Sunteți pe pagina 1din 12

Electr Eng (2010) 92:173184

DOI 10.1007/s00202-010-0173-3

ORIGINAL PAPER

Combined economic and emission dispatch problems using


biogeography-based optimization
Provas Kumar Roy S. P. Ghoshal S. S. Thakur

Received: 11 July 2009 / Accepted: 30 August 2010 / Published online: 15 September 2010
Springer-Verlag 2010

Abstract In this work, biogeography-based optimization


(BBO) is presented for solving different constrained economic load dispatch (ELD) problems combined with
economic emission aspects in power system. Nonlinear characteristics of generators like valve point discontinuities,
ramp rate limits and prohibited operating zones are considered in the problem. The simulation results show that
the proposed BBO algorithm based solutions prove to be
the best near-global optimal as compared to the solutions
based on NewtonRaphson, Tabu search, genetic algorithm
(GA), non-dominated sorting genetic algorithm (NSGA),
fuzzy logic controlled genetic algorithm (FCGA), particle
swarm optimization (PSO) and differential evolution (DE).
Keywords Biogeography-based optimization
Combined economic emission load dispatch
Genetic algorithm Migration Mutation Particle swarm
optimization
1 Introduction
Utilization of the existing resources in a planned way and
securing electricity at the cheapest rate without compromising the social benefits is of prime importance nowadays. The
main objective of ELD is to schedule the generating units output so as to meet the load demand at minimum operating cost
P. K. Roy (B) S. P. Ghoshal S. S. Thakur
Department of Electrical Engineering,
National Institute of Technology, Durgapur, West Bengal, India
e-mail: roy_provas@yahoo.com
S. P. Ghoshal
e-mail: spghoshalnitdgp@gmail.com
S. S. Thakur
e-mail: sst_nit_ee@yahoo.co.in

while satisfying all equality and inequality constraints. Since


environmental pollution is one of the major threats to the
human society, it has almost become compulsory to supply
the electricity not only at the cheapest rate, but also at a minimum level of emission. In Combined Economic and Emission Dispatch (CEED) problems, minimum emissions have
also been considered as one of the objectives along with cost
economy. Several classical methods, such as lambda iteration
method, gradient method, linear programming, interior point
method [1], etc. have been applied to solve economic load
dispatch assuming quadratic cost function. But unfortunately,
these methods lack feasibility in practical systems due to
nonlinear characteristics of the cost function. Non linear programming [2], which is fast and reliable along with powerful
mathematical optimization technique, has been employed to
solve the problems of ELD. A dynamic programming (DP)
[3] method may also solve the ELD problem with valve-point
discontinuities. However, the DP method suffers from dimensionality problem, thus imposing enormous computational
burden. In the recent years, complex constrained ELD problems have been solved by many population-based optimization techniques like genetic algorithms [46], evolutionary
programming (EP) [7], simulated annealing (SA) [8], particle swarm optimization (PSO) [912], hybrid evolutionary
programming (HEP) [13] and chaotic ant swarm optimization (CASO) [14].
Amongst the above population-based algorithms, the
inherent parallel search technique [6] of GA method has
made it faster than SA method. Real coded Genetic algorithm (RGA) [15] with arithmetic crossover, mutation and
elitism has been employed to solve the ELD problems more
efficiently than binary coded GA. Particle swarm optimization (PSO) is one of the modern heuristic algorithms and has
a great potential [11] to solve complex optimization problems. PSO algorithm is highly robust yet remarkably simple

123

174

Electr Eng (2010) 92:173184

to implement. Thus, it is quite pertinent to apply the PSO


with new modifications [16] to achieve better optimization
and handle the power system problems efficiently.
BBO [17] has some common features like other population-based algorithms. Like GA and PSO, BBO also shares
information amongst the solutions but BBO does not involve
in reproduction like GA. While GA solutions are lost at the
end of each iteration, BBO maintains its set of solution from
one iteration to next like PSO. But PSO solutions do not
change directly; initially the velocities next the positions
(solutions) of the particles are changed, whereas BBO solutions are changed directly via migration from other solutions.
In this paper BBO algorithm has been employed to four
different CEED problems having linear/non-linear operating constraints, smooth/non-smooth cost functions and
compared its simulation results to those of PSO, GA, differential evolution (DE) [18], NewtonRaphson (NR) [19],
fuzzy logic controlled genetic algorithm (FCGA) [20] and
non-dominated sorting genetic algorithm (NSGA) [21].

Step 1: Calculate h i for each unit which is given by:


hi =

FCi (Pgimax )/(Pgimax )


E i (Pgimax )/(Pgimax )

i = 1, 2, . . . , n
(4)

Step 2: h i values are arranged in the ascending order.


Step 3: Maximum capacity of each generating unit is added
one at a time starting from the unit having smallest
 max
Pgi PD .
h i until
Step 4: h i associated with the last unit is the price penalty
factor h for the given load.
2.2 Constraints
(i) The sum of the generated powers of all generating
units must be equal to sum of power demanded by the
load and the total transmission loss.
n


2 Mathematical problem formulation

Pgi = PD + PL

(5)

i=1

2.1 Fitness function


Emission dispatch may be associated with conventional ELD
problem by adding the emission cost with the fuel cost. This is
Combined Economic and Emission Dispatch (CEED) problem. Mathematically, the fitness function of CEED can be
described by introducing a price penalty factor [22] as follows:
Minimize TC = FC(Pg ) + h E(Pg )

FC(Pg ) =
E(Pg ) =

i=1
n


FCi (Pgi ),

(2)

