(2)
Where the subscripts G, L, and T indicate generation, load,
and injection respectively,
i
V and
i
B are voltage and old/new
shunt of bus i. The initial value of is 1. The objective
function (1) minimize (- ) so is maximized.
The IPM algorithm is applied to search solution. For IPM,
lagrange function is following.
u w l z g u x g w
g l x g z x h x f L
T T T
T T
+
=
) ) ( (
) ) ( ( ) ( ) ( ) (
max
min
(3)
Where, g h f , , are objective function (1), equality
constraints (2) and inequality constraints. w z, , are dual
variables and u l, are slack variables.
max min
, g g are
upper/lower limits of inequality constraints.
In (3), dual variables mean increamental cost which
means increamental value of objective function when load
increase. The following equation means concept of .
L
Q
L
P
Q
f
P
f
= , (4)
If the capacitor is allocated at bus i, it means that the
reactive load at bus i is decreased. Because the objective
function is maximum loadability, (-
Q
) represents the amount
of increased loadability according to allocated capacitor. If the
maximum loadability is less than desired margin, the capacitor
is allocated at the bus presents the largest (-
Q
). Similarity, if
the maximum loadability is greater than desired margin and
too many capacitors are allocated, the capacitor is not
allocated at the bus presents the lowest (-
Q
). So, the
Q
can
be applied to direction vector for the capacitor allocation
studies. However, the IPM algorithm performs to search of the
local optimal solution, so
Q
has to be applied for the local
optimal solution.
To calculate the maximum loadability, CPF or OPF has to
be performed. To search the local optimal solution, the local
search method has to be performed. However, local search
method is needed many search time. If we employ the OPF to
calculate the maximum loadability, we can obtain the
information of
Q
and this information can be applied to
search the local optimal solution as the form of direction
vector.
B. Formulation of PSO problem
The PSO is modern heuristic algorithm. It can search the
global solution and is easy to deal with integer variables. So,
the PSO is useful to the optimal capacitor placement problem.
However, it is not converged to the unique solution, namely,
the solution is different each time it running. So, to obtain
reasonable solution, we need to search the local optimal
solution. In this section, introduce the traditional PSO, and
then suggest the improved PSO algorithm.
The procedure of PSO is very simple.
1. Generate the particle
Each particle is generated by the random variables
2. Calculate the objective function
Decision of global best and each local best
3. Calculate the velocity of the each particles
4. Update the each particle
5. Go to step2 until all particles meet at the one solution
The procedure of the suggested algorithm is same with
ordinary PSO algorithm. Only the calculation method of the
velocity is different. For the OCP, the each particle contains
the information of the capacitor amount of the each location.
This amount is generated by the random variables. The
objective function for OCP is following.
set location candidate S
w ortherwise
value rge la w then rgin ma desired if
w B f
S
i
new
i
=
=
=
=
1 ,
,
) ( min
1
(5)
Where,
new
i
B is the allocated capacitor amount at the bus i,
is the voltage stability margin (PV margin), and wis the
weighting factor. The OCP considering voltage stability
margin (VSM) has to satisfy desired margin. So, if the VSM
dont satisfy the desired margin or the OPF is not converge
because of the lack of the allocated capacitor, the weighting
factor ( w) has to have very high cost.
After calculation of OBJ, the each particle updates the each
local best and the global solution is updated also. And then,
the velocity is calculated to update the each particle. The
equation of velocity is following.
s B
r C
P G r C P P r C v w v
size
average
Q
C des
average
Q
Q
C best C best
t t
+ + =
3 3
2 2 1 1
1
) ( ) (
(6)
Where, w is inertia weight.
3 2 1
, , r r r are random numbers
between 0 and 1.
best best
G P , are the local best and global best.
C
P is the current location of the each particle.
3 2 1
, , C C C are
positive constants called as the cognitive and social parameters
(acceleration parameters) respectively. This acceleration
factors are pull the solution towards
best
P and
best
G positions.
average
Q Q
, are the dual variable from the OPF and the average
value of dual variables.
C des
, are the desired VSM and the
VSM of the each particle.
size
B is the shunt capacitor amount
per one bank. s is the total number of the candidate location.
