Electrical Power and Energy Systems 73 (2015) 456–464

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

Teaching learning based optimization for economic load dispatch

problem considering valve point loading effect
Sumit Banerjee a,⇑, Deblina Maity b, Chandan Kumar Chanda c
a
Electrical Engineering Department, Dr. B C Roy Engineering College, Durgapur 713206, India
b
Electrical Engineering Department, Netaji Subhash Engineering College, Kolkata 700152, India
c
Electrical Engineering Department, Indian Institute of Engineering Science & Technology, Shibpur, Howrah 711103, 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 a novel teaching learning based optimization (TLBO) technique to solve economic
Received 21 June 2014 load dispatch (ELD) of the thermal unit without considering transmission losses. The proposed
Received in revised form 10 May 2015 methodology can take care of ELD considering nonlinearity such as valve point loading. The objective
Accepted 12 May 2015
of economic load dispatch is to determine the optimal power generation of the units to meet the load
demand, such that the overall cost of generation is minimized, while satisfying different operational
constraints. TLBO is a recently developed evolutionary algorithm based on two basic concepts of
Keywords:
education namely teaching phase and learning phase. At first, learners improve their knowledge through
Economic load dispatch
Genetic algorithm
the teaching methodology of teacher and finally learners increase their knowledge by interactions among
Particle swarm optimization themselves. The effectiveness of the proposed algorithm has been verified on three different test systems
Teaching learning based optimization with equality and inequality constraints. Compared with the other existing techniques demonstrates the
Valve point loading effect superiority of the proposed algorithm.
Ó 2015 Elsevier Ltd. All rights reserved.

Introduction Pothiya et al. proposed a novel and efficient optimization approach

based on ant colony optimization for solving the economic dis-
Recently the electrical power market becomes more competi- patch problem with non-smooth cost functions [5]. An efficient
tive. In order to survive in this situation, the optimal power gener- chaotic self-adaptive differential harmony search algorithm is pro-
ation is required which minimize the total cost. Economic load posed to solve the complicated dynamic economic dispatch prob-
dispatch determines low cost operation of a power system by lem including valve point effect, ramp rate limits and prohibited
dispatching the power generation resources to supply the load. operating zone [6]. An integrated algorithm based on evolutionary
The main objective of the ELD is to minimize the total cost of programming and simulated annealing is developed for solving
generation while satisfying the operational constraints. ELD problem in [7]. Hota et al. presented a newly developed
In the traditional ELD problem, the cost function for each gener- optimization approach involving a modified bacterial foraging
ator has been presented by a quadratic function and is solved using algorithm to solve economic load dispatch problem [8]. Multiple
mathematical programming based optimization techniques such tabu search algorithm is used to solve the economic dispatch prob-
as lambda iteration method and gradient-based method [1]. Basu lem by taking valve-point effects into consideration [9]. Enhanced
proposed artificial bee colony optimization technique to solve eco- cross-entropy method is also proposed to solve dynamic economic
nomic dispatch problem considering transmission losses, multiple dispatch problem with valve-point effects [10]. In [11], Meng pro-
fuels, etc. [2]. Problems of economic load dispatch including trans- posed quantum-inspired particle swarm optimization to solve the
mission losses are solved using dynamic programming method [3]. ELD problem. The method has stronger search ability and quicker
But there was a problem of ‘curse of dimensionality’ or local opti- convergence speed with the ability to be used as a reliable tool
mality. To overcome this problem several alternative methods are for solving ELD problem. Biogeography-based optimization algo-
developed such as differential evolution, tabu search, and particle rithm is used to solve the ELD problems of thermal plants [12].
swarm optimization. Economic Load Dispatch that includes wind Bhattacharya and Chattopadhyay proposed a hybrid technique
power has been solved using quantum genetic algorithm [4]. combining differential evolution with the biogeography-based
optimization algorithm in [13]. The searching ability of DE is
⇑ Corresponding author. Tel.: +91 9434475618; fax: +91 343 2503424. improved by using BBO algorithm. The ELD problems are solved
E-mail address: [email protected] (S. Banerjee). by using seeker optimization algorithm in [14] which exploits

