Electrical Power and Energy Systems 80 (2016) 325–332

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Kinetic gas molecule optimization for nonconvex economic dispatch

problem
M. Basu ⇑
Department of Power Engineering, Jadavpur University, Kolkata 700098, India

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents kinetic gas molecule optimization (KGMO) algorithm to solve economic dispatch
Received 3 November 2014 problems with non-smooth/non-convex cost functions. KGMO is based on kinetic energy and the natural
Received in revised form 28 November 2015 motion of gas molecules. The effectiveness of the proposed method has been verified on four different
Accepted 4 February 2016
non-convex economic dispatch problems with valve-point effects, prohibited operating zones with
transmission losses, multiple fuels with valve point effects and the large-scale Korean power system with
valve-point effects and prohibited operating zones. The results of the proposed approach are compared
Keywords:
with those obtained by other evolutionary methods. It is found that the proposed KGMO based approach
Kinetic gas molecule optimization
Economic dispatch
is able to provide better solution.
Prohibited operating zones Ó 2016 Elsevier Ltd. All rights reserved.
Valve-point loading
Multi-fuel option

Introduction The valve-point loading, prohibited operating zones, ramp-rate

limits and other constraints turn the decision space into disjoint
Economic dispatch (ED) is an important optimization task in subsets, transforming the ED problem into a difficult non-
power system operation for allocating generation among the com- smooth, non-convex optimization problem.
mitted generating units in the most economical manner while sat- The calculus-based methods fail to address these types of prob-
isfying various physical constraints. The input–output lems. The dynamic programming (DP) approach [8] imposes no
characteristics or cost functions of a generator are approximated restriction on the nature of the cost curves and can solve ED prob-
by using quadratic or piecewise quadratic functions, under the lems with non-smooth and discontinuous cost curves. However,
assumption that the incremental cost curves of the units are mono- this method suffers from the curse of dimensionality or local
tonically increasing piecewise-linear functions [1]. However, real optimality.
input–output characteristics show higher-order nonlinearities Modern meta-heuristic algorithms are a promising alternative
and discontinuities due to valve-point loading in fossil fuel fired for solution of complex ED problems. Genetic algorithms (GAs)
generating plants [2]. The valve-point loading effect has been mod- [3–6], Hopfield neural network (HNN) [9], simulated annealing
eled in [2,3] as a recurring rectified sinusoidal function, such as the (SA) [10–12], evolutionary programming (EP) [13,14], improved
one shown in Fig. 1. tabu search (ITS) [2], particle swarm optimization (PSO)
The discontinuous prohibited operating zones in the input–out- [1,15–18], evolutionary strategy optimization (ESO) [7], ant colony
put performance curve for a typical thermal unit can be due to optimization (ACO) [19], differential evolution (DE) [20–22], artifi-
vibration in a shaft bearing caused by a steam valve or can be due cial immune system (AIS) [23], bacterial foraging algorithm (BFA)
to faults in the machines themselves or the associated auxiliary [24], biogeography-based optimization (BBO) [25], continuous
equipment, such as boilers and feed pumps [4,5,7]. In practice, quick group search optimizer [26], etc. have been developed so
the shape of the input–output curve in the neighborhood of a pro- far and applied successfully to solve ED problems. Although these
hibited zone is difficult to determine by actual performance testing. methods do not always guarantee global best solutions, they often
In actual operation, the best economy is achieved by avoiding oper- achieve a fast and near global optimal solution.
ation in these areas [4,7]. Cost function that takes into account pro- Very recently, a new metaheuristic optimization concept, based
hibited operating zones, can be represented as in Fig. 2. on kinetic energy of gas molecules, has been proposed by Moein
and Logeswaran [27].
In this study, the (KGMO) is applied to solve four non-smooth/
⇑ Fax: +91 33 23357254.
non-convex ED problems with valve-point effects, prohibited
E-mail address: [email protected]

http://dx.doi.org/10.1016/j.ijepes.2016.02.005
0142-0615/Ó 2016 Elsevier Ltd. All rights reserved.
326 M. Basu / Electrical Power and Energy Systems 80 (2016) 325–332

where F i ðP i Þ is the fuel cost function of ith unit. ai , bi and ci are the
fuel cost coefficients of ith unit. N is the number of committed
D units; P i is the power output of ith unit.
Subject to the following constraints

(i) Power balance constraint:

C E
X
N

B Pi  PD  PL ¼ 0 ð2Þ
i¼1

The transmission loss P L may be expressed by using

B-coefficients as
X
N X
N X
N
PL ¼ Pi Bij Pj þ B0i P i þ B00 ð3Þ
i¼1 j¼1 i¼1
A
where P D is the system load demand. Bij , B0i and B00 are
B-coefficients.

