Académique Documents
Professionnel Documents
Culture Documents
CHAPTER 2
MULTI-OBJECTIVE GENETIC ALGORITHM (MOGA)
FOR OPTIMAL POWER FLOW PROBLEM INCLUDING
VOLTAGE STABILITY
2.1
INTRODUCTION
Voltage stability enhancement is an important task in power system
optimization
problem
with
cost
minimization
and
When an
optimization problem involves more than one objective function, the task of
finding one or more optimum solutions is known as multi-objective
optimization (Deb 2001). Since classical search and optimization algorithms
use a point by point approach, the outcome of using a classical approach is a
single optimized solution. However, most real world problems involve
simultaneous optimization of several often mutually concurrent objectives.
For multi-objective optimization, the preference based approach requires
multiple runs as many times as the number of desired optimal solutions.
Multi- objective Genetic Algorithm (MOGA) is proposed in this chapter to
solve the multi-objective OPF problem.
27
2.2
where
IG
YGG YGL
VG
IL
YLG YLL
VL
(2.1)
where
VL
Z LL F LG
IL
IG
K GL YGG
VG
-1
(2.2)
and
matrix.
The L-index of the j-th node is given by the expression,
28
Ng
Lj
F ji
i 1
where
Vi
Vj
ji
(2.3)
PROBLEM FORMULATION
The OPF problem is to optimize the steady state performance of a
29
Objective functions
Here the goal is to determine the optimal values of generator active
power, generator bus bar voltages, transformer tap settings and reactive power
sources to enhance the system stability level while minimizing the fuel cost
and maximizing the voltage stability margin.
2.3.1.1
(a i Pgi2
Minimize FC
bi Pgi
c i ) $ / hr
i 1
(2.4)
30
voltage stability margin. This is achieved by minimizing the Lmax value. This
is stated as,
Minimize L max
(2.5)
Equality constraints
In a power system with NB buses, at each bus i, the sum of the total
injected real and reactive power must be equal to zero. The specified power is
equal to the difference between the power generation and the load. These
constraints are mathematically represented as:
NB
Pi
V j (G ij Cos
Vi
ij
Bij Sin
ij
0 ;i
Bij Cos
ij
) 0 ;i
NB 1
(2.6)
N PQ
(2.7)
j 1
NB
Qi
V j (G ij Sin
Vi
ij
j 1
The set of equations formed by equation (2.6) for all system buses
except the slack bus and equation (2.7) for all load buses constitute the load
flow equations. The load flow equation determine the steady state condition
of the power system network for specified generations and load patterns,
calculate voltages, phase angles and flows across the entire system. When
solving the power flow equations iteratively, successive solutions will have a
31
mismatch between the specified and the injected power. So equation (2.6) and
(2.7) will not be satisfied. Hence, a tolerance is specified for the power flow
solutions.
2.3.3
Inequality constraints
The inequality constraints reflect the limits on physical device in the
Voltage limits
Too high or too low voltage magnitudes could cause problems to the
end user power apparatus or instability in the power system. The voltage
magnitude constraint is expressed as:
Vi min
(ii)
Vi
Vi max
;i
NB
(2.8)
(iii)
Q gi
max
Q gi
;i
Ng
(2.9)
32
(iv)
Q ci
Q cimax
;i
Nc
(2.10)
by the angle of difference of the terminal voltages and the flow of reactive
power is determined mainly by the magnitude difference of terminal voltages.
The value is modified in the search procedure among the existing tap
positions and expressed as:
t kmin
(v)
tk
t kmax
;k
NT
(2.11)
are given due to limitations of the branch material. Excessive power would
damage the transmission elements. This is stated as an inequality constraint
as:
Sl
S lmax
;l
Nl
(2.12)
33
and constraints, the problem can be mathematically formulated as a multiobjective non- linear constrained problem as follows:
Minimize F T
[ F C , Lmax ]
(2.13)
MULTI-OBJECTIVE OPTIMIZATION
Many real world problems involve the simultaneous optimization
of several objective functions. Generally, these functions are noncommensurable and often conflicting objectives.
In general, a multi-objective optimization problem is formulated as
follows:
Min / Max f i ( x)
g j ( x)
Subject to:
where
0 , j 1,. . .M
hk ( x) 0 , k
fi
i 1. . . N obj
1, . . . K
(2.14)
(2.15)
34
35
Dominated
solutions
f2 (minimize)
Pareto front
Pareto Solutions
f1 (minimize)
36
37
No
i=i+1
Meeting the
stopping rule
Yes
Output
38
2.4.2
Algorithm of MOGA
MOGA is an extension of the classical GA (Fonseca and Fleming.