http://dx.doi.org/10.1016/j.ijepes.2015.05.036
0142-0615/Ó 2015 Elsevier Ltd. All rights reserved.
 S. Banerjee et al. / Electrical Power and Energy Systems 73 (2015) 456–464 457

capability of human searching and understanding. In this algo- to solve dynamic economic dispatch problem [33]. Imperialist
rithm, the search direction is mapped on empirical gradient by competitive algorithm is proposed for solving non-convex dynamic
evaluating the response to the position changes and the step length economic power dispatch problem [36]. Alsumait et al. proposed a
is based on uncertainty reasoning by using a simple fuzzy rule. In hybrid GA–PS–SQP method to solve power system valve-point eco-
[15], Chakraborty et al. presented quantum mechanics inspired nomic load dispatch problems [37]. A new approach and coding
particle swarm optimization which is used to solve the ELD prob- scheme is used for solving economic dispatch problems (ED) in
lem. An enhanced bee swarm optimization method is proposed in power systems through an effortless hybrid method (EHM) [38].
[16] to solve the dynamic economic dispatch problem of thermal This paper presents TLBO algorithm to solve ELD problem with
units considering the valve-point effects, ramp-rate limits, and valve point loading effect of thermal plants without considering
the transmission power losses. To solve both complex and non- transmission losses.
complex economic load dispatch (ELD) problems of thermal plant, Section ‘Economic load dispatch’ describes the economic load
a memetic algorithm, namely, aBBOmDE, is proposed in [17]. To dispatch, Section ‘Teaching learning based optimization technique’
solve the economic load dispatch problem reinforcement learning deals with teaching learning based optimization algorithm, and
approaches is proposed in [18]. An artificial immune system based section ‘Implementation of TLBO algorithm for ELD’ discusses the
on the clonal selection principle is proposed by Basu [19] for solv- implementation of TLBO algorithm to ELD, Section ‘Results and dis-
ing dynamic economic dispatch problem. In [20] clonal selection cussions’, presents the simulation results and performance analysis
based artificial immune system algorithm is used to solve the and section ‘Conclusion’, the conclusion.
dynamic economic dispatch problem for generating units with
valve-point effect. In [21], Nima and Hossain proposed to solve Economic load dispatch
the economic dispatch problem with valve loading effect by a
new modified differential evolution algorithm. The proposed The primary objective of ELD involves the optimization of fuel
MDE algorithm is inspired from genetic algorithm, particle swarm cost. The problem is formulated as discussed below.
optimization and simulated annealing. An advanced parallelized
particle swarm optimization algorithm with modified stochastic
Objective functions
acceleration factors is proposed in [22] to solve large scale eco-
nomic dispatch problems with prohibited operating zones,
The classical economic dispatch problem of finding the optimal
ramp-rate limits and transmission losses. In [23], Kumar et al. pro-
combination of power generation which minimizes the total fuel
posed multi-agent based hybrid particle swarm optimization tech-
cost while satisfying the total required demand can be mathemat-
nique which is applied to solve the economic load dispatch
ically stated as follows:
problem. This algorithm recovers the problem of PSO that is the
tuning of variables, randomness and uniqueness of solution. n 
X 
Equal embedded algorithm has been used to solve the economic F i ðPi Þ ¼ ai P2i þ bi P i þ ci ð1Þ
i¼1
load dispatch problem with quadratic and cubic fuel cost functions
and transmission losses [24]. The problems of dynamic economic where
dispatch are also proposed to be solved by a modified particle Fi(Pi): Total fuel cost ($/h).
swarm optimization, which includes advantages of bacterial forag- ai, bi, ci: Fuel cost coefficients of generator i.
ing and PSO [25]. Vaisakh et al. proposed a heuristic optimization Pi: The generated power of generator i (MW).
methodology, namely, bacterial foraging PSO–DE algorithm which n: Number of generators.
is used to solve the economic load dispatch problems [26]. The
algorithm integrates bacterial foraging optimization algorithm,
Constraints
particle swarm optimization and differential evolution for solving
non-smooth non-convex dynamic economic dispatch problem. In
The optimization problem is bounded by two types of
[27], a heuristic algorithm is presented for solving economic dis-
constraints
patch problems including the valve-point effect, prohibited opera-
tion zones, ramp-rate constraints and transmission losses by
(i) Equality constraints
implementing iteration particle swarm optimization along with
(ii) Inequality constraints
time varying acceleration coefficients method. In [28], a differen-
tial harmony search algorithm is proposed by combining the
mechanisms of both differential evolution and harmony search to Equality constraints
solve ELD problem. A hybrid methodology integrating bee colony System power balance
optimization with sequential quadratic programming is proposed X
n
by Basu [29] for solving dynamic economic dispatch problem of ðP i  PD  PL Þ ¼ 0 ð2Þ
generating units considering valve-point effects. In [30] an opti- i¼1

