Professional Documents
Culture Documents
www.jatit.org
ABSTRACT
Reactive power dispatch (RPD) is one of the important tasks in the operation and control of power
system. This paper presents a Differential Evolution (DE) - based approach for solving optimal reactive
power dispatch including voltage stability limit in power systems. The monitoring methodology for
voltage stability is based on the L-index of load buses. The objective is to minimize the real power loss
subjected to limits on generator real and reactive power outputs, bus voltages, transformer taps and shunt
power control devices such as SVCs. The proposed algorithm has been applied to IEEE 30-bus system to
find the optimal reactive power control variables while keeping the system under safe voltage stability
limit and is found to be effective for this task. The optimal reactive power allocation results obtained
using DE are compared with other methods. It is shown that the objective function value is less than those
of other methods.
Keywords: Optimal Reactive Power Dispatch; Differential Evolution; Voltage Stability L-index; Power
System Loss
700
Journal of Theoretical and Applied Information Technology
www.jatit.org
Optimization (PSO), Ant Colony Optimization V L , I L are the voltage and current vectors at the
(ACO) and scatter search present great
convergence characteristics and capability of load buses
determining global optima. Evolutionary VG , I G are the voltage and current vectors at
algorithms have been successfully applied to many the generator buses
optimization problems within the power systems
area and to the economic dispatch problem in
Z LL , FLG , K GL ,YGG are the sub-matrices of
particular [6-23]. the hybrid matrix H.
Voltage Stability is becoming an increasing The H matrix can be evaluated from the Y bus
source of concern in secure operation of present- matrix by a partial inversion, where the voltages at
day power systems. The problem of voltage the load buses are exchanged against their
instability is mainly considered as the inability of currents. This representation can then be used to
the network to meet the load demand imposed in define a voltage stability indicator at the load bus,
terms of inadequate reactive power support or namely Lj which is given by,
active power transmission capability or both1.
Voltage collapse is a local load bus problem and V 0j
depends mostly on load conditions in the system. Lj = 1+ (2)
There exist two major techniques viz, static V j
approach and dynamic approach for this analysis. where,
Although not very accurate, yet the static
technique has gained wide acceptance for its V 0 j = −∑ F ji V i (3)
inherent virtues, eg, simplistic approach, faster i∈G
execution and less memory consumption. The
static voltage stability is primarily associated with The term V 0 j is representative of an equivalent
the reactive power support. The real power (MW) generator comprising the contribution from all
loadability of a bus in a system depends on generators.
reactive power support that the bus can receive The index Lj can also be derived and expressed
from the system. Several analytical tools have in terms of the power terms as the following.
been presented in the literature for the analysis of ∗
the static voltage stability of a system. This paper S j+
is mainly concerned with analysis and Lj = (4)
2
enhancement of steady state voltage stability based Y jj + Vj
on L-index [24]. An algorithm is proposed using
new operational load flow (OLF) and optimization where,
of reactive power control variables using LP S j + = S j + S jcorr (5)
technique. Simulated case studies conducted on
two Indian power networks of 82 and 217 buses * indicates the complex conjugate of the vector
∗
are presented for illustration purposes.
Z ji S i
2. VOLTAGE STABILITY L-INDEX
S jcorr = ( ∑
i∈Loads
∗
)V j (6)
i≠ j Z jj V i
Consider an n-bus system having 1, 2…g, 1
generator buses (g), and g+1,g+2…n the load Y jj + = (7)
buses(r=n-g-s) and t number of OLTC Z jj
transformers. The transmission system can be
represented using a hybrid representation, by the The complex power term component S jcorr
following set of equations represents the contributions of the other loads in
the system to the index evaluated at the node j.
⎡V L ⎤ ⎡ I L ⎤ ⎡ Z LL FLG ⎤ ⎡ I L ⎤ It can be seen that when a load bus approaches a
⎢ I ⎥ = H ⎢V ⎥ = ⎢ K YGG ⎥⎦ ⎢⎣VG ⎥⎦
(1) steady state voltage collapse situation, the index L
⎣ G⎦ ⎣ G ⎦ ⎣ GL approaches the numerical value 1.0. Hence for an
where overall system voltage stability condition, the
index evaluated at any of the buses must be less
701
Journal of Theoretical and Applied Information Technology
www.jatit.org
than unity. Thus the index value L gives an These constraints represent load flow equation
indication of how far the system is from voltage such as
Ng
collapse. This feature of this indicator has been
exploited in our proposed algorithm to evolve a Pi − Vi ∑ V j (Gij cos θ ij + Bij sin θ ij ) =0,
j =1
voltage collapse margin incorporated RPD routine.