(ii) Generation capacity constraints

The power generated by each unit should be within its lower
Fig. 1. Example of valve-point cost function with 5 valves. A – Primary Valve. limit P min
i and upper limit Pmax
i , so that
B – Secondary Valve. C – Tertiary Valve. D – Quaternary Valve. E – Quandary Valve.
Pmin
i 6 Pi 6 Pmax
i i2N ð4Þ
(iii) Prohibited operating zone
The feasible operating zones of a unit with prohibited oper-
F Prohibited operating zone ating zones can be described as follows:
u
Pmin 6 Pi 6 Pli;1
e i

l Pui;j1 6 Pi 6 Pli;j ; j ¼ 2; 3; . . . ; ni ð5Þ

Pui;ni 6 Pi 6 Pmax
i ; i2N
c
o where j represents the number of prohibited operating zones of i
s the unit. Pui;j1 is the upper limit of ðj  1Þth prohibited operating
t zone of i the unit. Pli;j is the lower limit of jth prohibited operating
zone of i the unit. Total number of prohibited operating zone of i
Power output
the unit is ni .
Fig. 2. Input–output curve with prohibited operating zones.
Economic dispatch problem considering valve-point effects and
transmission losses
operating zones with transmission losses, multiple fuels with valve
point effects and the large-scale Korean power system with valve- The ED problem can be described as a minimization process
point effects and prohibited operating zones. The performance of with the objective:
the proposed method has been compared with other evolutionary
methods reported in the literature. It is found that the proposed X
N X
N
Min F i ðPi Þ ¼ ai þ bi P i þ ci P2i þ jdi  sinfei  ðP min
i  Pi Þgj ð6Þ
KGMO based approach provides better solution. i¼1 i¼1

where di and ei are the fuel cost coefficients of ith unit with valve-
Problem formulation
point effects.
The above objective function is to be minimized subject to con-
The objective of the ED is to minimize the total generation cost
straints as mentioned in (2) and (4).
of a power system over some appropriate period while satisfying
various constraints. The practical non-smooth/non-convex ED
Economic dispatch problem considering valve-point effects and
problem considers generator nonlinearities such as valve-point
multiple fuels
loading effects, prohibited operating zones and multi-fuel options
along with system power demand, transmission loss and opera-
Since generators are practically supplied with multi-fuel
tional limit constraints.
sources [6], each generator should be represented with several
piecewise quadratic functions superimposed sine terms reflecting
Economic dispatch problem considering prohibited operating zones
the effect valve-point effect of fuel type changes and the generator
and transmission losses
must identify the most economical fuel to burn. The fuel cost func-
tion of the ith generator with N F fuel types is expressed as
The ED problem can be described as a minimization process
with the objective: F i ðPi Þ ¼ aij þ bij Pi þ cij P2i þ jgij  sinfdij  ðPmin
ij  Pi Þgj ð7Þ
X
N X
N
Min F i ðPi Þ ¼ ai þ bi Pi þ ci P2i ð1Þ if Pmin 6 P i 6 Pmax for fuel type j and j ¼ 1; 2; . . . ; NF
ij ij
i¼1 i¼1
 M. Basu / Electrical Power and Energy Systems 80 (2016) 325–332 327

where P min and Pmax are the minimum and maximum power gener- where P is the pressure exerted by the gas, V is the volume of the
ij ij
ation limits of the ith unit with fuel type j respectively. aij ; bij ; cij ; gij container, and N is the number of molecules in the gas.
The kinetic energy equation is stated as follows:
and dij are the fuel-cost coefficients of unit i for fuel j.
The problem can be described as minimization process with the Dk ¼ w ¼ F Ds ¼ maDs ð13Þ
objective:
where Dk is the difference of the gas molecule’s kinetic energy
X
N
Min F i ðPi Þ ð8Þ between the old and new positions, w is the work energy expended,
i¼1 F is the Newton force applied, a is the acceleration of the gas mole-
cule, Ds is the difference in the positions of the gas molecule in a
The above objective function is to be minimized subject to con-
unit time interval, and m is the mass of the gas molecule.
straints as mentioned in (2) and (4).
From the kinematics equations, it can be stated as:
Calculation of slack generator v 2 ¼ v 20 þ 2aDs ð14Þ

