. After twenty seven iterations no improvement in objective function was found. Table 8 gives
optimized set of control variables. In this situation all load bus voltages are also given in the same table. Total reactive reserve available
2.4526 pu.Continuation power ow was carried with optimized set
of reactive power control variable and static voltage stability limit
was obtained as 20.132291 pu. Best initial solution (particle)
selected as V1 = 1.1373 pu, V2 = 1.0740 pu, V3 = 1.0210 pu, V4 =
1.0488 pu, V5 = 1.1002 pu. Reactive reserves at bus Nos. 15 with 5

Table 4
Optimized set of control variables and all bus voltages for 6-bus system.
Sr. No.

Control variables

1
2
3
4
5
6

V1
V2
BSH4
BSH6
TAP4
TAP7

Optimized control variables magnitudes (pu)

PSO

DFP

1.0868
1.0661
0.0497
0.0526
0.9455
0.9872

1.1042
1.0711
0.0474
0.0542
0.9529
0.9870

Total load Pd = 1.35pu, Qd = 0.32 pu.

Proximity indicator using (PSO) s = 0.5523.
Proximity indicator using (DFP) s = 0.5590.

Load bus voltages

Load bus voltages magnitudes (pu)

PSO

DFP

V3
V4
V5
V6

0.9586
0.9699
0.9530
0.9500

0.9608
0.9714
0.9500
0.9509

373

L.D. Arya et al. / Electrical Power and Energy Systems 32 (2010) 368374

best initial solution were 0.2841 pu, 0.4094 pu, 0.3092 pu,
0.2527 pu, and 0.1723 pu. Whereas with optimized solution these
reactive reserves are obtained as 0.5826 pu, 0.8121 pu, 0.4609 pu,
0.2969 pu, and 0.3001 pu respectively. Weighting factors for ve
generations are 3.9328, 1.2648, 5.5761, 6.2756 and 5.8215 obtained
by sensitivity analysis. Magnitude of proximity indicator with optimized solution is s = 0.3920.
The reactive power reserve maximization problem as formulated in Section 2 has been solved using one of the most popular
non-linear optimization techniques known as DevidonFletcher
Powells (DFP) method and accounting inequality constraints by
an exterior penalty function method [15]. This gradient based
method has been extensively used for non-linear optimization
problem solution in power system studies. Mechanization of DFP
method is much more involved than PSO technique. Various sensitivities are required to evaluate the gradient vector. A penalty
parameter is selected at a low initial value say 0.5 and if violation
persists than this is increased in step size by 10% each time till all
inequality constraints are satised.
Results obtained with this method for 6-bus test system are given in Table 4. This table gives value of control variable as obtained
by DFP method. It is seen; control variables as obtained by both the
methods are in close agreement. Value of Max J obtained using DFP
method is 5.9619 as against 6.0529 obtained using PSO. Reactive
power reserve obtained as 0.4551 pu, and 0.9307 pu. Similar results have been obtained for 25-bus system. Control variables as
obtained using DFP method has shown in Table 8 along with those
obtained using PSO. Value of objective function obtained using DFP
method is 9.1749 as against those obtained using PSO as 9.4987
(Table 7). Results obtained by both the method are in close agreement. Reactive reserves obtained at all ve buses obtained for 25bus test system are 0.5973 pu, 0.8035 pu, 0.4816 pu, 0.3011 pu,
and 0.2677 pu.

Table 5
Load ow solution for 25-bus test system under base case condition.
Sr.
No.

Control
variables

Control variables
magnitudes (pu)

Load bus
voltages

Load bus voltages

magnitudes (pu)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

V1
V2
V3
V4
V5

1.0000
1.0000
1.0000
1.0000
1.0000

V6
V7
V8
V9
V10
V11
V12
V13
V14
V15
V16
V17
V18
V19
V20
V21
V22
V23
V24
V25

0.9467*
0.9421*
0.9374*
0.9192*
0.9398*
0.9312*
0.9333*
0.9472*
0.8845*
0.8809*
0.9065*
0.9309*
0.9305*
0.9764
0.9378*
0.8903*
0.8425*
0.8991*
0.8130*
0.8282*

Total load Pd = 12.41 pu, Qd = 3.876 pu.