E i (Pgi ),

(3)

i=1

active power generation of the


ith generating unit, MW
TC
fitness function as total operating cost, $/h
the total fuel cost, $/h
FC(Pg )
the total amount of emission, lb/h (kg/h)
E(Pg )
FCi (Pgi ) the fuel cost for supplying the specified
load of the ith generating unit, $/h
the amount of emission of
E i (Pgi )
the ith generating unit, lb/h (kg/h)
h
price penalty factor, $/lb ($/kg)
n
number of generators
The penalty factor [22] for a given load may be found out
by the following steps:
Pgi

123

PL =

n 
n


Pgi Bi j Pg j +

i=1 j=1

n


Bi0 Pgi + B00

i=1

(6)

(1)

where
n


where PL is the transmission loss formula known as


Krons loss formula or B coefficient formula is given
by:

where B00 , Bi0 , Bi j are the loss coefficients which


may be assumed to be constant under normal operating condition.
(ii) The power output of each generating unit must be
greater than or equal to the minimum power permitted and also be less than or equal to maximum power
permitted on that specified unit. Thus the inequality
constraint is expressed as:
Pgimin Pgi Pgimax

(7)

where Pgimin , Pgimax are the minimum and the maximum


power outputs of ith unit respectively, MW.
(iii) The active power output of each generating unit is limited by ramp up/down rate at each hour as in (8)


Max Pgimin , Pgi0 DRi Pgi


Min Pgimax , Pgi0 + URi

(8)

Electr Eng (2010) 92:173184

175

where Pgi0 is the previous operating active power output of ith generating unit, MW; DRi , URi are the down
rate and up rate limits of ith generating unit respectively, MW/h.
(iv) Each unit must avoid operation in prohibited zones.
The operating zone of ith unit may be described as
follows:
l
Pgimin Pgi Pgi,1
u
l
Pgi,
j1 Pgi Pgi, j , j = 2, 3, . . . , n i
u
Pgi,n
i

Pgi

where
ni
l
Pgi,
j
u
Pgi,
j1

(9)

Pgimax

number of prohibited zones


of ith generating unit;
lower active power limit of prohibited zone
j of ith generating unit, MW;
upper active power limit of prohibited zone
j 1 of ith generating unit, MW.

where ai ($/MW2 /h), bi ($/MW/h), ci ($/h) are the fuel


cost coefficients of the ith generating unit.
2.3.2 Cost function considering valve point discontinuity
For more practical and accurate model of the cost function,
the above expression needs to be modified. Modern thermal power plants consist of generating units having multiple valve steam turbines in order to incorporate flexible
operational facilities. The generating units with multi valve
turbines have very different cost curves compared to those
defined by Eq. (10). Total cost of generating units with valve
point loading is given by:
FC(Pg ) =

n


ai Pgi2 + bi Pgi + ci

i=1







+ di sin ei Pgimin Pgi 

(11)

where di ($/h), ei (rad/MW) are the fuel cost coefficients of


the ith generating unit reflecting the valve point discontinuity.

2.3 Cost function

2.4 Emission function

The cost function of a thermal generator with multi-valve


steam turbines has a very different generator fuel cost curve
(as shown in Fig. 1) as compared to the conventional smooth
one. In general, the valve-point effect shows ripple when
every steam valve begins to open. This fuel cost curve is a
non-convex function. In the border points between any two
adjacent ripples of the curve, the cost function becomes nondifferentiable and non-smooth, since its left and right derivatives are not equal at these points.

The emission generated by each generating unit may be


approximated as a quadratic function of the power output
of the generator. The total amount of emission released is
given by:
E(Pg ) =

n 


i Pgi2 + i Pgi + i

(12)

i=1

where i (Ib/MW2 /h, kg/MW2 /h ), i (Ib/MW/h,kg/MW/h ),


i (Ib/h, kg/h) are the emission coefficients of the ith generating unit.

2.3.1 Quadratic cost function


Total cost of generating units without valve point effect is
given by:

3 Biogeography

n 


ai Pgi2 + bi Pgi + ci
FC(Pg ) =

Biogeography is an area of scientific study that describes


natures way of distributing species. Figure 2 describes a
model of species abundance in a habitat [23]. The immigration curve of Fig. 2 shows that immigration rate to the habitat
is maximum (I ), when there are zero species in the habitat.
As the number of species increases, the habitat becomes more
crowded; fewer species are able to successfully survive immigration to the habitat and the immigration rate decreases. The
largest possible number of species that the habitat can support is Smax , at which point the immigration rate becomes
zero. The emigration curve of Fig. 2 depicts that if there are
no species in the habitat, the emigration rate will be zero.
As the number of species increases, the habitat becomes
more crowded; more species are able to leave the habitat
to explore other possible residences and the emigration rate

i=1

Generator
Cost ($/hr)

Power Output (MW)


Fig. 1 Generators cost curve with valve-point effect

(10)

123

176

Electr Eng (2010) 92:173184

(s +s )Ps +s+1 Ps+1

PS = (s +s )Ps +s1 Ps1 +s+1 Ps+1

Immigration s

(s +s )Ps +s1 Ps1

S =0
1 S Smax 1
S = Smax .