N committed generating units deliver their power output sub- where v is the velocity of the gas molecule in the new position, v 0 is
ject to the power balance constraint (2) and the respective capacity the velocity of the gas molecule in the old position.
constraints (4). Assuming the power loading of first (N  1) gener- Combining Eqs. (13) and (14), it is acquired that
ators are known, the power level of the Nth generator (i.e. the slack  2 
v  v 20
generator) is given by Dk ¼ m ð15Þ
2
X
N 1
PN ¼ PD þ PL  Pi ð9Þ and expanding Eq. (15), it is acquired that
i¼1
1 1
The transmission loss PL is a function of all generator outputs Dk ¼ mv 2  mv 20 ð16Þ
2 2
including the slack generator and it is given by
! As the kinetic energy of an object at rest is zero, an object’s
X
N1 X
N1 X
N1 X
N1
kinetic energy is expressed by
PL ¼ Pi Bij Pj þ 2PN BNi Pi þ BNN P 2N þ B0i P i
i¼1 j¼1 i¼1 i¼1 1
k¼ mv 2 ð17Þ
þ B0N PN þ B00 ð10Þ 2
Expanding and rearranging, Eq. (9) becomes A closed vertical cylinder, within which the pressured air sup-
! ports the weight of the piston on top [29], as shown in Fig. 3, is
X
N1
considered to demonstrate the theory behind the KGMO algorithm.
BNN P2N þ 2 BNi Pi þ B0N  1 PN
Based on the law of ideal gases, with constant pressure as the
i¼1
! temperature decreases, the gas molecules converge together in
X
N1 X
N1 X
N1 X
N1
þ PD þ Pi Bij P j þ B0i Pi  Pi þ B00 ¼0 ð11Þ the part of the container that has a lower temperature because
i¼1 j¼1 i¼1 i¼1 the lower temperature will cause less motion. This law is portrayed
in Fig. 3, where the lower left part of the container has the lowest
The loading of the slack generating unit (i.e. Nth) can then be temperature, and the gas molecules converge there over several
found by solving Eq. (11) using standard algebraic method. time intervals.

Kinetic theory of gas molecules

Kinetic gas molecule optimization
The basic concepts of gas molecule laws form the fundamental
principle of the suggested kinetic gas molecule optimization This section describes development of kinetic gas molecule
(KGMO) algorithm [27]. Boyle, Charles and Gay-Lussac developed optimization. In kinetic gas molecule optimization (KGMO) [27]
gas laws based on empirical observations to express the macro- as proposed by Moein and Logeswaran, the agents are gas mole-
scopic behavior of gas molecules [28]. The atomic theory of gases
states that each substance is composed of a large number of mole-
cules or atoms. Basically, all of the properties of the gases, includ-
ing the pressure, volume and temperature, are the effect of the Piston with weight
action of the molecules that compose the gas [29]. There are five
postulates which describe the behavior of molecules in a gas. The
kinetic molecular theory of ideal gases is stated as follows [28]:
Lower temperature
1. A gas comprises a collection of molecules that travel in straight- m
line motion based on Newton’s Laws. m v1 m
2. The molecules in a gas are points and they occupy no volume. v2
3. The collisions between molecules are completely elastic and no m
energy is gained or lost during collision.
v3
m
4. There are no attractive or repulsive forces between the
v4
molecules. v
5. The average kinetic energy of a molecule is 3kT=2, where T is
the absolute temperature, and k is the Boltzmann constant.
Lower temperature
The ideal gas law is stated as follows [28]:
Fig. 3. Under constant pressure, kinetic energy of gas molecules decreases by
PV ¼ NkT ð12Þ decreasing the velocity.
328 M. Basu / Electrical Power and Energy Systems 80 (2016) 325–332

Table 2
Initialize positions and velocities of all agents randomly, number Unit generation (MW) and power loss (MW) for test system 1.
of agents in the container Ν Ρ and set iteration count iter = 0
Unit KGMO Unit KGMO
GEN GEN
1 454.9835 9 25.0000
Evaluate objective function of each agent 2 454.9998 10 28.0022
3 130.0000 11 78.1456
4 130.0000 12 80.0000
Yes 5 235.7674 13 25.0000
Is stopping 6 460.0000 14 15.0018
End 7 464.9957 15 15.0023
criteria satisfied?
8 60.0000 Ploss 26.8983

No
4
Update agent’s best position pbest and global best x 10
3.3
position among all agents in the container gbest
3.295