mization methodology is proposed which is based on hybrid shuf- where

fled differential evolution algorithm which combines the benefits
of shuffled frog leaping algorithm and differential evolution, to PD: Total load (MW),
solve economic dispatch problem considering valve point loading PL: Transmission losses (MW).
effects. In [31] a solution for multi-objective economic dispatch
problem with transmission losses is provided by semi-definite pro- The transmission losses can be represented as
gramming formulation. A new method of solving ELD problem is
presented in [32] by integrating the classical gradient-based opti- X
n X
n
PL ¼ Pi Bij Pj ð3Þ
mization technique and a new enhanced simplified swarm opti-
i¼1 j¼1
mization algorithm. Abbas et al. proposed an efficient real-time
approach based on optimality condition decomposition technique where Bij transmission losses coefficient.
458 S. Banerjee et al. / Electrical Power and Energy Systems 73 (2015) 456–464

Inequality constraints subjects offered to learners and the learners’ result is analogous
Maximum and minimum limits of power generation to the ‘‘fitness’’, as in other population-based optimization tech-
The generation power of each generator should lie between niques. As the teacher is considered the most learned person in
maximum limit and minimum limit. That is the society, the best solution is analogous to teacher in TLBO. The
algorithm of TLBO is divided into two parts. The first part consists
Pmin
i 6 Pi 6 Pmax
i ð4Þ of ‘‘teacher phase’’ and the second part consists of ‘‘learner phase’’.
where P min and P max are the minimum and maximum generation of The ‘‘teacher phase’’ means learning from the teacher and the ‘‘lear-
i i
ner phase’’ means learning through the interaction between learners
power.
in a class. Now, implementation of TLBO is described below.
Economic load dispatch with valve point loading. In ELD with ‘‘valve
Initialization
point loadings’’, objective function F is represented by a more com-
plex formula, given as (5). Variation of fuel cost ‘‘Fi(Pi)’’ due to valve
The population X is randomly initialized by a search space
point loading with the change of generation value is shown in
bounded by matrix of N rows and D columns.
Fig. 1. The objective of economic load dispatch with valve point
Where
loading is to minimize (5) subject to generator operating limit.
!
X
n
N number of learners in a class i.e. ‘‘class size’’.
F ¼ min F i ðP i Þ D number of courses offered to the learners i.e. ‘‘no of designed
i¼1
! variables’’.
X
n  n  o
  MAXIT maximum number of allowable iterations.
¼ min ai P2i þ bi Pi þ ci þ ei  sin f i  Pmin
i  Pi  ð5Þ
i¼1
The population X is randomly initialized which is bounded by
where ai, bi, ci, di, ei are the cost coefficients of unit i. matrix of N rows and D columns. The jth parameter of the ith lear-
ner is assigned values randomly using the equation,
Economic load dispatch with valve point loading for multiple  
fuels. For a power system with n generators and nF fuel options X 0i;j ¼ X min
j þ rand  X max
j  X min
j ð7Þ
for each unit, the cost function of the generator with valve-point
where rand represents a uniformly distributed random variable
loading is expressed as (6), where Pmin
ik and P max
ik are the minimum
and maximum power generation limits of the ith generator with within the range (0, 1), X min
j and X max
j represents the minimum
fuel option k, respectively; aik, bik, cik, dik, eik and fik are the and maximum value for jth parameter.
fuel-cost coefficients of generator i for fuel k.
 n  o Teacher phase
 
