2004 International Conference on

-
Power System Technology POWERCON 2004
Slngapore, 21-24 November 2004

Reactive Power Planning using Hybrid Method

P . h a v u k k a r a s u , R.Prasanna, T.Aruldoss Albert Victoire

Abstract: The papw proposes an application of Hybrid techniques. The EAs have attractive features such as the
Evolutiorwy Programming io Reactive Power Planning. Reactive simplicity of the algorithm, the ability to handle all sorts of
power planning is a non-smooth and nondifferentiable optimization functional representations of problems, including very
problem for a miti-objectivefinction. The objectivefinctions dmls complex function relationships. Despite their superior search
with the minimization of operating cost by reducing r e d power loss. ability, EAs fail to meet the high expectation that theory
improving the voltage profire and minimizing the allocafion cost of
additional reactive power sources. The proposed method is
predicts for the quality and efficiency of the solution. Genetic
developed in such a way that a standard e w l u t i o n a ~programming algorithms (GAS)have been given [3,6] for the global optimal
(EP)meihod is acting m a base level search, which makes a quick solutions of reactive power optimization problems. Simulated
decision to direct the search towards rhe optimal region, and local annealing (SA) would be more likely to either prematurely
optimization by Sequential Quadratic Programming is next employed converges or keep searching without direction GAS
to do Jne hming. The Reoctive Power Planning using Hybrid consistently perForms better than SA. Evolutionary
EVOlUtjOM?y programming is demonstrated with the IEEE 30-bus probymmhg robust and randomized algorithms, which can
system. The comprehensive simulation results show that Hybrid adaptively search the globally optimal point. One
Evolutionary Programming is a suitable method to solve the Reactive disadvantage of EP in problems is its slow convergence to a
power-planning problem.
-
Key words Evolutionary P r o ~ ~ Hybrid g , Algorithms
good near optitnum.
In general, Hybrid methods involve interwoven levels of
Optimization, Reactive pawer planning.
optimization. Namely evolution by any one of the
I. INTRODUCTION evolutionary algorithms and individual learning by
conventional optimization methods, which cooperate in a
I n general, Reactive power planning (RPP)can be defined as
to determine the amount and location of shunt reactive
power compensation devices needed for minimum cost while
global process of optimization.
The paper proposes an application of a two-phase
keeping an adequate voltage profile. RPP is one of the most evolutionary programming method based on the hybrid
complex problems o f power systems, as it requires the method has been developed. This method was also not able to
simultaneous “ i z a t i o n of two objective functions. The fmd the optimal solution for difficult constrained non-linear
first objective fimction deals with the minimization of real optimization probIems consistently, because once again
power losses in reducing the operation cost and improving the modified evolutionary algorithm was employed in the phase-2
voltage profile. The second objective h c t i o n minimizes the of the algorithm. To get a compromise between the
allocation cost of the additional reactive power sources. RPP computational expenses and the accuracy of the solution, a
is a non-linear optimization problem for a large-scale system new Hybrid EP method is proposed in this paper to improve
with many uncertainties. During the last decades there has the speed and solution quality so as to take challenging non-
been a growing concern about FWP problems. Conventional linear constrained optimizations problems. The propose hybrid
calculus based optimizations algorithms have been used in techniques codd outer fkom the conventional EP solution by
RPP for many years [l]. Most conventional optimization using EP as a base level search towards the optimal region
methods are based on successive liberalizations, which use the with limited computing h e . Subsequently, a local search
first and secund derivatives of the objective functions and its method an optimization by Sequential Quadratic
constmints equations as the search directions. Because the Programming is employed to do the fine-tuning to obtain an
formulae of the RPP problem are hyper quadric functions, exact global optimum with light computational expenses. The
such linear and quadratic treatments induce many local procedure handles either inequality constraints or equality
minima. Furthermore, the conventional methods cannot deal constraints and the feasible region do not have to be convex.
with the nondifferentiable factor in the reactive power Gradient and other auxiliary irhoimation’sare not required.
sources instaIIation function in RPP. The conventional In order to prove the effectiveness of the proposed hybrid
optimization methods can only lead to local minimum and algorithm, this method was applied to Reactive power
sometimes result in divergences in solving in RPP problems. planning. The IEEE 30-bus system is used in this paper as the
Recently, new methods based on artificial intelligences have simulation system. The results obtained were compared with
been used in RPP to solve local minimum problems and the results of Evolutionary programming methods. The
uncertainties [3,5]. outcome of the study clearly demonstrates the effectiveness
E h offer new tools for the optimization of complex and robustness that a hybrid evolutionary programming (HEP)
problem, Evolutionary algorithms have a number of can achieve over evolutionary algorithms.
advantages over other conventional optimization and search
Nomenclat w e
N, = set of numbers of load level duration
P.Thirunawkkarasu, R.Prasanna, T.Aruldoss Albert Victoire are with the NE = set of branch numbers
Department of Electrical Engineering, Government College of Techuology, N, = set of numbers of possible YAR Source installment buses
Coimbatore, India.
Email Id: thima 789@- [email protected] N, =set o f numbers of buses adjacent to bus i, including bus i
Nm= set of numbers of PQ-buses, which are load buses with