3.29
Update velocity and position of all agents using
equations (21) and (29) respectively 3.285

3.28

Cost ($)
iter = iter + 1 3.275

3.27
Fig. 4. Flowchart of kinetic gas molecule optimization.
3.265

cules moving in the search space and they are subject to the kinetic 3.26
theory of gases, which defines the laws for gas molecule interac- 3.255
tions in the model. The gas molecules move in the container until
they converge in the part of the container which has the lowest 3.25
0 10 20 30 40 50 60 70 80 90 100
temperature and kinetic energy. The gas molecules attract each
Iteration
other based on weak electrical intermolecular Van Der Waal forces,
where the electrical force is the result of positive and negative Fig. 5. Convergence characteristic for test system 1.
charges in the molecules [30]. In the KGMO algorithm, each gas
molecule, i.e. agent has four specifications position, kinetic energy,
velocity and mass. The kinetic energy of each gas molecule deter- Table 3
mines its velocity and position. In this algorithm, the gas molecules Unit generation (MW) for test system 2.
explore the whole search space to attain the point that has the low- Unit KGMO Unit KGMO Unit KGMO Unit KGMO
est temperature. GEN GEN GEN GEN
The position of the ith agent of a system containing N P agents 1 110.9759 11 168.8005 21 523.2837 31 190.0000
i.e. gas molecules is defined by 2 110.8856 12 94.0005 22 523.2826 32 190.0000
3 97.4171 13 214.7599 23 523.2804 33 190.0000
X i ¼ ðx1i ; . . . ; xdi ; . . . ; xni Þ; i 2 NP ð18Þ 4 179.7342 14 394.2794 24 523.2875 34 164.8433
5 87.9716 15 394.2800 25 523.2856 35 200.0000
where xdi represents the position of the ith agent in the dth 6 140.0000 16 304.5197 26 523.2805 36 200.0000
7 259.6016 17 489.2793 27 10.0012 37 110.0000
dimension. 8 284.6052 18 489.2800 28 10.0000 38 110.0000
The velocity of the ith agent is stated as follows: 9 284.6173 19 511.2797 29 10.0001 39 110.0000
10 130.0000 20 511.2797 30 96.6080 40 511.2797
V i ¼ ðv 1i ; . . . ; v di ; . . . ; v ni Þ; i 2 NP ð19Þ

where v di represents the velocity of the ith agent in the dth where N is the number of gas molecules, b is the Boltzmann con-
dimension.
stant, and T di ðtÞ is the temperature of the ith agent in the dth dimen-
The movement of the gas molecules in the cylinder is based on
sion at time t.
the Boltzmann distribution [28], which means that its velocity is
The velocity of the molecule is brought up to date as follows:
proportional to the exponential of the molecules’ kinetic energy.
This kinetic energy is stated as follows: v di ðt þ 1Þ ¼ T di ðtÞwv di ðtÞ þ C 1 randi ðtÞðgbestd  xdi ðtÞÞ
3 d
d
ki ðtÞ ¼
d
NbT i ðtÞ;
1 d n
K i ¼ ðki ; . . . ; ki ; . . . ; ki Þ; i 2 NP ð20Þ þ C 2 randi ðtÞðpbest i ðtÞ  xdi ðtÞÞ ð21Þ
2
T di ðtÞ reduces exponentially with time and is computed as
follows:
Table 1
Comparison of performance for test system 1.
T di ðtÞ ¼ 0:95  T di ðt  1Þ ð22Þ
Techniques Best cost ($) Average cost ($) Worst cost ($) CPU time (s)
1 2 n
KGMO 32548.1736 32548.2163 32548.3755 7.24 The vector pbest i ¼ ðpbest i ; pbest i ; . . . ; pbest i Þ
signifies the
IPSO [14] 32704.4514 32704.4514 32704.4514 16.2 best previous position of the ith gas molecule and gbest ¼
PSO [1] 32858.00 33039.00 33331.00 26.5 1 2 n
ðgbest ; gbest ; . . . ; gbest Þ is the best previous position among all
 M. Basu / Electrical Power and Energy Systems 80 (2016) 325–332 329