F i ðPi Þ ¼ aik P2i þ bik Pi þ cik þ eik  sin f ik  Pmin
ik  P ik  ð6Þ
The mean parameter of each subject of the learners in the class
If Pmin
ik 6 Pi 6 P max
ik for fuel option k; k ¼ 1; 2; . . . nF . at generation g is given as
h i
Teaching learning based optimization technique M g ¼ m1g ; m2g ; . . . ; mjg ; . . . ; mDg ð8Þ

The learner with minimum objective function is represented as

This optimization method is based on the relationship between
the ‘Teacher’ (XTeacher). The teacher tries to enhance the results of
teacher and student in the class. It is influenced by the effect of a
other individuals (Xi) by increasing the mean result of the class-
teacher on the output of learners in a class. It is a population based
room (Mg) toward his/her position XTeacher. To obtain a new set of
method and like other population based methods it uses a
learners a random weighted differential vector is formed from
population of solutions to get the global solution. A group of
the current mean and the desired mean parameters and added to
learners constitute the population in TLBO. In any optimization
the existing population of learners. The equation is
algorithms there are numbers of different design variables. The  g 
different design variables in TLBO are represented as different Xnewig ¼ X ig þ rand X Teacher  TF  M g ð9Þ

TF is the teaching factor. Value of TF can be either 1 or 2. The value of

TF is decided randomly with equal probability as,
T F ¼ round½1 þ randð0; 1Þf2  1g ð10Þ
where TF is not a parameter of the algorithm. Its value is randomly
decided by the algorithm using above Equation. However, the algo-
rithm is found to perform much better if the value of TF is either 1 or
2 and hence to simplify the algorithm, the teaching factor is sug-
gested to take either 1 or 2 depending on the rounding up criteria.
If Xnewig is found to be better than X ig in generation g, than it
replaces on X ig otherwise it remains X ig .

Learner phase

In the learner phase, the learners attempt to increase their

information by interacting with others. Therefore, an individual
learns new knowledge if the other individuals have more knowl-
edge than him/her. The random interaction among the learners
Fig. 1. Input–output curve with valve-point loading. a, b, c, d, e—valve points. improves his or her knowledge. For a learner X ig , randomly select
 S. Banerjee et al. / Electrical Power and Energy Systems 73 (2015) 456–464 459

another learner X rg as i – r. The ith learner of the matrix Xnewig is for solving economic load dispatch using TLBO technique are
modified according the following equation. described below.
   
Xnewig ¼ X ig þ rand  X ig  X rg if f X ig < f ðX rg Þ ð11Þ Step 1
     
Xnewig ¼ X ig þ rand  X rg  X ig if f X ig > f X rg ð12Þ Read the system data which consists of fuel cost curve coeffi-
cients of generators, power generation limits and power demand.
All generators’ generation is to be initialized within power genera-
Algorithm termination tion limits. Set time count t as one and repeat the following steps
for the scheduled iteration.
When the stopping criteria that means when MAXIT iteration is
completed, then the algorithm is stop, otherwise repeat from Step 2
‘Teacher Phase’.
Teacher phase will start. Mean value of all generators is deter-
Flowchart of TLBO algorithm mined. Calculate cost value of all population size. As ELD problem

Initialize number of students (population), termination criterion

Calculate the mean of each design variables

Identify the best solution (teacher)

Teacher Phase
Modify solution based on best solution
Xnew = Xold + r (Xteacher – (TF) Mean)

No
Is new solution Yes
Reject Accept
better than existing?

Select any two solutions randomly Xi and Xj

Yes No
Is Xi better than Xj?

Student Phase
Xnew = Xold + r (Xi - Xj) Xnew = Xold + r (Xj – Xi)

No Is new solution Yes

Reject Accept
better than existing?

Is termination No
criteria satisfied?