0-7803-8610-8/04/$20.00 0 2004 IEEE 1694

 constant P and Q injections Where reactive power flow equations are used as equality
Ng = set of generator bus numbers constrains.YAR source installment restrictions, reactive power
NT = set of numbers of tap-setting transformer branches generation restrictions, Transformer tap-setting restrictions
NB = set of numbers of total buses and bus voltage restrictions are used as inequality constraints.
h =per unit energy cost (%/Rmh) Qdh can be less than zero and if Qci is selected as a negative
di = duration of load level (h) value, say in the light load period, variabfe reactance should
gk = conductanceof branch k b.u) be installed at bus i. The transformer tap-setting T, generator
Vi = voltage magnitude at bus [email protected]) bus voltages V, and VXR source installments Qc are control
Qv = voltage angle difference between bus i and busj (rud variables so they are self-restricted. The load bus voltages
ei = fixed reactive power source installation cost at bus i K d and reactive power generations a are state variables,
C c I =per-unit VAR source purchase cost at bus i which are restricted by adding them as the quadratic penalty
Qi= reactive power injected into network at bus i terms to the objective function to form a penalty function,
Qi = reactive power source installation at bus i @.U) Equation (4)is therefore changed to the following generalized
PiQi=real and reactive powers, respectively, injected into objective function.
network at bus i @..U)
G,&= mutual conductance and susceptance,respectively
between bus i and bus j @.U)
N B -set~ ofnumber of total buses excluding a slack bus
Gipii =self conductance and susceptance, respectively, of
bus i &.U)
Qgi= reactive power generation at bus i s.t,o=Qi-vi (V . G i j S i n g' .I- -'BJ. . c o s g9. . ) i & N p e ( 5 )
Tk=tap-setting transformer branch k I
Npgl,,,=set of numbers of buses at which reactive power J& Ni
generations violate the limits. Where Aur and & are the penalty factors, which can be
increased in the optimization procedure; and are "i'"'
II. PROBLEM FORMULATION defined in the following equations:

The objective function in RPP problem comprises two

terms [6].The first term represents the total cost of energy IOSS
as follows:
wc = E dlPioss,l (1)
1.5 nl (7)
Where PI,J is the network real power,loss during the period
, l be expressed in the following
of load level 1. The P l ~ ~can
equation in the duration dl: It can be seen that the generalized objective function F, is a
non-linear and non-continuous function. Furthermore, it
Pjoss= c gk(v~+Y~-vivjcosL9y) (2) contains of a lot of uncertainties because of uncertain loads
k& nE and other factors.
The second term represents the cost of reactive power source
installments, which has two components, namely, fmed III. EVOLUTIONARY PROGRAMMING
installment cost and the purchase cost.
EP is different from conventional optimization methods. It
IC= c
iE?lC
( e j + CCieci) (3 1
does not need to differentiate cost function and constraints. It
uses probability transition rules to select generations. Each
The objective function, therefore, can be expressed as individual competes with other individuals in a combined
foIlows: population of the old generation and the mutated old
generation. The competition results are valued ushg a
&fc= IC+ wc probabilistic d e . The Winners of the same number as the
S.f,O=Qi-vi v j ( G..SinBij-B~COSBi;i)i&NpQ
!I individuals in the old generation, constitute the next
j E Ni
generation [X, 111. The EP procedure is briefly listed as
follows:
QFIQCi S Q y i & N r
Inifializution: The initial population of control variables is
selected randomly &om a set of uniformly distributed control
Qmin
gi Qgi < QFgr is N g
(4) variables ranging over their upper and lower limits. The
=Fin
5 Tk 2 Tp' iE N T
fitness scoref; is obtained according to the objective function
and the environment.
v p 5 vi < v imax i E N B

1695
Sutktics: The maximum fitness fmm minimum fitness fmmthe maximum generation number can be used for convergence
sum of fitness & and average fitness fm9 of this generation condition. Other criteria, such as the ratio of the average and
calculated as follows: the maximum fitness of the population is computed and
generations are repeated until

f,
f max
26 (11)

Where 6 should be very close to 1, which represents the

degree of satisfaction. If the convergence has reached a given
accuracy, an optimal solution has been found for an
f,= c fi
m

i=l
optimization problem.
IV.HYBRID EVOLUTIONARY PROGRAMMING
In general, local search techniques have the advantage of
solving the problem quickly, though their results are very
much dependent on the initial starting point; therefore they
Mrrialion: Each selected parent, for example Pi,is mutated can be easily trapped in a local optimum. On the other hand in
and added to its Population following the equation. evolutionary programming samples a large search space,