ðdv i Þ
d
of the molecules in the container. The velocity and the position of
each particle are initialized by random vectors within the corre- adi ¼ ð24Þ
dt
sponding ranges. Here, ½v min ; v max  is used as the limits of the gas
From Eq. (17), it can be written as
molecules’ velocity. w is the inertia weight that imitates the gas
molecule’s resistance to slow its movement. randi ðtÞ is a uniform 1 d 2
mðdv i Þ
d
dki ¼ ð25Þ
random variable in the interval [0, 1] at time t, which is utilized to 2
endow with a randomized characteristic to the search algorithm. From Eqs. (24) and (25), the acceleration is defined as
C 1 and C 2 are two acceleration constants. qffiffiffiffiffiffiffiffiffi
ffi
d
The mass m of each gas molecule is a random number within 2ðdki Þ
m
the range 0 < m 6 1 in each execution of the algorithm and this adi ¼ ð26Þ
dt
is same for all of the gas molecules in an execution because the
container is assumed to contain only one type of gas at any one In the time interval Dt, Eq. (26) can be amended as
time. qffiffiffiffiffiffiffiffiffiffi
d
2ðDki Þ
The position of the molecule from the equations of motion in m
adi ¼ ð27Þ
physics [31] is stated as follows: Dt
1 d Thus, in a unit time interval, the acceleration is
xdi ðt þ 1Þ ¼ a ðt þ 1Þ  t2 þ v di ðt þ 1Þ  t þ xdi ðtÞ ð23Þ sffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 i d
2ðdki Þ
where adi represents the acceleration of the ith agent in the dth adi ¼ ð28Þ
m
dimension.
From the acceleration equation, it is acquired that From Eqs. (23) and (28), the position of the molecule is com-
puted as follows:
1 d
Table 4 xdi ðt þ 1Þ ¼ a ðt þ 1Þ  Dt 2 þ v di ðt þ 1Þ  Dt þ xdi ðtÞ
Comparison of performance for test system 2. 2 si ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
d
ð29Þ
1 2ðDki Þ
Techniques Best cost ($) Average cost Worst cost CPU time (s)
xdi ðt þ 1Þ ¼ ðt þ 1Þ  Dt 2 þ v di ðt þ 1Þ  Dt þ xdi
($) ($) 2 m
KGMO 121378.71 121385.54 121394.27 8.03
The position is brought up to date for the unit time interval sta-
IPSO [18] 121403.53 121445.32 121525.49 19.3
NPSO-LRS 121664.43 122209.31 122981.59 – ted as follows:
[16] sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
d
BBO [25] 121426.95 121503.33 121688.66 11.74 2ðDki Þ
CQGSO [26] 121412.55 121423.52 121438.68 8.55 xdi ðt þ 1Þ ¼ ðt þ 1Þ þ v di ðt þ 1Þ þ xdi ðtÞ ð30Þ
m

5
x 10
1.3 Table 6
Comparison of performance for test system 3.
1.29
Techniques Best cost ($) Average cost Worst cost CPU time (s)
($) ($)
1.28
KGMO 608.1096 608.1590 608.1833 2.73
1.27 IPSO [18] 623.826 623.827 623.829 3.2
NPSO-LRS 624.127 – – –
Cost ($)

1.26 [16]
CQGSO [26] 623.827 623.834 623.850 8.76
1.25 IGA_MU [6] 624.517 – – 7.25

1.24
650
1.23
645
1.22
640
1.21
0 50 100 150 200 250 300
635
Iteration
Cost ($)

630
Fig. 6. Convergence characteristic for test system 2.
625

Table 5 620
Unit generation (MW) for test system 3.
615
Unit KGMO Unit KGMO
GEN F GEN F 610
1 230.3725 2 6 244.0957 3
605
2 215.1462 1 7 307.0458 1 0 10 20 30 40 50 60 70 80 90 100
3 293.9596 1 8 242.6440 3
Iteration
4 137.9261 1 9 440.0000 3
5 297.0623 1 10 291.7478 1
Fig. 7. Convergence characteristic for test system 3.
330 M. Basu / Electrical Power and Energy Systems 80 (2016) 325–332