i ∈ NB −1 (12)
The L -indices for a given load condition are
Ng
computed for all load buses. The equation for the Qi − Vi ∑ V j (Gij sin θ ij − Bij cos θ ij ) =0,
L -index for j-th node can be written as j =1
i=g
Vi i ∈ NB −1 (13)
L j = 1.0 − ∑ F ji ∠θ ji + δ i − δ j (8) Inequality Constraints:
i =1 Vj These constraints represent the system operating
constraints. Generator bus voltages (Vgi), reactive
i=g
Vi
L j = 1 .0 − ∑
power generated by the capacitor (Qci),
( F jir + jF jim ) (9)
transformer tap setting (tk), are control variables
i =1 Vj
and they are self-restricted. Load bus voltages
* Indicates the complex conjugate of the vector (Vload) reactive power generation of generator (Qgi)
Vi = Vi ∠δ i , V j = V j ∠δ j and line flow limit (Sl) are state variables, whose
limits are satisfied by adding a penalty terms in the
F ji = F ji ∠θ ji objective function. These constraints are
formulated as
F jir = F ji cos(θ ji + δ i − δ j ) , (i) Voltage limits
Vi min ≤ Vi ≤ Vi max ;i ∈ NB (14)
F m
ji = F ji sin(θ ji + δ i − δ j ) (10)
(ii) Generator reactive power capability limit
It can be seen that when a load bus approaches a Q gimin ≤ Q gi ≤ Q gimax ; i ∈ N g (15)
steady state voltage collapse situation, the index
L approaches the numerical value 1.0. Hence for (iii) Capacitor reactive power generation limit
an overall system voltage stability condition, the Qcimin ≤ Qci ≤ Qcimax ; i ∈ N c (16)
index evaluated at any of the buses must be less (iv) Transformer tap setting limit
than unity. Thus the index value L gives an t kmin ≤ t k ≤ t kmax ; k ∈ N T (17)
indication of how far the system is from voltage
collapse. This feature of this indicator has been (v) Transmission line flow limit
exploited in our proposed algorithm to evolve a S l ≤ S lmax ; l ∈ N l (18)
voltage collapse margin incorporated in RPD (vi) Voltage stability constraint
routine.
L j ≤ Lmax ; j ∈ N PQ (19)
3. FORMULATION OF ORPD PROBLEM The equality constraints are satisfied by running
the power flow program. The active power
The objective of RPD is to identify the reactive generation (P) (except the gi generator at the slack
power control variables, which minimizes the real bus), generator terminal bus voltages (V) and
transformer tap-settings (t) are the optimization gi
power loss ( Ploss ) of the system. This is
k variables and they are self-restricted by the
mathematically stated as follows: optimization algorithm. The active power
generation at the slack bus (Pgs ), load bus
Minimize F= [ f 1 ] voltages (V) and reactive power generation ( Q)
f1 = Ploss = ∑g
k∈N l
k (Vi 2 + V j2 − 2ViV j cosθij ) and voltage stability load gi level (L) are state
variables which are restricted through penalty
k = (i , j ) function approach.
(11)
4. OVERVIEW OF DIFFERENTIAL
The reactive power optimization problem is EVOLUTION
subjected to the following constraints.
Equality Constraints: One extremely powerful algorithm from
702
Journal of Theoretical and Applied Information Technology
www.jatit.org
evolutionary computation due to it’s excellent • Generate trial vectors applying the
convergence characteristics and few control selected crossover scheme
parameters is differential evolution. Differential • Select next generation members
evolution solves real valued problems based on the according to competition performance.
principles of natural evolution [11-15] using a
B. DE Optimization Process
population P of Np floating point-encoded
individuals that evolve over G generations to reach
1) Initialization
The first step in the DE optimization process is to
an optimal solution. In differential Evolution, the
create an initial population of candidate solutions
population size remains constant throughout the
by assigning random values to each decision
optimization process. Each individual or candidate parameter of each individual of the population.
solution is a vector that contains as many Such values must lie inside the feasible bounds of
parameters as the problem decision variables D. the decision variable and can be generated by Eq.