Proximity indicator s = 0.30.
*
is the buses are violating the bus voltage limit.

Table 6
Initial solutions for 25-bus test system for PSO.
Sr. No.

V1 (pu)

V2 (pu)

V3 (pu)

V4 (pu)

V5 (pu)

1
2
3
4
5

1.1396
1.1406
1.1373
1.1427
1.1362

1.0903
1.1021
1.0740
1.0628
1.1028

1.0318
1.0195
1.0210
1.0363
1.0861

1.0527
1.0481
1.0488
1.0511
1.0137

1.1043
1.1105
1.1002
1.0993
1.0987

2.93852
3.11468
3.29528
3.04620
2.87043

Table 7
various parameters in PSO for 25-bus test system.
Case

Wmax

Wmin

C1

C2

No. of initial particles

Max J

No. of iterations for convergence

1.0

0.5

1.0

1.0

05

9.4987

27

Table 8
Optimized set of control variables and all bus voltages for 25-bus system.
Sr. No.

Control variables

Optimized control variables magnitudes (pu)

PSO

DFP

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

V1
V2
V3
V4
V5

1.1354
1.0719
1.0161
1.0488
1.0954

1.1394
1.0813
1.0268
1.0475
1.0942

Total load Pd = 12.41 pu, Qd = 3.876 pu.

Proximity indicator using (PSO) s = 0.3920.
Proximity indicator using (DFP) s = 0.3997.

Load bus voltages

V6
V7
V8
V9
V10
V11
V12
V13
V14
V15
V16
V17
V18
V19
V20
V21
V22
V23
V24
V25

Load bus voltages magnitudes (pu)

PSO

DFP

0.9897
1.0326
1.0306
1.0171
1.0402
1.0345
1.0309
0.9823
0.9820
0.9940
1.0297
1.0368
1.0226
1.0488
1.0017
0.9751
0.9539
1.0358
0.9500
0.9741

1.0026
1.0376
1.0350
1.0199
1.0412
1.0355
1.0341
0.9949
0.9871
0.9974
1.0326
1.0378
1.0231
1.0487
1.0004
0.9745
0.9531
1.0371
0.9500
0.9750

374

L.D. Arya et al. / Electrical Power and Energy Systems 32 (2010) 368374

Results obtained by both the methods are in close agreement

and this justies the use of PSO technique for its ease of implementation and computational efciency, which is an established
advantage for such evolutionary algorithm [16].

6. Conclusions
This paper discusses the management of reactive power reserves
in order to improve static voltage stability. This has been achieved
via a modied PSO algorithm. Advantage of PSO algorithm is that
its mechanization is simple without much mathematical complexity. Moreover global optimal solution is obtained and local optimal
solution is avoided via search procedure. Important about the methodology is that not only reactive reserve is optimized but inequality
constraint on proximity indicator guarantee required static voltage
stability margin. Network as well as source capabilities are important from voltage instability viewpoint. Further contribution to
reactive reserve has been considered from participation viewpoint
by weighting factors. This is important aspect, which has been considered since large reactive reserve available at a generator bus,
which is not utilized in a load increased scenario, is not of great signicance. Hence a generator participating to a larger extent has
been given lesser weight in the PSO algorithm. The algorithm has
been implemented on 6-bus and 25-bus sample test systems.

References
[1] Park JB, Lee KS, Shin JR, Lee KY. A particle swarm optimization for economic
dispatch with non-smooth cost functions. IEEE Trans Power Syst
2005;20(1):3442.
[2] Yoshida H, Kawata K, Fukuyama Y, Takayama S, Nakanishi Y. A particle swarm
optimization for reactive power and voltage control considering voltage
security assessment. IEEE Trans Power Syst 2000;15(November):12329.
[3] Abido MA. Optimal design of power system stabilizers using particle swarm
optimization. IEEE Trans Energy Convers 2002;17(3):40613.
[4] Kasabadis IN, El-Sharkawi MA, Marks RJ, Moulin LS, da Silva APA. Dynamic
security border identication using enhanced particle swarm optimization.
IEEE Trans Power Syst 2002;17(August):7239.
[5] Nedwick P, Mistr Jr AF, Croasdale EB. Reactive management a key to survival in
the 1990s. IEEE Trans Power Syst 1995;10(2):103643.