(16)

Rate

Emmigration s

S0

4 Algorithms

S max

No of Species

4.1 Genetic algorithm

Fig. 2 Species model of a single habitat

increases. The maximum emigration rate is E, which occurs


when the habitat contains the largest number of species
(Smax ). The equilibrium number of species is S0 , at which
point the immigration and the emigration rates are equal.
The immigration and the emigration curves in Fig. 2 [17] are
shown as straight lines but, in general, they might be more
complicated curves. Nevertheless, this simple model gives
us a general description of the process of immigration and
emigration.
From Fig. 2, the immigration rate, s and the emigration
rate, s for S number of species may be formulated as follows:


S
(13)
s = I 1
Smax
ES
s =
(14)
Smax
Mathematically, the concept of emigration and immigration
may be represented by a probabilistic model. Let us consider
the probability Ps of a habitat that contains exactly S species.
Ps changes from time t to time t + t as follows [17]:
Ps (t + t) = Ps (1 s t s t) + Ps1 s1 t
+Ps+1 s+1 t

(15)

where s and s are the immigration and the emigration rates


when there are S species in the habitat. This equation holds
because in order to have S species at time t + t, one of the
following conditions must hold:
(1) there are S species at time t and no immigration or emigration occurs between t and t + t;
(2) there are (S 1) species at time t and one species immigrates;
(3) there are (S + 1) species at time t and one species emigrates.
If time t is small enough so that the probability of more
than one immigration or emigration can be ignored, then
taking the limit of (15) as t 0 gives the following equation [17]:

123

Genetic algorithm (GA) [4,5] is an increasingly popular optimization technique being applied to many fields of endeavor,
motivated by Darwins theory of evolution and the concept
of survival of the fittest. GA uses processes analogous to
genetic recombination and mutation to promote the evolution of a population that best satisfies a predefined goal and
objective. Crossover refers to the mixing of information from
both parents to create the children. Thus, in the crossover
process, more fit individuals are allowed to produce more
off-springs than less fit individuals, which tend to homogenize the population and improve the average result as the
algorithm progresses. Subsequent mutations of the off-spring
add diversity to the population and explore new areas of the
parameter search space. The probability that a particular element is mutated is governed by the mutation probability.
Steps of GA as implemented for optimization are:
Step 1. Initialization of binary chromosome strings of n p
population, each consisting of the parameters to be
optimized
Step 2. Decoding of strings and evaluation of fitness of each
string
Step 3. Selection of the elite strings in order of increasing
fitness from minimum value
Step 4. Apply crossover and mutation to generate offspring
Step 5. Calculate the objective function value and find the
fitness of the offspring
Step 6. Adopt the elitist strategy i.e. replace the worst offspring by the elite string
Step 7. If the generation cycle number (k) reaches the maximum limit, go to Step 8. Otherwise, set generation
cycle number, k = k + 1, and go back to Step 3
Step 8. Terminate the GA operation.
4.2 Particle swarm optimization
It is a population-based evolutionary algorithm. It is the simulation of the social behavior that motivates PSO. Here, the
population is called swarm. In PSO, multiple candidate
solutions coexist. Here each candidate solution is associated with a velocity. Each candidate solution, called a particle, flies in the problem space (similar to the search process
for food of a bird swarm) looking for the optimal position.

Electr Eng (2010) 92:173184

177

A particle with time adjusts its position according to its own


memory and cognitive experience, while adjusting to the
social experience of neighboring particles as well. If a particle discovers a promising new solution, all the other particles
will move closer to it, exploring the region more thoroughly
in the process requiring less computational burden and generally few lines of code [10]. Based on PSO concept, mathematical equations for the searching process are:
Velocity updating equation:
vik+1

=w

vik

+ c1 r 1 ( pBesti

+c2 r 2 (gBest xik )

xik )
(17)

Position updating equation:


xik+1 = xik + vik+1

(18)

where
vik

velocity of the of ith particle at kth generation


cycle,
k+1
modified velocity of the ith
vi
particle at (k + 1)th generation cycle,
w
inertia weight factor,
pBesti the position best of the ith particle,
gBest the group best,
c1, c2 the cognitive and the
social parameters respectively,
r 1, r 2 random numbers in the interval [0, 1],
the current position of the ith particle at
xik
the kth generation cycle,
the modified position of the ith
xik+1
particle at the (k + 1)th generation cycle,
Main steps of PSO techniques are given below:
Step 1. The particles are randomly generated between the
operating limits of the generators.
Step 2. The values of the fitness function of the particles are
evaluated using (1) and the dimensions (variables)
of the particles are initialized as pBest (s).
Step 3. The best value of pBest(s) is represented as gBest.
Step 4. The particles velocities and positions are updated
using (17) and (18) respectively.
Step 5. The new fitness function values are evaluated using
the updated positions of the particles. If the current
position of the particle is better than its previous
pBest, the pBest is updated by the current particle,
otherwise not updated. The updated gBest is the best
among all the pBest(s).
Step 6. If the stopping criterion is satisfied, go to Step 7,
otherwise, go to Step 4.
Step 7. The particle that generates the latest gBest yields
the optimal variables.

4.3 Biogeography-based optimization (BBO)


Biogeography-based optimization technique (BBO) [17] has
been developed based on the theory of biogeography. BBO
concept is mainly based on migration and mutation. The concept and mathematical formulation of migration and mutation steps are given below:
A. Migration
This BBO algorithm [17] is similar to other population-based
optimization techniques where population of candidate solutions is represented as vector of real numbers. Each real
number in the array is considered as one suitability index
variable (SIV). Fitness (in BBO, a term called habitat suitability index (HSI)) of each candidate solution is evaluated
using its SIVs. HSI represents the fitness of each candidate
solution. High HSI solution represents better quality solution and low HSI solution represents inferior solution in the
optimization problem. The emigration rate and the immigration rate of each solution are used to probabilistically
share information between habitats. The immigration rate
(s ) and the emigration rate (s ) can be evaluated by Eq. (13)
and (14) respectively. Using habitat modification probability each solution is modified based on other solutions. The
immigration rate, s of each solution is used to probabilistically decide whether or not to modify each SIV in that
solution. If a SIV in a given solution is selected for modification, emigration rates, s of other solutions are used to
probabilistically select which of the solutions should migrate
a randomly selected SIV to that solution. The main difference between the recombination approach of evolutionary
strategies (ES) and the migration process of BBO is that
in ES, the global recombination process is used to create
a completely new solution, while in BBO; the migration is
used to bring changes within the existing solutions. In order
to prevent the best solutions from being corrupted by the
immigration process, few elite solutions are kept in BBO
algorithm.
B. Mutation
Due to some natural calamities or other events, HSI of a natural habitat may change suddenly and it may deviate from its
equilibrium value. In BBO, this event is represented by the
mutation of SIV and species count probabilities are used to
determine mutation rates. The species count probability can
be calculated using the differential equation (16). Each population member has an associated probability, which indicates
the likelihood that it exists as a solution for a given problem.
Mutation rate of each set of solution can be calculated in
terms of species count probability using Eq. (19) [17]:

123

178

Electr Eng (2010) 92:173184


m (s) = m max

1 PS
Pmax


(19)

where m(s) the mutation rate for habitat contains S species,


m max maximum mutation rate, Pmax maximum probability.
From (19), it can be inferred that if the probability (Ps )
of a given solution is very low, then that solution is likely to
mutate to some other solution. Similarly, if the probability
(Ps ) of a solution is high, then that solution has very little
chance to mutate. Now, both habitats having very high HSI
values and very low HSI values have low species count probability (Ps ) [17] and therefore are equally probable for mutation i.e. they have high chances to produce much improved
SIVs in the later stage. But, medium HSI solutions have
higher values of species count probability (Ps ) [17] and less
chance to create better solutions after mutation operation.
This mutation scheme tends to increase diversity among the
populations. Without this modification, the highly probable
solutions will tend to be more dominant in the population.
This mutation approach makes both low and high HSI solutions likely to mutate, which gives a chance of improving
both types of solutions in comparison to their earlier values. Few elite solutions are kept in the mutation process to
save the best solutions, so if a solution becomes inferior after
mutation process then, previous solution (solution of that
set before mutation) can revert back to that place again, if
needed. So, mutation operation is a high-risk process. It is
normally applied to both poor and better solutions. Since
medium quality solutions are in improving stage, so it is better not to apply mutation on medium quality solutions.
Here, mutation of a selected solution is performed simply
by replacing it with a randomly generated new solution. Other
than this any other mutation scheme that has been implemented for GA can also be implemented for BBO.

where SIViq is the active power output of the qth unit of the
ith individual. The dimension of the habitat is s m. All
these components in each individual are real values. HSIi
indicates the objective function of the ith habitat containing
m active power outputs (SIVs) in $/h.
The algorithm of the proposed method is as enumerated
below.
Initialization of the BBO input parameters.
The initial values of power outputs i.e. SIVs of each habitat H i are randomly selected while satisfying different
equality and inequality constraints of the CEED problem. A population size (n) of habitats H is generated.
Each habitat represents a potential solution.
Calculate each HSIi i.e. value of objective function for
each i-th habitat of the population set n for given emigration rate s , immigration rate, s and species, S.
Based on the values of HSI, elite habitats are identified.
Each habitat is modified by performing probabilistically
immigration and emigration operations as described in
Sect. 4.3.A and HSI of each modified habitat is recalculated. Feasibility of a solution is verified when each SIV
satisfies equality and inequality constraints of generator
as mentioned in the specific problem.
Species count probability of each habitat is updated using
(16). Mutation operation is performed on the habitats as
described in Sect. 4.3.B and HSI value of each new habitat
is computed.
The worst habitats of the population set are replaced by
the elite habitats.
Feasibility of a problem solution is verified i.e. each SIV
should satisfy equality and inequality constraints of generator as mentioned in the specific problem.
Iteration updating or stop iterations after a predefined
number of iterations.

4.3.1 BBO algorithm for the CEED problem


5 Input parameters
In this paper, BBO algorithm has been employed to solve
the constrained CEED problem and to find the optimal or
near optimal active power outputs satisfying both equality
and inequality constraints.
In BBO algorithm, applied to the CEED problem, each
habitat is defined by a vector with SIVs where each SIV represents the active power output of each generating unit. Each
individual habitat within the total n habitats represents a candidate solution. For example, if there are m generating units
that must be operated to provide power to the loads,
Then, the ith individual habitat H i can be defined as follows:
H i = SIViq = [SIVi1 , SIVi2 , . . . , SIVim ]
where i = 1, 2, . . . , S; q = 1, 2, . . . , m

123

(20)

Since the performance of the algorithms sometimes depends


on input parameters, these should carefully be chosen. A
population size of 100 and number of generations of 300 are
considered for all the algorithms. The input control parameters of the algorithms are given below:
5.1 For GA: mutation probability = 0.004, crossover probability = 0.75, number of elite strings = 5.
5.2 For PSO: The choice of weighting factors c1 and c2
affects the performance of PSO. The best and commonly used values, c1 = c2 = 2.05 [16] are explained
below.
5.3 For BBO: habitat modification probability, P mod = 1;
mutation probability = 0.005, maximum immigration
rate, I = 1, maximum emigration rate, E = 1, step

Electr Eng (2010) 92:173184


Table 1 GA performance with
different input parameters (test
system III)

Italic entries are the optimal


input parameters setting of GA

179

GA parameters

Total cost (average of


20 random trials) ($/h)

GA parameters
Mutation
probability

Total cost (average of 20


random trials) ($/h)

Mutation
probability

Crossover
probability

Crossover
probability

0.01

0.50

22,219

0.004

0.50

22,187

0.01

0.75

22,206

0.004

0.75

22,164

0.01

1.00

22,227

0.004

1.00

22,199

0.005

0.50

22,192

0.003

0.50

22,194

0.005

0.75

22,179

0.003

0.75

22,181

0.005

1.00

22,205

0.003

1.00

22,211

Table 2 PSO performance with different input parameters (test system III)
PSO parameters

Total cost
(average of 20
random trials)
($/h)