The basic strategy employs the difference of two (21). In case a preliminary solution is available,
randomly selected parameter vectors as the source adding normally distributed random deviations to
of random variations for a third parameter vector. the nominal solution often generates the initial
In the following, we present a more rigorous population.
description of this new optimization method.
P = [Y1( G ) ......................YNp
(G )
] (19)
Yi ,( 0j ) = Y jmin + η j (Y jmax − Y jmin ) (21)
i = 1,2,…, Np , j = 1,2,…, D
Yi ( G ) = [ X 1(iG ) , X 2( Gi ) ,........., X Di
(G )
] min max
Where Y j and Y j are respectively, the
i =1,2,… Np (20)
Extracting distance and direction information lower and upper bound of the j th decision
from the population to generate random deviations parameter and ηj is a uniformly distributed
result in an adaptive scheme with excellent
random number within [0,1] generated anew for
convergence properties. Differential Evolution
creates new offsprings by generating a noisy each value of j .
replica of each individual of the population. The 2) Mutation
individual that performs better from the parent After the population is initialized, this evolves
vector (target) and replica (trail vector) advances through the operators of mutation, cross over and
to the next generation. selection. For crossover and mutation different
This optimization process is carried out with types of strategies are in use. Basic scheme is
three basic operations: explained here elaborately. The mutation operator
• Mutation is incharge of introducing new parameters into the
• Cross over population. To achieve this, the mutation operator
• Selection creates mutant vectors by perturbing a randomly
First, the mutation operation creates mutant selected vector ( Ya ) with the difference of two
vectors by perturbing each target vector with the other randomly selected vectors ( Yb and Yc ). All
weighted difference of the two other individuals
selected randomly. Then, the cross over operation of these vectors must be different from each other,
generates trail vectors by mixing the parameters of requiring the population to be of at least four
the mutant vectors with the target vectors, individuals to satisfy this condition. To control the
according to a selected probability distribution. perturbation and improve convergence, the
Finally, the selection operator forms the next difference vector is scaled by a user defined
generation population by selecting between the constant in the range [0, 1.2]. This constant is
trial vector and the corresponding target vectors commonly known as the scaling constant ( S ).
those that fit better the objective function. Yi ' (G ) = Ya( G ) + S (Yb(G ) − Yc( G ) )
A. DE Algorithm i =1,2,…… Np (22)
• Initialize population Where Ya , Yb , Yc , are randomly chosen vectors
• While stopping criteria are not satisfied,
• Create mutant vector with the difference ∈ {1,2,.........Np } and a ≠ b ≠ c ≠ i
vector and scaling constant Ya , Yb , Yc are generated anew for each parent
703
Journal of Theoretical and Applied Information Technology
www.jatit.org
vector, S is the scaling constant. For certain be in the range of [0, 1]. A crossover constant of
problems, it is considered a = i . one means the trial vector will be composed
3) Crossover entirely of mutant vector parameters. A crossover
The crossover operator creates the trial vectors, constant near zero results in more probability of
which are used in the selection process. A trail having parameters from the target vector in the
vector is a combination of a mutant vector and a trial vector. A randomly chosen parameter from
parent (target) vector based on different the mutant vector is always selected to ensure that
distributions like uniform distribution, binomial the trail vector gets at least one parameter from the
distribution, exponential distribution is generated mutant vector even if the crossover constant is set
in the range [0, 1] and compared against a user to zero.
defined constant referred to as the crossover ⎧⎪ X i',(jG ) if η 'j ≤ C R or j = q
constant. If the value of the random number is less X = ⎨ (G )
'' ( G )
i, j
or equal than the value of the crossover constant, ⎪⎩ X i , j otherwise
the parameter will come from the mutant vector,
otherwise the parameter comes from the parent
Where i = 1, 2 … Np (23)
vector. Figure 3 shows how the crossover j = 1, 2 … D
operation is performed. q is a randomly chosen index
∈ {1,2,...........D} that guarantees that the trial
vector gets at least one parameter from the mutant
vector; η 'j is a uniformly distributed random
number within [0, 1) generated anew for each
Yb value of j . X i(,Gj ) is the parent (target) vector,
Yc X i',(jG ) the mutant vector and X i", (jG ) the trial
Ya vector.
Yi ' Another type of crossover scheme is mentioned
in [11].