The suggested algorithm uses the dual variables from the
OPF result. In (4), it can be rewritten in following.
i L
Q
B Q
f
=
(7)
= s B
B
s B
size
i
size
average
Q
(8)
In (8), s B
size
average
Q
means when one shunt
capacitor bank is allocated at each and all location. So, in (6),
s B
size
average
Q
C des
means how much shunt capacitor banks are
needed at each and all location. And in (6),
average
Q
Q
is pull
the solution towards more sensitive buses, namely it performs
the function of the local search. So,
s B
size
average
Q
C des
average
Q
Q
=
2 1 1
2
2
) (
) (
2
1
C best
iter
C
iter
best
P G
P G
w (11)
Where, iter is the iteration number which starts from
zero.
1
,
best
iter
best
G G are the global best of each iteration and first
iteration.
1
,
C
iter
C
P P are the location of particle of each
iteration and first iteration.
1
w is decreased at each iteration
but if global best is updated, iter is reset to zero.
2
w is
decreased as each particle move to global best. So, w is
changed from 1.5 to nearby zero.
From (6), the position of particle is updated.
1 1 + +
+ =
t t
C
t
C
v P P (12)
C. Solution procedure of OCP
The optimal capacitor placement has to consider the
contingency. The following flowchart shows the overall
procedure.
Fig. 1. Flowchart of OCP
The one particle is calculated contingency number times.
The OBJ of particle has to be updated by the greatest value of
all contingency, namely the most sever contingency.
III. CASE STUDY
This section provides an example applying the proposed
method, OCP, into NewEngland 39-buses test system to verify
the method. The particle number is 50 and the candidate
location is the all load buses, and 5 banks (10Mvar per bank)
can be allocated at each and all load bus. The desired voltage
stability margin ( ) is 1.8 and the contingency is not consider
to reduce the progress time. All
3 2 1
, , C C C are 1. FORTRAN
is used as a front end language and the simulations are carried
out on a Intel core2 CPU, 2.4GHz, 2GB RAM processor.
The test case1 is the ordinary velocity calculation method
and the case2 is the suggested method. The next table (1) and
figure (2) show the OBJ results about case1 and case2. For the
accuracy of result, total 20 times simulation per each case is
performed.
TABLE I
OBJ RESULT OF EACH CASE
1 2 3 4 5 6 7
Case1 20.19 20.19 21.19 28.19 20.19 21.19 24.19
Case2 20.19 21.19 21.19 21.19 20.19 21.19 21.19
8 9 10 11 12 13 14
Case1 20.19 22.19 24.19 24.19 24.19 21.19 20.19
Case2 21.19 21.19 20.19 20.19 20.19 20.19 20.19
15 16 17 18 19 20
Case1 20.19 21.19 21.19 22.19 21.19 20.19
Case2 20.19 21.19 20.19 22.19 20.19 20.19
Fig. 2. OBJ result of each case
In figure (2), we can compare with the two results. The
horizontal axis is the number of times of simulation and the
vertical axis is the OBJ value which has to show the low result.
The result of case1 is not uniform, namely, the possibility to
find the optimal solution is so low. The other side, case2 can
improve the possibility to find the optimal solution.
The next table (2) and figure (3) show the CPU times about
case1 and case2.
TABLE II
CPU TIME OF EACH CASE
1 2 3 4 5 6 7
Case1 596 280 626 438 774 761 545
Case2 131 268 191 237 64 191 231
8 9 10 11 12 13 14
Case1 809 490 519 475 640 641 505
Case2 176 185 164 201 163 151 184
15 16 17 18 19 20
Case1 728 499 548 520 531 744
Case2 207 164 220 94 171 213
Fig. 3. CPU time of each case
In figure (3), we can compare with the two results.
The horizontal axis is the number of times of simulation and
the vertical axis is the CPU time (second). The result of case2
is less than case1. The suggested algorithm reduces the CPU
time about average three times.