PSO parameters

c1

c2

22,228

2.0

2.0

2.0

22,206

2.0

2.05

22,198

2.0

Total cost
(average of 20
random trials)
($/h)

PSO parameters

Total cost
(average of 20
random trials)
($/h)

c1

c2

c1

c2

1.9

1.9

22,186

2.05

2.05

1.9
1.9

2.05

22,175

2.05

2.1

22,172

2.1

22,184

2.1

2.0

22,179

1.9

2.10

22,202

2.05

1.9

22,193

2.1

2.05

22,171

2.0

1.9

22,197

2.05

2.0

22,181

2.1

2.1

22,188

22,167

Italic entries are the optimal input parameters setting of PSO

size for numerical integration, =1, maximum generation cycles = 300, elitism parameter = 2, number of
SIVs of BBO algorithm = number of generating units
and number of habitats = 100.
The above best parameters are set after investigating their
effects on the performances of PSO, GA and BBO in Test system III for the total load of 1,000 MW. For each parameter
setting, 20 random trials are performed and average results
are shown in Tables 1, 2 and 3 for GA, PSO and BBO respectively.
6 Simulation and results
The proposed algorithm is implemented in four test systems
and its performance is compared to other population-based
optimization techniques like NR, Tabu search, GA, NSGA,
FCGA, DE and PSO which were already tested and reported
by earlier authors. The performance of each system has been
judged out of 50 trials. The programming was written in
MATLAB language, using MATLAB 7.1 on Pentium IV processor, 1.6 GHz with 256 MB RAM.
(a) Test system I: A case of three generating units with a total
demand of 850 MW is used to test the effectiveness of
BBO algorithm. The system data are the same as [24].

Two objective functions namely, fuel cost and emission


cost are used in this case. The performance of BBO is
compared to that of Tabu search [24] and non dominated
sorting genetic algorithm-II (NSGA-II) [21]. The computation test results of active power output schedule,
total active power output (TG), transmission loss (TL),
fuel cost (FC), emission (E) and total operational cost
(TC) given in Table 4 show that the BBO gives the best
fuel cost (italic) amongst three algorithms though the
changes in the fuel costs are not very significant because
of very small system (three generating units only). The
best emission found by BBO is better than that found
by NSGA-II [21] but it is inferior to that found by Tabu
search [24] though the changes are very insignificant
because of very small system. Convergence characteristic for best fuel cost is shown in Fig. 3.
(b) Test system II: This system consists of six generating units having quadratic cost and emission functions. The cost and emission functions coefficients and
transmission loss coefficients are taken from [21]. The
penalty price factors (PF) for 500, 700 and 900 MW
loads are 43.8983, 44.788 and 47.8222 respectively
which are computed from the algorithm given in [18].
The sample test results for NewtonRaphson (NR) [19],
fuzzy logic controlled genetic algorithms (FCGA) [20],
non-dominated sorting genetic algorithm (NSGA) [21]

123

180

Electr Eng (2010) 92:173184

Table 3 BBO performance with different input parameters (test system III)
BBO input parameters

Total cost
(average of 20
random trials)
($/h)

Mutation
probability

Maximum
immigration
rate

Maximum
emigration
rate

0.01

0.75

0.75

0.01

0.75

0.90

0.01

0.75

0.01

BBO input parameters

Total cost
(average of 20
random trials)
($/h)

Mutation
probability

Maximum
immigration
rate

Maximum
emigration
rate

22,189

0.005

0.90

1.00

22,136

22,182

0.005

1.00

0.75

22,147

1.00

22,177

0.005

1.00

0.90

22,138

0.90

0.75

22,180

0.005

1.00

1.00

22,124

0.01

0.90

0.90

22,171

0.004

0.75

0.75

22,174

0.01

0.90

1.00

22,164

0.004

0.75

0.90

22,160

0.01

1.0

0.75

22,175

0.004

0.75

1.00

22,157

0.01

1.0

0.90

22,168

0.004

0.90

0.75

22,162

0.01

1.0

1.0

22,163

0.004

0.90

0.90

22,154

0.005

0.75

0.75

22,166

0.004

0.90

1.00

22,149

0.005

0.75

0.90

22,154

0.004

1.00

0.75

22,155

0.005

0.75

1.00

22,149

0.004

1.00

0.90

22,147

0.005
0.005

0.90
0.90

0.75
0.90

22,158
22,142

0.004

1.00

1.00

22,139

Italic entries are the optimal input parameter settings of BBO


Table 4 Simulation results of
best fuel cost and best emission
of 3-generator system with
different algorithms

Italic entries are the objective


function value obtained by BBO

Unit

Best fuel cost

Best emission

Tabu Search [24] NSGA-II [21] BBO

Tabu Search [24] NSGA-II [21] BBO

Pg1 (MW) 435.69

435.885

435.195

502.914

505.810

Pg1 (MW) 298.828

299.989

299.972

254.294

252.951

250.446

Pg3 (MW) 131.28

129.951

130.662

108.592

106.023

105.724

TG (MW)

865.826

865.829

865.8

864.784

864.746

15.8

14.784

14.7459

865.798

TL (MW)

15.798

15.826

15.8290

FC ($/h)

8,344.598

8,344.598

8,344.592 8,371.143

8,363.627

8,365.113

E (kg/h)

0.09863

0.09860

0.09869

0.09593

0.09592

8345.4

Objective Function

8345.3
8345.2
8345.1
8345
8344.9
8344.8
8344.7
8344.6
8344.5

50

100

150

200

250

300

No Of Iterations
Fig. 3 Convergence characteristic of objective function using BBO
(for test system Ibest fuel cost)