[
. . + N 0,pk i . m - x j -
P i + m , j = p 3.j )L)J
f,, = 1,2,....n (9)
climbs faany peaks in parallel, and is likely to lead the search
towards the most promising area. However, an EP faces
difficulties in a tine-tuning of local search: it spends most of
Where n is the number of decision variables in an individual, the time competing between different hills, rather than
Pii denotes the "J element of the J" individual: N ( ~ , )d improving the solution along a single hill that the optimal
represents a Gaussian random variable with mean p and point locates. Hence, if one makes use of the advantages of
variance 8.f- i!; the maximum fitness value of the old both the local and EP techniques, one can improve the search
generation, which is obtained in statistics. ,x,i is maximum algorithm both effectively and efficiently.
fitness value of the old generation which is obtained in fitness The proposed hybrid EP combines a standard EP and the
value of the old generation which is obtained in statistics. x-, phase3 of the conventional search technique. EP takes the
and xjm. are the maximum and minimum limits of t h e P place of the phase-1 of the search, providing the potential near
element; and p is the mutation scale, O<&l, that could be optimum solution, and the phase-2 of a search technique using
adiiptively decreaqed during generations. If any mutated optimization by Sequential Quadratic Programming Phase-2
value exceeds its limit, it will be given the limit value, The algorithm is applied to rapidly generate a precise solution
Mutation process (1) allows an individual With larger fitness under the assumption that the evolutionary search has
to produce more offspring for the next generation. generated a solution near the global optimum.

Comp&Xon: Several individuals @), which have the best Need for Phase-2 AIgoritlhm
fitness, are kept as the parents for the next generation. Other The Phase-2 algorithm is employed for h d m g the most
individuals in the combined population of size (2m-k) have to optimal solution. Normally, this algorithm is used €or rehing
compete with each other to get their chances for the next the solution obtained through the Evolutionary programming
generation.A weight value W, of the ibindividual is calculated method. In other words, it is used for fine tuning of optimal
by the following competition as follows: solution. The local search is performed using the phase-2
algorithm. The hybridization of the Evolutionary
programming method and the conventional technique
(Sequential Quadratic Programming) yields better results.
r=1[
0, Otherwike
Phuse-1 Algorithm
Where q is the number of wmpetition;f, is the fitness value of
the ihrandomly selected competitor in the combined Reactive power planning using Evolutionary Programming
population;f; is the fitness value of p b ui is randomly selected method is solved [111.
from a uniform distribution set, U (0,l). When all individuals
p c i =1,2 ...Zm,obtain their competition weights, they will be Phase-2 Algorithm
ranked in descending order of their corresponding value W, After the phase-1 is halted, satisfying the halting condition
.The first m individuals are transcribed along with their described in the previous section, optimization by Sequential
corresponding fitness values f; to be the basis of the next Quadratic Programming method is next employed in the
generation. The values of maximum, minimum and average phase-2. In the light of the solution accuracy, the success rate,
fitness and sum of fitness of this generation are then calculated and the computation time, the best vector obtained from the
in step 2. phase-lis used as an initial point for phase-2.
This Sequential Quadratic Programming procedure (phase-
Convergence test: If the convergences condition is not met, 2 algorithm) is implemented as follows [SI:
the mutation and the competition process will run again. The

1696
programming procedure for the solution of the approximate SB(MVA) 1 h($Kwh) I e,($) 1 C,i($/kVAR)
model here developed. 100 0.06 1000 1 30

BUS I 2 5 8 11 13
Let x=O.Repeat Koax
II 0.596 0.480 0.6 0.53 0.15 0.155
1. For each considered load level, a harmonic power Qamin -0.298 -0.24 -0.3 -0.265 -0.075 -0.078