(2.16)
(2.17)
Sh (d ij )
nc i
j 1
where
Sh (d ij )
is the sharing
39
Sh (d ij )
1 (
Sh (d ij )
d ij
if d ij
share
share
share
(2.18)
otherwise
distance between any two solutions before they can be considered to be in the
same niche .
is a scaling factor less than or equal to 1.
d ij
d ij
k 1
where
fk
f kmax
( j)
fk
f kmin
(2.19)
fkmax and fkmin are the maximum and minimum objective function
share.
assigned fitness to a solution by its niche count. Although all solutions of any
particular rank have an identical fitness, the shared fitness value of a solution
residing in a less crowded region has a better shared fitness. This produces a
large selection pressure for poorly represented solutions in any rank. Hence
with scaling factor value equal to 1 compute the sharing function value and
niche count for all population members using the equation (2.17) and (2.18).
Thus, it is clear that in a population, a solution may not get any sharing effect
40
from some solutions and may get partial sharing effect from few solutions and
will get a full effect from itself. It is important to note that niche count is
always greater than or equal to 1. Then calculate the shared fitness
using F j'
Fj
nc j
reduces the fitness of each solution. In order to keep the average fitness of the
solutions in a rank the same as that before sharing, these fitness values are
scaled using equation (2.20) so that their average shared fitness value is the
same as the average assigned fitness value.
F jsc
F j (r )
(r )
F j'
(2.20)
F k'
k 1
where
F j'
(r ) is
GA IMPLEMENTATION
When applying GA to solve a particular optimization problem,
41
2.5.1
Pg 3
Pgn
Vg 1
Vg 2
1 4 3
Vgn Qc1 Qc 2
4 2 1
Qcn t1 t 2
3
tk
42
Fitness Evaluation
Genetic algorithm searches for the optimal solution by maximizing
following equations:
PS K
K S PS K
PS max
, if PS K
PS max
K S PS K
PS max
, if PS K
PS min
, otherwise
(2.21)
43
PVi K
KV Vi K
Vi max
, if Vi k
Vi max
KV Vi K
Vi min
, if Vi k
Vi min
PQ gK
, otherwise
K q Q gK
Q gmax
, if Q gk
Q gmax
K q Q gK
Q gmin
, if Q gk
Q gmin
where
(2.23)
, otherwise
K l S lK
PLKl
(2.22)
S lmax
, if S lk
S lmax
(2.24)
, otherwise
KS, Kv, Kq, and Kl are the penalty factors for the slack bus power
output, bus voltage limit violation, generator reactive power limit violation
and line flow violation respectively. Thereafter, all the penalty terms are
added together to get the overall penalty function.
K 0 i 1
NC
Nl
PLK
l
PQ iK
PVi K
PSK
K 0
NC N g
NC N P Q
NC
PF
K 0 i 1
(2.25)
K 0 l 1
Genetic Operators
The GA uses three main operating schemes namely, selection,
the population according to their fitness; the higher the fitness, the more
44
chance an individual has to be selected for the next generation. There are a
number of methods proposed in the literature for the selection operation.
Fitness proportionate selection is used in this work. Fitness proportionate
selection (Eshelman and Schaffer 1993), also known as roulette-wheel
selection, is a genetic operator used in genetic algorithms for selecting
potentially useful solutions for recombination. In this approach, the fitness
level is used to associate a probability of selection with each individual
chromosome. If fi is the fitness of the individual i in the population, its
probability of being selected is
Pi
fi
(2.26)
fj
j 1
45
2.5.3.2
in the population. Crossovers for real parameter GAs have the interesting
feature of having tunable parameters that can be used to modify their
exploration power. In the proposed approach each individual in the population
consists of two types of variables: real and integer. Hence a two-part
crossover which takes advantage of the special structure of the problem
representation was developed. First, the two parents are represented on the
floating point and integer parts. The BLX (Blended crossover operator)
(Eshelman and Schaffer 1993, Devaraj 2005, 2010) is used for real variables
and the standard single point crossover is applied for the integer part. Figure
2.3 represents the BLX- crossover operation for the one dimensional case. In
the figure, u1 and u2 are the selected individuals and umin and umax are the
lower and upper limits respectively and I= (u2-u1). In the BLX- crossover,
the offspring (y) is sampled from the space [e1, e2] as follows:
e1
repeat
(e 2
e 1 ) ; if u
min
(2.27)
max
sampling ; otherwise
where :
where :
(u 2
u1 )
e1
u1
e2
u2
(u 2
uniform
random
u1 )
number
[0 ,1 ]
u max
u min
e2
e1
u1
u2
Crossover
46
the population. Mutation randomly changes the new offspring. In this work,
the Non Uniform Mutation operator (Deb 2001) is applied to the mixed
variables with some modifications.