123

508.576

0.0958

and BBO for 500 MW, 700 MW & 900 MW loads are
illustrated in Table 5. The simulation results show that
BBO gives better fuel cost and emission (italic) than NR
[19] for all three load demands. It is also observed that
though BBO gives slightly higher fuel cost than FCGA
and NSGA, the emission quality is minimum (italic)
with BBO algorithm in all three cases.
(c) Test system III: In this example, six generating units
with the constraints of ramp rate limits and prohibited
zones of all units are considered for CEED to test the
effectiveness of the BBO algorithm. The data for cost
coefficients, active power limits, ramp rate limits and
prohibited operating zones are taken from [25]. The
data of emission coefficients are shown in Table 6. Simulation results of GA, PSO and BBO for total loads of
1000 and 1200 MW with 300 iterations are shown in

Electr Eng (2010) 92:173184

181

Table 5 Simulation results of 6-generator system with different algorithms


Load (MW)

Algorithms

Pg1 (MW)

Pg2 (MW)

Pg3 (MW)

Pg4 (MW)

Pg5 (MW)

Pg6 (MW)

TL (MW)

FC ($/h)

E (kg/h)

500

NR [19]

59.873

39.651

35.000

72.397

185.241

125.000

17.162

28,550.15

312.513

FCGA [20]

65.23

24.29

40.44

74.22

187.75

125.48

17.41

28,231.06

304.90

NSGA [21]

54.048

34.25

54.497

80.413

161.874

135.426

20.508

28,291.119

284.362

700

900

BBO

49.894

33.908

62.987

82.881

153.405

139.0324

22.1073

28,318.5060

279.3092

NR [19]

85.924

60.963

53.909

107.124

250.503

176.504

34.927

39,070.74

528.447

FCGA [20]

80.16

53.71

40.93

116.23

251.20

190.62

32.85

38,408.82

527.46

NSGA [21]

86.286

60.288

73.064

109.036

223.448

184.111

36.234

38,671.813

484.931

BBO

84.018

61.9268

78.7284

110.9188

211.9852

187.8924

39.262

38,828.266

476.408

NR [19]

122.004

86.523

59.947

140.959

325

220.063

54.498

50,807.24

864.060

FCGA [20]

111.40

69.33

59.43

143.26

319.40

252.11

54.92

49,674.28

850.29

NSGA [21]

120.0587

85.202

89.565

140.278

288.614

233.687

57.405

50,126.059

784.696

BBO

115.7087

100.755

101.445

145.0146

282.058

212.417

57.400

50,297.271

765.087

Italic entries are the fuel cost (FC) and emission (E) obtained by BBO algorithm

Table 6 Data of emission coefficients of 6-generating unit


Unit no.

(lb/MW2 /h)

(lb/MW/h)

(lb/h)

0.013

1.7

22.050

0.019

1.8

21.90

0.020

1.249

22.983

0.033

1.36

24.313

0.015

2.1

25.1

0.018

1.9

27.01

Table 7. Results show the total cost obtained from the


BBO (italic) is the lowest for each load. The emission
dispatches are also lowest for both loads (italic) though
the fuel cost is not the lowest for 1,000 MW load. The
Table 7 Simulation results of
6-generator system having ramp
rate limits and prohibited
operating zones with different
algorithms

Unit

penalty price factors (PF) for 1,000 and 1,200 MW loads


are computed from the algorithm given in [18]. Convergence characteristics of the total cost using GA, PSO
and BBO for 1,200 MW load shown in Fig. 4 also show
that the total cost obtained from the BBO is the lowest
and thus verify that the BBO yields better solution.
(d) Test system IV: In this test system, the combined economic and emission load dispatch problem (CEED) of
14 generating units with transmission loss is considered.
To incorporate more realistic cost function, valve point
discontinuity is included in cost function of each unit.
The full system data of [18] are employed. The comparative test results with DE [18], GA, PSO and BBO
algorithms for the total load of 1,500, 2,000 and 2,500
MW and the convergence characteristics of the total

Algorithms
Demand (1000 MW)
PSO

GA

Demand (1200 MW)


BBO

PSO

GA

BBO

Pg1 (MW)

320.0000

320.0000

320.0000

329.6465

329.1083

349.2169

Pg2 (MW)

164.6135

140.0000

160.0000

198.1324

198.0966

200.0000

Pg3 (MW)

138.1051

142.1484

137.7791

240.0493

243.8283

211.7796

Pg4 (MW)

77.3720

90.0000

80.0000

124.792

121.3974

131.0000

Pg5 (MW)

189.9837

197.5012

191.8292

199.5031

199.8071

200.0000

Pg6 (MW)

119.5071

119.9993

120.0000

119.9729

119.4460

120.0000

TG (MW)

1,009.5813

1,009.6489

1,009.6084

1,212.116

1,212.1726

1,211.9965

TL (MW)

9.5813

9.6489

9.6084

12.116

12.1823

11.9964

PF

6.1152

6.1152

6.1152

12.8718

12.8718

12.8718

TC ($/h)

22,130

22,130.8

22,108.3

50,650.0

50,728.5

50,330.7

FC ($/h)

12,205.6

12,199.8

12,206.2

14,725.4

14,725.9

14,715.4

E (lb/h)

1,622.85

1,623.98

1,619.27

2,790.95

2,797.0

2,766.9

123

182

Electr Eng (2010) 92:173184


4

Total Operational Cost ($/hr)

Table 9 Ranking in terms of frequencies of occurrence of GA, PSO


and BBO with total 100 runs for test system III and test system IV
together

x 10

5.6

BBO
GA
PSO

5.5

Algorithms

Rank
First

Second

GA

12

40

48

PSO

10

44

46

BBO

78

16

5.4

5.3

Third

5.2

5.1

50

100

150

200

250

300

No Of Iterations

Fig. 4 Convergence characteristics of the total cost for 1200 MW load


using GA, PSO and BBO (for test system III)

cost using GA, PSO and BBO for 1500 MW load are
illustrated in Table 8 and Fig. 5, respectively. The simulation results show that the total cost, the fuel cost and
the emission found by BBO (bold) are reduced by 3.24,