IV. CONCLUSIONS
This paper suggests an optimal capacitor placement
considering voltage stability margin based on improved PSO
algorithm which is useful to power systems var planning. The
power system planning must consider the voltage stability
margin. The suggested algorithm can find the solution well to
insure the VSM. To solve OCP problem, PSO algorithm is
applied because PSO can find global solution and is easy to
handle integer variable. But, PSO is limited to solve the
optimal solution because of several reasons, so suggested
algorithm complements a weak point of PSO.
This paper demonstrates the quality of suggested algorithm
through the test system. The improved PSO shows good
characteristics about accuracy and simulation running time.
However, as ever, it is too late to apply to large system. So,
hereafter, the studies for more accurate and fast running time
have to be advanced.
ACKNOWLEDGMENT
This work is the outcome of a Manpower Development
Program for Energy & Resources supported by the Ministry of
Knowledge and Economy (MKE) and supported in part by the
Korea Electric Power Research Institute.
REFERENCES
[1] Thomas,W.R;, Dunnett, R.M.; Dixon, A.M.; Schaff, G.; Cheng, D.T.Y.;
Thorp, J.D, Optimal reactive planning with security constraints, IEEE
Power Industry Computer Applications Conference, p79-84, 1995
[2] Xihui Yan, Victor H.Quintana, Improving An Interior Point Based OPF
by Dynamic Adjustments of Step Sizes and Tolerances, IEEE Trans. on
Power Systems, Vol.14, No.2, 709-717, 1999
[3] H. Wei, H. Sasaki, J. Kubokawa, and R. Yokoyama, An Interior Point
Nonlinear Programming for Optimal Power Flow Problems with A
Novel Data Structure, IEEE Trans. on Power Systems, Vol.13, No.3,
870-877, August 1998
[4] Victor Quintana, Geraldo Torres, An Interior Point Method for
Nonlinear Optimal Power Flow Using Voltage Reactangular
Coordinates, IEEE Trans. on Power Systems, Vol.13, No.4, 1211-1218,
1998
[5] Yu-Chi Wu, Atif S.Debs, Roy E.Marsten, "A Direct Nonlinear
Predictor-Corrector Primal-Dual Interior Point Algorithm for Optimal
Power Flows", IEEE Trans. on Power Systems, Vol.9, No.2, 876-883,
1994
[6] Sergio Granville, Abib Lima.Maria Candida, Application of
decomposition techniques to VAr planning: methodological &
computational aspects, IEEE Transactions on Power System, v.9, n.4,
p1780-1786, 1994
[7] Gan.Degiang, Qu.Zhihua, Cai.Hongzhi, Large-scale var optimization
and planning by tabu search, Electric Power Systems Research, v.39,,
n.3, p195-204, Dec.. 1996
[8] R.C.Eberhart and Y.Shi, Particle swarm optimization : developments,
applications and resources, Proc. Congr.Evol.Comput, p81-86, 2001
[9] M.Clerc and J.Kennedy, The particle swarm-Explosion, stability and
convergence in a multidimensional complex space, IEEE Tr Recent
approaches to global optimization problems through particle swarm
optimization, Natural compution, p235-306, 2002
[10] H. Song, B. Lee, S. Kwon, V. Ajjarapu, Reactive Reserve Based
Contingency Constrained Optimal Power Flow (RCCOPF) for
Enhancement of Voltage Stability Margins, IEEE Tran. Vol. 18, p1538-
1546, Nov. 2003
[11] Malihe M.Farsangi, Hossein Nezamabadi-pour and Kwang Y.Lee,
Implementation of GCPSO for Multi-objective VAr Planning with SVC
and Its Comparion with GA and PSO, Intelligent Systems Applications
to Power Systems, ISAP 2007, International Conference, p1-6, Nov.
2007
[12] Eghbal.M, EI-Araby, E.E., Ito.Y., Yorino.N, A PSO Approach for
VAR Planning Considering the Slow and Fast VAR Devices Prices,
IEEE International Conference, Vol.3, p1849-1854, Oct., 2006
[13] Jong-Young Park, Jin-Man Sohn, Hong-Keun Park, Optimal Capacitor
Allocation in a Distribution System Considering Operation Costs,
Power Systems, IEEE Transactions, Vol.24, Issue1, p462-468, Feb.
2009