Period I Case 1 I Case 2

The main reason for the success of phase-2 algorithm is that

lies in its focal search ability. Since the values for the variables OUTPUTGRAPH
for always chosen around the best point determined in the
previous iteration, there is a more likelihood of convergence to
the optimum solution.
V.NUMERICAL RESULTS
In this section, E E E 30-bus system has been used to show
the effectiveness of the algorithm. The network consists of 6
germator-buses, 21 load buses and 43 branches, of which 4
branches, (6, 9), (6, lo), (4, 12) and (28,27) are under load
tap-setting transformer branches. The parameters and variable
limits are listed in table 1.The possible VAR source VI. CONCLUSION
installment buses are Buses 6, 17, 18 and 27. All power and The use of Hybrid Evolutionary Programming for the
voltage quantities are per-unit values and the base power i s Reactive power planning of power systems has been reported.
used to compute the energy cost. Reactive power planning is an optimization problem of a non-
Two cases have been studied. Case 1 is of light loads linear, non-smooth and non-continuous function. This type of
whose loads are the same as [lo ].case 2 is of heavy load less well behaved function is met with in most engineering
whose loads are. twice as those of casel.one year energy loss problems, so the devising, testing and r e h i n g of new
cost is used to access the possibility of installing the VAU techniques for finding optimal solutions has become more
sources, so the duration of load level is given as in table 4. important with the advent of even more powerful computers.
The Hybrid Evolutionary Programming approach has been
Initid Conditiutr: All transformer taps are set to 1.0. The evaluated on the E E E 30-Bus power network. The
initial generator voltages are set to 1.O.The generator bus simulations show that Hybrid Evolutionary Programming
voltages are set in different values to share the Ioads evenly always leads to satisfactory results for multi- objective
among the generators as table 2. The loads are given as: Reactive power planning, especially in non-continuous and
Plod= 2.834 Qlo&=1.262. non-smooth situations. The comparison shows that the Hybrid
evolutionary Programming method is more powerful for
The minimum cost obtained using the Hybrid Technique is global optimization problems of the non-smooth, non-
shown in Table 5 . continuous function. To prove the effectiveness of the
proposed hybrid algorithm, it was applied to reactive power

1697
planning. On comparison of the results it was observed that [e]. Kwang.Y.Lee and Frank F.Yang, ‘Qpoptimal Reactive power planning
using EvolutionaryAlgorithm: A comparative study for Evolutionary
the hybrid evolutionary programming was more effective than P r o g ” h g , Evolutionary strategy, Genetic Algorithm, and Linear
the evolutionaryprogramme method. P m g r a m ” : IEEE Trmuaiom on Power System, Vol. 13, No. 1, Feb
1998.
Vn. REFERENCES [9]. Ignacio Ignacicez Abd and Jose angle G o d e z pUintero VAR “

Compensationby Sequential Quadratic Programming’’, IEEE Tmactiom

on Power Systems, Vol. 18,No.1, Feb 2003
[l]. k Kishore and E.F.HiU, ‘Static Optimization of reactive power source9 [lOl.KY.Lee, Y.M.Park and J.L. Orti% “A United approach to optimal
by use of sensitivityp-eteW, IEEE Tmnractions on Power Apporafm
real and reactive power Dispatch”, IEEE Tranrnctfons on Power upparam
ad$wtem, vol. PAS-90,No.,pp.1166-1173,1971.
and+sfems, Vol. PAS-104,No 5Nay 1985.
[Z]. S.Sa&deva and RBillintoq “Optimum network VAR planning by [I 11. V.Gapalakrishnan and P.”avukkamsu, ‘%Reactive Power Plamhg
nonlinear Programming”, IEEE ?hmuctiom on Power Appurarus und whg Evolutionary Programming Method“,Proceedings of Natiand
$!atem, volPAS-92 No.,pp1217-1225, 1973.
conference of Power Elecfronicsin Energv Conservation,Feb 2004.
[3]. SSwarup, M.Yoshimi and Y . h ,“Genetic algorithm approach to
reactive power planning in power systems”, Proceedings of the 5’Annrrui
conferenceofPowerrmdEnergvsOciely L E E J ~ npp-119-124,
, 1994. WI. BIOGRAPHIES
[4].RH.AbduaI-Rahman and S.M. Shahidehpour, “Application of rupY
sets io optimal d v e power planning with security conshains”, IEEE P,Tbirnnavokkamsu obtahd his Depee of Engiaeaing from M a d d
Trunsacmns on Puwer S)mew, vol PWRS-9,No-2, pp 589-597. May Kamaraj University and Master of E n g i n k g from Anna University in 2000
1994. and 2004 respectively. His field of interest is Power System optimization,
[5]. Y.T.Hsiao.C.C.Liu, H.D.Cbiang and Y.LChm, “A new approach for Power Quality and Power system Control.
optimal VAR sources Planning in large scale electric power system”,
IEEE Transactions on Power S y s t m , Vol.PwRs-S.No.3, pp.988-996, RPrasanna obtained his Degree of Engheering from Madurai Kamaraj
Aug.1993. University and Master of Engineering from Aana University in Z00l and
[q.L.L.Lai and J.T.Ma, “Application of evolutionary programming to 2004 respectively. His field of interest is Power System optimization, Power
reactive power planning - Comparison with Nonliar Programming System Reshucturing, Power Quality a d Power System control.
Approach”,EEE Trunracrionson Power Systm, Vol 12, No 1, Feb 1997.
[7]. LJ.Foge1, “The Feature of Evolufio~ary Programming”, proceading of
the 2 6 AsiIumar Cnnfmmces on Signals, S y s f m nnd Computers, Pacific
Grove, CA, 1990.

1698