Each gas molecule endeavors to amend its position ðxdi Þ by using The unit generation and power loss corresponding to best fuel
d
the distance between the current position and pbesti and the dis- cost among 100 runs of solutions obtained from proposed KGMO
are shown in Table 1. The best, average and worst cost and average
tance between the current position and gbesti .
CPU time among 100 runs of solutions obtained from proposed
Fig. 4 portrays the flowchart of kinetic gas molecule
KGMO are summarized in Table 2. The cost obtained from particle
optimization.
swarm optimization [1] and improved particle swarm optimization
(IPSO) [18] is also shown in Table 2. The convergence characteristic
Application of the proposed method of this test system in case of proposed KGMO is depicted in Fig. 5. It
is seen from Table 2 that the fuel cost found by using KGMO is the
The proposed kinetic gas molecule optimization (KGMO) is lowest among all other methods.
applied to four different power systems (1) 15-unit system with
prohibited operating zones and transmission losses; (2) 40-unit
system with valve-point effects; (3) 10-unit system considering Test system 2
multiple fuels with valve-point effects; and (4) 140-unit Korean
power system with valve-point effects and prohibited operating This system consists of 40 generating units and the input data is
zones. For each case 100 runs are conducted to compare the solu- taken from [14]. The load demand of this system is 10,500 MW.
tion quality. The computational results have been used to compare The problem is solved by using KGMO. Here, the parameters are
the performance of the proposed KGMO approach with that of taken as N P ¼ 100, C 1 ¼ 1 and C 2 ¼ 3, itermax ¼ 300. The inertia fac-
other evolutionary methods. The proposed KGMO algorithm used tor ðwÞ is decreased linearly from 0.85 to 0.2 and T is decreased
in this paper is implemented by using MATLAB 7.0 on a PC exponentially from 0.95 to 0.1.
(Pentium-IV, 80 GB, 3.0 GHz). The unit generation corresponding to best fuel cost among 100
runs of solutions obtained from proposed KGMO is shown in
Test System 1 Table 3. The best, average and worst cost and average CPU time
among 100 runs of solutions obtained from proposed KGMO are
This system consists of 15 generating units. Here prohibited summarized in Table 4. The cost obtained from improved particle
operating zones and transmission losses are considered. Units 2, swarm optimization (IPSO) [18], new particle swarm optimization
5 and 6 have three prohibited operating zones and unit 12 has with local random search (NPSO-LRS) [16], continuous quick group
two prohibited operating zones. These prohibited zones result in search optimizer (CQGSO) [26] and biogeography-based optimiza-
four disjoint feasible sub-regions for each of units 2, 5, and 6 and tion (BBO) [25] is also shown in Table 4. The convergence charac-
three disjoint feasible sub-regions for unit 12. Hence, those zones teristic of this test system in case of proposed KGMO is depicted in
result in a non-convex decision space which consists of 192 convex Fig. 6. It is seen from Table 4 that the fuel cost found by using
sub-spaces for this system. The load demand of this system is KGMO is the lowest among all other methods.
2630 MW. The input data and B coefficients for transmission losses
are taken from [1]. Test system 3
The problem is solved by using KGMO. Here, the parameters are
taken as N P ¼ 30, C 1 ¼ 1 and C 2 ¼ 3, itermax ¼ 100. The inertia fac- This system consists of 10 generating units with valve-point
tor ðwÞ is decreased linearly from 0.85 to 0.2 and T is decreased effects and multi-fuel sources. The input data is taken from [6].
exponentially from 0.95 to 0.1. The load demand is 2700 MW.

Table 7
Unit generation (MW) for Test System 4.

Unit KGMO Unit KGMO Unit KGMO Unit KGMO Unit KGMO
GEN GEN GEN GEN GEN
1 119.0000 29 490.1614 57 103.0000 85 115.0000 113 94.0000
2 180.7165 30 491.2084 58 198.0000 86 207.0000 114 94.0000
3 190.0000 31 506.0000 59 255.0900 87 254.8467 115 244.5116
4 159.4037 32 503.7455 60 153.0000 88 226.1198 116 244.0000
5 90.0000 33 497.8549 61 164.2026 89 195.1145 117 255.2947
6 90.0000 34 506.0000 62 95.0000 90 240.6012 118 95.0000
7 490.0000 35 495.6174 63 511.0000 91 175.0000 119 95.0000
8 490.0000 36 499.9293 64 160.0000 92 580.0000 120 116.0000
9 496.0000 37 241.0000 65 480.5710 93 640.7335 121 75.7069
10 495.9340 38 240.7858 66 328.0236 94 980.8292 122 2.1608
11 496.0000 39 771.8933 67 490.0000 95 978.0000 123 4.0000
12 488.5684 40 769.0000 68 196.0000 96 681.9991 124 15.0000
13 506.0000 41 6.3951 69 130.1320 97 718.5009 125 9.0000
14 506.0215 42 6.3384 70 344.3487 98 718.0000 126 12.5153
15 505.9789 43 240.9178 71 167.0911 99 716.0806 127 10.3184
16 502.9080 44 228.6591 72 137.0000 100 957.7208 128 112.3814
17 506.0000 45 217.7606 73 242.9033 101 954.7347 129 7.8492
18 505.8375 46 247.5269 74 176.9010 102 1006.1632 130 5.0000
19 503.1142 47 201.6792 75 324.9975 103 997.7745 131 5.0000
20 505.0000 48 231.9561 76 294.4351 104 1008.0740 132 79.4814
21 505.0000 49 216.4365 77 238.2067 105 1020.0000 133 5.5327
22 505.0000 50 212.6734 78 330.0000 106 953.1784 134 42.0000
23 504.9557 51 165.0000 79 531.0000 107 943.7348 135 42.0000
24 493.6675 52 165.0000 80 516.1927 108 1006.0000 136 41.0000
25 525.6857 53 166.3243 81 200.0000 109 1012.7469 137 19.6460
26 536.5553 54 214.8663 82 132.0000 110 1018.7499 138 9.4835
27 544.7023 55 187.7986 83 115.3559 111 1008.6801 139 7.0000
28 546.4826 56 180.0000 84 132.6763 112 119.5441 140 26.0000
 M. Basu / Electrical Power and Energy Systems 80 (2016) 325–332 331