[6] Vaahedi E, Masour Y, Fuchs C, Granville S, Latore MDL, Hamadoni Zadheh H.

Dynamic security constrained optimal power ow/Var planning. IEEE Trans on
Power Syst 2001;16(1):3843.
[7] Menezes T, da Silva LC, da Costa VF. Dynamic Var sources scheduling for
improving voltage stability margin. IEEE Trans Power Syst 2003(May):46971.
[8] Arya LD, Sakravadia DK, Kothari DP. Corrective rescheduling for static voltage
stability control. Int J Elect Power Energy Syst 2005;27(January):312.
[9] Dong F, Chowdhury BH, Crow ML, Acar L. Improving voltage stability by
reactive power reserve management. IEEE Trans PS 2005;20(1):33844.
[10] Vlachogiannis JG, Hatziargyriou ND, Lee KY. Ant colony system-based
algorithm for constrained load ow problem. IEEE Trans PS 2005;20(3):
12419.
[11] Kundur P. Power System Stability and Control. Mc Graw-Hill Inc.; 1994.
[12] Kreyszig EW. Advanced Engineering Mathematics. John Wiley; 2001.
[13] HE S, Prempain E, Wu QH. An improved particle swarm optimization for
mechanical design optimization problems. J Eng Optim 2004;36(5):585605.
[14] Bijwe PR, Kothari DP, Arya LD. Alleviation of line over loads and voltage
violations by corrective rescheduling. Proc IEE Part C 1993;140(4):24955.
[15] arya LD, Choube SC, Kothari DP. Emission constrained secure economic
dispatch. Int J Elect Power Energy Sys 1997;19(5):27985.
[16] Deb K. Multi Objective Optimization Using Evolutionary Algorithms. John
Wiley and Sons; 2003.
[17] Lof PA, Andersson G, Hill DJ. Voltage dependent reactive power limits for
voltage stability studies. IEEE Trans Power Syst 1995;10(1):2208.
[18] Van Cutsem T. A method to compute reactive power margins with respect to
voltage collapse. IEEE Trans Power Syst 1991;6(1):14556.
[19] Johansson SG, Daalder JE, Popovic D, Hill DJ. Avoiding voltage collapse by fast
active power rescheduling. Int J Elect Power Energy Syst 1997;19(8):5019.
[20] Ajjarapu V, Christy C. The continuation power ow: a tool for steady state
voltage stability analysis. IEEE Trans PS 1992;7(1):41622.
[21] Tinney WF, Walker JW. Direct solution of sparse network equations by
optimally ordered factorization. Proc IEE 1967;55:18015.
[22] Wallach Y. Calculations and Program for Power System network. Engle Wood
Cliffs (NJ): Prentice-Hall, Inc.; 1986.
[23] Yang N, Yu CW, Wen F, Chung CY. An investigation of reactive power planning
based on chance constrained programming. Int J Elect Power Energy Sys
2007;29(9):6506.
[24] Wu H, Yu CW, Xu N, Lin XJ. An OPF based approach for assessing the minimal
reactive power support for generators in deregulated power systems. Int J
Elect Power Energy Sys 2008;30(1):2330.
[25] He R, Taylor GA, Song YH. Multi-objective optimal reactive power ow
including voltage security and demand prole classication. Int J Elect Power
Energy Syst 2008;30(5):32736.
[26] Varadarajan M, Swarup KS. Differential evolutionary algorithm for optimal
reactive power dispatch. Int J Elect Power Energy Syst 2008;30(8):43541.
[27] Zhang W, Liu Y. Multi-objective reactive power and voltage control based on
fuzzy optimization strategy and fuzzy adaptive particle swarm. Int J Elect
Power Energy Syst 2008;30(9):52532.
[28] Zhang T, Elkasrawy A, Venkatesh B. A new computational method for reactive
power marked clearing. Int J Elect Power Energy Syst 2009;31(6):28593.