⎧⎪Xi',(jG) for j = n D, n +1 D,...., n + L −1 D
Xi'',(jG) = ⎨ (G)
Figure 1: Mutation operator
⎪⎩Xi, j otherwise
(24)
Mutant Trail Parent Where the acute brackets denote the
D
modulo function with modulus D. The starting
index n is a randomly chosen integer from the
interval [0, D-1]. The integer L is drawn from
interval [0, D-1] with the probability Pr (L=v) =
(CR) v. CR ∈ [0,1] is the crossover probability
and constitutes a control variable for the DE
scheme. The random decisions for both n and L
are made anew for each trial vector.
4) Selection
The selection operator chooses the vectors that
are going to compose the population in the next
generation. This operator compares the fitness of
Figure 2: Crossover operator the trial vector and fitness of the corresponding
target vector, and selects the one that performs
The crossover operation maintains diversity in better.
the population, preventing local minima
convergence. The crossover constant ( CR ) must
704
Journal of Theoretical and Applied Information Technology
www.jatit.org
( G +1)
⎧⎪Yi" ( G ) if f (Yi " ( G ) ) ≤ f (Yi ( G ) )
Yi = ⎨ (G ) The minimization objective function given by
⎪⎩Yi otherwise equation (26) is transformed to a fitness function
(f) to be maximized as, where k is a large constant.
i =1, 2… Np (25)
This is used to amplify, the value of 1/F which is
The selection process is repeated for each pair usually small, so that the fitness value of the
of target/ trail vector until the population for the chromosome will be in a wider range.
next generation is complete.
5. DIFFERENTIAL EVOLUTION 6. SIMULATION RESULTS
SOLUTION TECHNIQUE
The details of the simulation study carried out
In the ORPD problem, the elements of the on IEEE 30-bus system using the proposed DE-
solution consist of all the control variables, based method are presented here. It is chosen as it
namely, generator bus voltages (V), the gi is a benchmark system, has more control variables
transformer tap-setting (tk ), and the reactive and provides results for comparison of the
power generation ( Qci ). These variables are proposed method. The approach can be
represented continuous variables in the DE generalized and easily extended to large-scale
population. systems. IEEE 30-bus system consists of 6
generator buses, 24 load buses and 41 transmission
Fitness Function: In the ORPD problem under lines of which 4 branches (6-9), (6-10), (4-12) and
consideration the objective is to minimize the total
(28-27) are with the tap-setting transformer.
power loss satisfying the constraints given by
Generator parameters are given in the Appendix.
equations (12) to (19). For each individual, the
equality constraints given by equations (12) and The transmission line parameters of this system
(13) are satisfied by running Newton-Raphson and the base loads are given in [1].
algorithm and the constraints on the state variables For the ORPD problem, the candidate buses for
are taken into consideration by adding a quadratic reactive power compensation are 10, 12, 15, 17,
penalty function to the objective function. 20, 21, 23, 24 and 29. The DE-based ORPD
With the inclusion of penalty function, the new algorithm was implemented using MATLAB code
objective function then becomes, and was executed on a PC. Two different studies
were performed with this system to show the
significance of the proposed method and the use of
Min F =
the algorithm in a bigger system. In case 1 RPD
N PQ Ng
lim
⎧⎪Vi min ; if Vi < Vi min
Vi = ⎨ max
⎪⎩Vi ; if Vi > Vi max
⎧⎪Qgimin ; if Qgi < Qgimin
Qgilim = ⎨ max (27)
⎪⎩Qgi ; if Qgi > Qgi
max
Figure 3: IEEE 30-bus system
705
Journal of Theoretical and Applied Information Technology
www.jatit.org
The DE parameters used for the optimal power and the objective function (Ploss) are plotted
flow solution are given in Table III. They are against the number of generations in Figure 2.
treated as continuous controls. The results of these From the figure it can be seen that the proposed
simulations are summarized next. algorithm converges rapidly towards the optimal
solution. This shows the effectiveness of the
proposed method for the ORPD problem.