The problem is solved by using KGMO. Here, the parameters are swarm optimization (IPSO) [18], continuous quick group search
taken as N P ¼ 30, C 1 ¼ 1 and C 2 ¼ 3, itermax ¼ 100. The inertia fac- optimizer (CQGSO) [26] and differential evolution based on trun-
tor ðwÞ is decreased linearly from 0.85 to 0.2 and T is decreased cated Levy-type flights and population diversity measure (DEL)
exponentially from 0.95 to 0.1. [21] is also shown in Table 8. The convergence characteristic of this
The unit generation corresponding to best fuel cost among 100 test system in case of proposed KGMO is shown in Fig. 8. It is seen
runs of solutions obtained from proposed KGMO is shown in from Table 8 that the fuel cost found by using KGMO is the lowest
Table 5. The best, average and worst cost and average CPU time among all other methods.
among 100 runs of solutions obtained from proposed KGMO are
summarized in Table 6. The cost obtained from improved particle
Conclusion
swarm optimization (IPSO) [18], continuous quick group search
optimizer (CQGSO) [26], new particle swarm optimization with
In this paper, kinetic gas molecule optimization has been suc-
local random search (NPSO-LRS) [16] and improved genetic algo-
cessfully implemented to solve four non-smooth/non-convex eco-
rithm with multiplier updating (IGA_MU) [6] is also shown in
nomic dispatch problems considering valve-point effects,
Table 6. The convergence characteristic of this test system in case
prohibited operating zones with transmission losses, multiple fuels
of proposed KGMO is depicted in Fig. 7. It is seen from Table 6 that
with valve-point effects and the large-scale Korean power system
the fuel cost found by using KGMO is the lowest among all other
with valve-point effects and prohibited operating zones. The
methods.
results have been compared with those obtained by other evolu-
tionary algorithms reported in the literature. It is seen from the
Test system 4 comparisons that the proposed kinetic gas molecule optimization
performs better than other evolutionary algorithms in the
This is the Korean power system consisting of 140 generating literature.
units. The input data is taken from [18]. The load demand is
49,342 MW. Here, twelve generators have the cost function with
References
valve-point effects and four generators are considered the prohib-
ited operating zones. [1] Gaing Z-L. Particle swarm optimization to solving the economic dispatch
The problem is solved by using KGMO. Here, the parameters are considering the generator constraints. IEEE Trans Power Syst 2003;18
taken as N P ¼ 200, C 1 ¼ 1 and C 2 ¼ 3, itermax ¼ 400. The inertia fac- (3):1187–95.
[2] Lin Whei-Min, Cheng Fu-Sheng, Tsay Ming-Tong. An improved Tabu search for
tor ðwÞ is decreased linearly from 0.85 to 0.2 and T is decreased economic dispatch with multiple minima. IEEE Trans Power Syst 2002;17
exponentially from 0.95 to 0.1. (1):108–12.
The unit generation corresponding to best fuel cost among 100 [3] Walter DC, Sheble GB. Genetic algorithm solution of economic dispatch with
valve point loading. IEEE Trans Power Syst 1993;8(August):1325–32.
runs of solutions obtained from proposed KGMO is shown in
[4] Cheng PH, Chang HC. Large scale economic dispatch by genetic algorithm. IEEE
Table 7. The best, average and worst cost and average CPU time Trans Power Syst 1995;10(4):1919–26.
among 100 runs of solutions obtained from proposed KGMO are [5] Orero SO, Irving MR. Economic dispatch of generators with prohibited
operating zones: a genetic algorithm approach. IEE Proc Gener Transm
summarized in Table 8. The cost obtained from improved particle
Distrib 1996;143(6):529–33.
[6] Chiang C-L. Improved genetic algorithm for power economic dispatch of units
with valve-point effects and multiple fuels. IEEE Trans Power Syst 2005;20
Table 8 (4):1690–9.
Comparison of performance for Test System 4. [7] Pereira-Neto A, Unsihuary C, Saavedra OR. Efficient evolutionary strategy
optimization procedure to solve the nonconvex economic dispatch problem
Techniques Best cost ($) Average cost ($) Worst cost ($) CPU time (s) with generator constraints. IEE Proc Gener Transm Distrib 2005;152
KGMO 1583944.60 1583952.14 1583963.52 28.14 (5):653–60.
IPSO [18] 1657962.73 1657962.73 1657962.73 150 [8] Liang ZX, Glover JD. A zoom feature for a dynamic programming solution to
economic dispatch including transmission losses. IEEE Trans Power Syst
CQGSO [26] 1657962.72 1657962.74 1657962.77 31.67
1992;7(May):544–9.
DEL [21] 1657962.71 1658001.70 – 57.98
[9] Su CT, Lin CT. New approach with a Hopfield modeling framework to economic
dispatch. IEEE Trans Power Syst 2000;15(2):541.
[10] Wong KP, Fung CC. Simulated annealing based economic dispatch algorithm.
IEE Proc Gen, Trans, Distrib 1993;140(6):509–15.
[11] Wong KP, Wong YW. Genetic and genetic/simulated annealing approaches to
6
x 10 economic dispatch. IEE Proc Gen Trans Distrib 1994;141(5):507–13.
1.85 [12] Wong KP, Wong YW. Thermal generator scheduling using hybrid genetic/
simulated-annealing approach. IEE Proc Gen, Trans, Distrib 1995;142(4):372–
80.
1.8 [13] Yang HT, Yang PC, Huang CL. Evolutionary programming based economic
dispatch for units with nonsmooth fuel cost functions. IEEE Trans Power Syst
1996;11(February):112–8.
1.75 [14] Sinha N, Chakrabarti R, Chattopadhyay PK. Evolutionary programming
techniques for economic load dispatch. IEEE Trans Evol Comput 2003;7
(1):83–94.
Cost ($)