3.11 and 3.8% respectively as compared to the corresponding worst results (italic) for the load of 1,500 MW.
For the load of 2,000 MW, the total cost, the fuel cost
and the emission obtained by BBO (bold) are reduced
by 3.84, 0.15 and 11.76% respectively as compared to
the corresponding worst results (italic). For the load of
2500 MW, the total cost, the fuel cost and the emission
obtained by BBO (bold) are reduced by 4.25, 0.59 and
9.33% respectively as compared to the corresponding
worst results (italic).
(e) Robustness test: A meaningful and acceptable conclusion about the optimizing techniques may only be

Table 8 Simulation results of 14-generator system with 1,500, 2,000 and 2,500 MW loads
Unit

Demand (1,500 MW)

Demand (2,000 MW)


BBO

DE [18] PSO

Demand (2,500 MW)

DE [18] PSO

GA

Pg1 (MW)

150.20

150.000

150.0000 150.0000 239.76

239.7598 264.0103 239.7594 329.53

329.5196 339.1189 329.5199

Pg1 (MW)

150.23

150.000

150.0000 150.0000 150.01

150.0000 150.0000 150.0000 219.54

150.0000 226.0364 157.3679

Pg3 (MW)

96.51

94.7998

94.8789

94.9987

129.9988 102.7986 126.7956 129.99

130.0000 103.5845 130.0000

Pg4 (MW)

120.01

130.000

75.0000

119.7331 119.75

119.7331 119.6997 119.7331 120.06

129.9990 122.9847 130.0000

Pg5 (MW)

150.10

150.000

150.0000 150.0000 199.86

150.0000 200.0000 150.0000 249.75

231.14

250.0000 249.7482

Pg6 (MW)

135.25

135.000

138.6355 135.0000 284.59

281.3344 277.2228 284.5995 384.25

384.33

420.8942 434.2207

Pg7 (MW)

135.07

135.0000 135.0000 135.0000 234.86

184.8666 238.0072 184.8665 284.43

284.5997 258.3177 284.7409

Pg8 (MW)

60.59

109.8660 109.0982 109.8671 159.73

159.7331 157.7048 159.7331 209.56

300.0000 240.0000 209.8249

Pg9 (MW)

125.10

124.7330 124.8268 124.7331 124.89

161.9994 124.4686 161.8835 161.74

162.0000 162.9238 162.0000

Pg10 (MW) 139.09

99.7998

153.9034 99.7997

137.32

160.0000 136.7911 160.0000 159.89

160.0000 138.9684 160.0000

Pg11 (MW) 58.77

23.2322

57.3947

57.3999

66.65

79.9986

65.0000

80.0000

79.95

80.0000

68.6673

Pg12 (MW) 79.97

80.0000

57.1477

77.7971

79.95

80.0000

79.7628

80.0000

79.93

80.0000

79.9905

80.0000

Pg13 (MW) 64.07

85.0000

74.2493

62.2823

84.97

85.0000

62.5000

85.0000

84.84

85.0000

84.5621

85.0000

95.21

GA

BBO

DE [18] PSO

GA

BBO

80.0000

Pg14 (MW) 52.95

52.3999

52.5945

52.3999

52.43

52.3999

52.3999

52.3999

52.65

52.3999

52.4009

54.9998

TG (MW)

1,517.9

1,519.83

1522.7

1519.01

2030.0

2034.8

2030.4

2034.8

2546.1

2559.0

2548.4

2547.42

TL (MW)

17.88

19.8314

22.7290

19.0109

29.98

34.8236

30.3658

34.7706

46.11

58.9872

48.4494

47.4223

PF

1.2191

1.2191

1.2191

1.2191

1.5299

1.5299

1.5299

1.5299

1.5716

1.5716

1.5716

1.5716

TC ($/h)

8,503.7

8,754.5

8,745.4

8,470.9

14,537

14,941

15,017.3

14,462.5

22,797

23,152.9

23,704.6

22,738.4

FC ($/h)

6,869.9

7,109.9

7,062.1

6,849.3

9,592.4

10,166

9,889.3

9,874.3

12,871

13,427.0

13,301.8

13,223.7

E (lb/h)

1,340.2

1,349.0

1,380.8

1,330.2

3,232.2

2,990.6

3,351.8

2,999.0

6,316.3

6,188.5

6,619.2

6,054.2

Bold entries are the total cost (TC), fuel cost (FC) and emission (E) values obtained by BBO algorithm
Italic entries are the total cost (TC), fuel cost (FC) and emission (E) values obtained by the worst algorithm (GA)

123

Electr Eng (2010) 92:173184

183

10200

Total Operational Cost ($/hr)

erators and it is observed that the proposed algorithm exhibits


better performance compared to other techniques like GA and
PSO which are known as best until now. In the last CEED
problem it is revealed that biogeography-based optimization
(BBO) shows least values of total cost (objective function) as
compared to DE, PSO and GA for all three different loads.
Considering all these results with different loads, different
constraints and different cost functions, it may finally be concluded that the proposed algorithm works very well for the
near-global optimization of combined economic and emission dispatch problems.