TABLE I
SYSTEM DESCRIPTION OF CASE STUDY TABLE IV
Sl.No. Variables 30-bus CONTROL VARIABLES FOR THE 30-BUS SYSTEM
system I. Generator II. Shunt III. Transformer
1 Buses 30 voltages Compensation taps
Gen Value SVC Value Tran. Value
2 Branches 41 Tap
bus
3 Generators 6 1 1.0700 Qc10 0.0426 T 0.9000
Qc12 0.0260 6 −9
4 Generator buses 6 2 1.0629 0.9000
5 1.0446 Qc15 0.0275 T6−10
5 Shunts reactors 2 8 1.0430 Qc17 0.0282 1.0093
1.0974 Qc20 0.0458 T4−12
6 Tap-Changing 4 11 1.0119
Qc21 0.0380
transformers 13 1.0613
Qc23 0.0531
T28− 27
Qc24 0.0258
TABLE II Qc29 0.0309
LIMITS OF VARIABLES FOR IEEE 30-BUS SYSTEM
No. Description Units Lower Upper
Limits Limits
1 Voltage PQ- Pu 0.95 1.05
bus
2 Voltage PV- Pu 0.90 1.10
bus
3 Trans. taps Pu 0.90 1.10
TABLE III
DE PARAMETERS FOR BEST RESULTS OF OPTIMAL POWER FLOW
FOR IEEE 30-BUS SYSTEM
Parameters of Differential evolution
Sl.No. Parameters Values
1 Population 20
2 Generations 100
Figure 4: Fitness function value Vs Generations for case 1
706
Journal of Theoretical and Applied Information Technology
www.jatit.org
TABLE V
CONTROL VARIABLES FOR THE 30-BUS SYSTEM
I. Generator II. Shunt III. Transformer
voltages Compensation taps
bus Value SVC Value Tran. Value
No Tap
. Figure 7: Objective function value Vs Generations for case 2
1 1.0700 Qc10 0.0140 T6−9 1.0284
2 1.0625 Qc12 0.0554 7. CONCLUSIONS
0.9000
5 1.0387
Qc15 0.0421 T6−10
Qc17 0.0260 1.0137
8 1.0403 Qc20 0.0484 T4−12 This paper presents a DE solution to the optimal
0.9850 reactive power allocation problem and is applied
11 1.0863 Qc21 0.0159
Qc23 0.0194
T28− 27 to an IEEE 30-bus power system. The main
13 1.0646
Qc24 0.0497 advantage of DE over other modern heuristics is
Qc29 0.0288 modeling flexibility, sure and fast convergence,
less computational time than other heuristic
methods. And it can be easily coded to work on
parallel computers. The main disadvantage of DE
TABLE VI is that it is heuristic algorithms, and it does not
PERFORMANCE PARAMETERS provide the guarantee of optimal solution for the
Parameter Values RPD problem. The DE approach is useful for
Case 1 Case 2 obtaining high-quality solution in a very less time
Pg1 (pu)(slack bus) 0.9985 1.0236 compared to other methods.
Differential evolution algorithm is a stochastic
Lmax 0.1310 0.1800 optimization technique was employed as the
Ploss (pu) 0.0485 0.0507 optimization approach in determining the optimum
values for the reactive power to be dispatched to
707
Journal of Theoretical and Applied Information Technology
www.jatit.org
708
Journal of Theoretical and Applied Information Technology
www.jatit.org
[18] Raul E. Perez-Guerrero, Jose R. Cedeno- [21] IEEE Committee Report, “Present Practices in
Maldonado, “Differential Evolution Based the Economic Operation of Power Systems.”
Economic Environmental Power Dispatch” IEEE Transactions on Power Apparatus
Pp.191-197. Systems, Vol. PAS-90, July/August 1971, pp.
[19] R.Balamurugan and S.Subramanian ,“Self 1768-1775.
Adaptive Differential Evolution Based Power [22] A.Wood. B. Woolenberg, Power Generation,
Economic Dispatch Of Generators With Valve Operation and Control, New York: Wiley,
Point Effects And Multiple Fuel Options”, 1996.
International Journal Of Computer Science [23] D.C. Walters, G.B.Sheble, “Genetic
And Engineering ,Vol. 1, No. 1, 2007, ISSN Algorithm Solution of Economic Dispatch
1307-3699, pp. 10-17. with valve point loadings,” IEEE trans. Power
[20] Raul E. Perez-Guerrero and Jose R. Cedeno- systems, Vol. 8, No. 3, pp. 1325-1332, August
Maldonado, “Economic Power Dispatch With 1993.
Non-Smooth Cost Functions Using [24] P. Kessel, H. Glavitsch, Estimating the
Differential Evolution”, pp. 183-190. Voltage and Stability of a Power System,
IEEE Trans. Power Deliv. Vol. PWRD-1 (3)
(July 1986) 346–354.
709