[15] Park JB, Lee KS, Shin JR, Lee KY. A particle swarm optimization for economic
1.7 dispatch with nonsmooth cost function. IEEE Trans Power Syst 2005;20
(1):34–42.
[16] Selvakumar AI, Thanushkodi K. A new particle swarm optimization solution to
1.65 nonconvex economic dispatch problems. IEEE Trans Power Syst 2007;22
(1):42–51.
[17] Chaturvedi KT, Pandit M, Srivastava L. Self-organizing hierarchical particle
1.6 swarm optimization for nonconvex economic dispatch. IEEE Trans Power Syst
2008;23(3):1079.
[18] Park JB, Jeong YW, Shin JR, Lee KY. An improved particle swarm optimization
for nonconvex economic dispatch problems. IEEE Trans Power Syst 2010;25
1.55 (1):156–66.
0 50 100 150 200 250 300 350 400
[19] Hou YH, Wu YW, Lu LJ, Xiong XY. Generalized ant colony optimization for
Iteration economic dispatch of power systems. In: Proceedings of the international
conference on power system technology, Power-Con 2002, October 13–17, vol.
Fig. 8. Convergence characteristic for test system 4. 1; 2002. p. 225–9.
332 M. Basu / Electrical Power and Energy Systems 80 (2016) 325–332

[20] Nomana N, Iba H. Differential evolution for economic load dispatch problems. [25] Bhattacharya A, Chattopadhyay PK. Biogeography-based optimization for
Elect Power Syst Res 2008;78(3):1322–31. different economic load dispatch problems. IEEE Trans Power Syst May
[21] dos Santos Coelho Leandro, Bora TC, Mariani VC. Differential evolution based 2010;25(2):1064–77.
on truncated Levy-type flights and population diversity measure to solve [26] Moradi-Dalvand M, Mohammadi-Ivatloo B, Najafi A, Rabiee A. Continuous
economic load dispatch problems. Electr Power Energy Syst 2014;57:178–88, quick group search optimizer for solving non-convex economic dispatch
2014. problems. Elect Power Syst Res 2012;93:93–105.
[22] Wang SK, Chiou JP, Liu CW. Non-smooth/non-convex economic dispatch by a [27] Moein S, Logeswaran R. KGMO: a swarm optimization algorithm based on the
novel hybrid differential evolution algorithm. IET Gener, Transm Distrib kinetic energy of gas molecules. Inf Sci 2014;275:127–44.
2007;1(5):793–803. [28] Loeb LB. The kinetic theory of gases. NY: Dover Phoenix Editions; 2004.
[23] Panigrahi BK, Yadav SR, Agrawal S, Tiwari MK. A clonal algorithm to solve [29] Hopwood SJ, Jeans J. An introduction to the kinetic theory of
economic load dispatch. Elect Power Syst Res 2007;77(10):1381–9. gases. UK: Cambridge University Press; 2009.
[24] Panigrahi BK, Pandi VR. Bacterial foraging optimization: Nelder–Mead hybrid [30] Parsegian VA. Van der Waals forces: a handbook for biologists, chemists,
algorithm for economic load dispatch. IET Gen Trans Distrib 2008;2 engineers, and physicists. Cambridge University Press; 2006.
(4):556–65. [31] Holliday D, Resnick R. Fundamentals of physics. John Wiley and Sons; 1993.