uk
1 p
1 M
1 p
M
u max
uk . 1 r
uk
u 1k
uk
where
u min . 1 r1
if r1
0.5
(2.28)
if r1
0.5
47
After mutation, the new generation is complete, and the procedure begins
again with the fitness evaluation of the population.
2.6
SIMULATION RESULTS
The proposed GA approach for solving the optimal power flow
optimization problem was applied to the IEEE 30-bus test system. Figure 3.4
represent the schematic diagram of the IEEE 30-bus system. The IEEE 30bus system has 6 generator buses, 24 load buses and 41 transmission lines of
which four branches are (6-9),(6-10),(4-12) and (28-27) with tap setting
transformers. The upper and lower voltage limits at all buses except the slack
bus are taken as 1.10 p.u and 0.95 p.u respectively. The slack bus voltage is
fixed at its specified value of 1.06 p.u. The generator cost coefficients and the
line parameters are taken from Alsac and Scot (1974).
48
The possible locations for reactive power sources are buses 10, 12,
15, 17, 20, 21, 23, 24 and 29. Two different cases were considered for
simulation and the results are presented.
Case (i): OPF using GA
Here the GA-based algorithm was applied to identify the optimal
control variables of the system under base-load condition, with cost
minimization objective and without considering the voltage stability of the
system. The upper and lower voltage limits at all the busbars except the slack
bus were taken as 1.10 and 0.95 respectively. The slack busbar voltage was
fixed to its specified value 1.06 p.u. Here the contingencies are not considered
and the GA- based algorithm was applied to find the optimal scheduling of
the power system for the base case loading condition given in (Alsac and Scot
1974). Generator active power outputs, generator bus voltages, transformer
tap settings and reactive power generation of capacitor bank were taken as the
optimization variables. The optimization variables are represented as floating
point numbers and integers in the GA population.
The initial population was randomly generated between the
variables lower and upper limits. Tournament selection was applied to select
the members of the new population. Blend crossover and mutation were
applied on the selected individuals. The performance of the GA generally
depends on the GA parameters used, in particular the crossover and mutation
probabilities in the ranges 0.6- 0.9 and 0.001 0.01 respectively was therefore
evaluated. It was run with different control parameter settings and the
minimization solution was obtained with the following parameter setting:
Population size
50
Crossover rate
0.9
Mutation rate
0.01
Maximum generations :
150
49
50
Variable setting
P1 (MW)
175.2070
P2 (MW)
48.5714
P5 (MW)
21.6667
P8 (MW)
22.6984
P11 (MW)
12.5397
P13 (MW)
12.0000
V1 (P.U)
1.0500
V2 (P.U)
1.0381
V5 (P.U)
1.0119
V8 (P.U)
1.0214
V11 (P.U)
1.0976
V13 (P.U)
1.0881
T11
0.9286
T12
0.9286
T15
0.9571
T36
0.9286
QC10 (MVAr)
2.8571
QC12 (MVAr)
4.2857
QC15 (MVAr)
4.2857
QC17 (MVAr)
5.0
QC20 (MVAr)
0.0
QC21 (MVAr)
5.0
QC23 (MVAr)
4.2857
QC24 (MVAr)
5.0
QC29 (MVAr)
5.0
Cost ($/hr)
802.0392
51
Figure 2.5
802.62
804.019
Evolutionary Programming
(Somasundaram et.al. 2004)
802.40
802.465
Proposed method
802.0392
52
100
Population size :
50
Crossover rate
0.8
Mutation rate
0.01
Variable
25
53
0.115
0.11
0.105
0.1
0.095
0.09
0.085
800
805
810
815
820
825
830
Minimum operating
cost solution
175.8663
49.0139
20.9582
22.3892
11.9904
12.5718
1.0548
1.0383
1.0097
1.0157
1.0904
1.0897
0.9750
0.9250
0.9750
0.9750
2
2
5
5
5
5
0
0
5
802.1208
0.1117
54
2.7
CONCLUSIONS
This chapter presents a MOGA algorithm approach to obtain the