BBO
GA
PSO

10000
9800
9600
9400
9200
9000
8800
8600
0

50

100

150

200

250

300

No Of Iterations

References
Fig. 5 Convergence characteristics of the total cost for 1500 MW load
using GA, PSO and BBO (for test system IV)

drawn from the frequency of occurrence of convergence


towards a predefined specific value. An algorithm is
robust, if it gives similar results during all the runs.
Hundred independent trial runs for each GA, PSO and
BBO optimization algorithms for test system III and
test system IV are individually carried out to judge the
robustness of the algorithms. Frequencies of occurrence
of different optimization techniques for test system III
and test system IV in order of merit ranks are shown
in Table 9. From this table it is noticed that the performance of PSO and GA are quite similar but inferior
to BBO. Therefore, BBO yields more robust optimal
solutions.

7 Conclusion
In this paper, biogeography-based optimization has been
introduced to solve combined economic emission load dispatch problem with different kinds of cost functions. Its
results are compared to those of other well established algorithms like PSO, GA, DE [18], NR [19], FCGA [20], NSGA
[21] and Tabu search [24]. It is revealed in the first example
that among all the algorithms, BBO gives the best fuel cost
and it is also found that the best emission solutions of the
proposed algorithm are comparable to those of Tabu search
and NSGA. For the 6-generator system without prohibited
operating zones and ramp rate limits (second example), it is
observed that the fuel costs found with different algorithms
are comparable and emission level found with BBO technique is improved significantly for all three load demand
cases. In the third example, nonlinear characteristics of the
generators such as ramp rate limit and prohibited operating
zones are introduced for practical operation of thermal gen-

1. Wood AI, Woolenburg BF (1996) Power generation operation and


control. Wiley, New York
2. Nanda J, Hari L, Kothari ML (1994) Economic emission load dispatch with line flow constraints using a classical technique. IEE
Proc Gen Transm Distrib 141(1):110
3. Shoults RR et al (1986) A dynamic programming based method
for developing dispatch for developing dispatch curves when incremental heat rate curves are non-monotonically increasing. IEEE
Trans Power Syst 1(1):1016
4. Chiang C-L (2005) Improved genetic algorithm for power economic dispatch of units with valve-point effects and multiple fuels.
IEEE Trans Power Syst 20(4):16901699
5. Walters DC, Sheble GB (1993) Genetic algorithm solution of economic dispatch with valve point loading. IEEE Trans Power Syst
8:13251332
6. He H, Sykora O, Salagean A, Makinen E (2007) Parallelisation
of genetic algorithms for the 2-page crossing number problem.
J Parallel Distrib Comput 67(2):229241
7. Park JH, Yang SO, Lee HS, Park YM (1996) Economic load dispatch using evolutionary algorithms. In: Proceedings of the international conference on intelligent systems applications to power
systems, pp 441445 (1996)
8. Abido MA (2000) Robust design of multi-machine power system
stabilizers using simulated annealing. IEEE Trans Energy Convers
15(3):297304
9. Kennedy J, Eberhart R (1995) Particle swarm optimization. IEEE
Int Conf Neural Netw, pp 19421948
10. Boeringer DW, Werner DH (2004) Particle swarm optimization
versus genetic algorithms for phased array synthesis. IEEE Trans
Antennas Propag. 52:771779
11. Lu H, Sriyanyong P, Song YH, Dillon T (2010) Experimental
study of a new hybrid PSO with mutation for economic dispatch
with non-smooth cost function. Int J Electr Power Energy Syst
32(9):921935
12. Abido MA (2009) Multiobjective particle swarm optimization for
environmental/economic dispatch problem. Electr Power Syst Res
79(7):11051113
13. Swain AK, Morris AS (2000) A novel hybrid evolutionary programming method for function optimization. In: Proceedings of
the 2000 congress on evolutionary computation, vol 1, pp 699705
14. Cai J, Ma X, Li L, Yang Y, Peng H, Wang X (2007) Chaotic ant
swarm optimization to economic dispatch. Electr Power Syst Res
77:13731380
15. Ghoshal SP, Chatterjee A, Mukherjee V (2009) Bio-inspired fuzzy
logic based tuning of Power system stabilizer. Expert System with
Applications 36(5):92819292

123

184
16. Roy R, Ghoshal SP (2008) A novel crazy swarm optimized economic load dispatch for various types of cost functions. Int J Electr
Power Energy Syst 30(4):242253
17. Simon D (2008) Biogeography-based optimization. IEEE Trans
Evol Comput 12(6):702713
18. Mandal K, Chakraborty N (2008) Effect of control parameters
on differential evolution based combined Economic emission
Dispatch with valve point loading and transmission loss. Int J
Emerging Electric Power Syst 9(4), Art. 5
19. Dhillon JS, Parti SC, Khotari DP (1993) Stochastic economic load
dispatch. Electric Power Syst Res 26:179186
20. Song YH, Wang GS, Wang PY, Johns AT (1997) Environmental/economic dispatch using fuzzy logic controller genetic algorithms. IEE Proc Gen Transm Distrib 144(4):377382
21. Rughooputh Harry CS, King Robert TFAh (2003) Environmental/
economic dispatch of thermal units using an elitist multiobjectiveevolutionary algorithm. ICIT Maribor, Slovenia, IEEE
conference, pp 4853

123

Electr Eng (2010) 92:173184


22. Kulkarni PS, Kothari AG, Kothari DP (2000) Combined economic
and emission dispatch using inproved back-propagation neural network. Electric Power Components Syst 28:3144
23. MacArthur R, Wilson E (1997) The theory of biogeography.
Princeton University Press, Princeton, NJ
24. Roa-sepulveda CA, Salazar-Nova ER, Graciacaroca E, Knight
UG and Coonick A (1996) Environmental economic dispatch via
Hopfield neural network and taboo search. UPEC96 Universities Power Engineering Conference, Crete, Greece, pp 10011004
(1996)
25. Gaing Z-L (2003) Particle swarm optimization to solving the economic dispatch considering the generator constraints. IEEE Trans
Power Syst 18(3):11871195

S-ar putea să vă placă și