Yes
Final value of solution

Implementation of TLBO algorithm for ELD

Table 1
The TLBO is implemented to ELD utilizing four main features. Best power output for three generator system without considering transmission
losses (PD = 585 MW).
Firstly, all units are initialized within the generator limits. Then
teacher phase will start. In teacher phase teacher is selected and Unit power output TLBO Classical PSO (34)
new generator matrix is formed according to the teacher phase P1 (MW) 268.8938 268.89
equation. In learner phase, again new generator matrix is to be pre- P2 (MW) 234.2651 234.266
pared by interchanging the one generator with other. Lastly when P3 (MW) 81.8411 81.8412
Total generation cost ($/h) 5821.4 5821.44
stopping criteria is reached then algorithm is terminated. The steps
460 S. Banerjee et al. / Electrical Power and Energy Systems 73 (2015) 456–464

Table 2
Best power output for three generator system without considering transmission Pseudo code of TLBO
losses (PD = 700 MW). Set k = 1
Unit power output TLBO Classical PSO (34) D = no. of generators
P1 (MW) 322.9408 322.9451
N = number of learners in a class i.e. population size.
P2 (MW) 277.7256 277.7309 X imin ¼ Minimum value of generators.
P3 (MW) 99.3335 99.3354
X imax ¼ Maximum value of generators.
Total generation cost ($/h) 6838.4 6888.4
Generate initial students of the classroom i.e. generation of all
generators randomly.
is to minimize the cost function, fittest population size correspond- Calculate objective function f (X) for whole students of the
ing minimum cost is selected as ‘teacher’. Then new generator classroom i.e. all generators.
matrix is formed according to the following equation. WHILE (the termination conditions are not met)
 g  fTeacher Phaseg
Xnewig ¼ X ig þ rand X Teacher  TF  M g ð13Þ
Calculate the mean of each design variable Mg.
Xnewig = New generator matrix, Identify the best solution (teacher)
X ig = Initialization generator matrix, FOR i ! n
g
X Teacher = Fittest generator (teacher), Calculate teaching factor T iF ¼ round½1 þ randð0; 1Þf2  1g
Mg = Mean value of all generators. Modify solution based on best solution (teacher)
Xnewig ¼ X ig þ randðX Teacher
g
 TF  M gÞ
TF is the teaching factor. Value of TF can be either 1 or 2. The Calculate objective function for new mapped student f ðX inew Þ
value of TF is decided randomly with equal probability as, IF X inew is better than Xi, f ðX inew Þ < f ðX i Þ
T F ¼ round½1 þ randð0; 1Þf2  1g ð14Þ X inew ¼ X i
Check whether X inew is within limits.
If new generator matrix Xnewig is found to be a superior learner
than initialized generator matrix X ig in generation g, than it replaces IF X inew > X imax
inferior learner X ig in the matrix. X inew ¼ X imax
ELSE IF X inew < X imin
Step 3 X inew ¼ X imin :
END
Learner phase will start. Here generation is improved by inter- END IF fEnd of Teacher Phaseg
action with different learners. For a learner X ig (ith population size), fLearner Phaseg
another learner X rg (rth population size) is selected. The ith learner Randomly select another learner Xr, such that i – r
of the matrix Xnewig is modified according the following equation. IF Xi is better than Xr, i.e. f(Xi) < f(Xr)
      X inew ¼ X i þ randð0; 1ÞðX i  X r Þ
Xnewig ¼ X ig þ rand  X ig  X rg if f X ig < f X rg ELSE
      ð15Þ
Xnewig ¼ X ig þ rand  X rg  X ig if f X ig > f X rg X inew ¼ X i þ randð0; 1ÞðX r  X i Þ
END IF
Step 4 IF X inew is better than Xi, f ðX inew Þ < f ðX i Þ
X inew ¼ X i
If the termination criterion is satisfied, the iterative process is Check whether X inew is within limits.
stopped. The termination criteria used in this work is the maxi-
IF X inew > X imax
mum number of iterations. The best fitness and the corresponding
generation retained in the memory at the end of the algorithm is X inew ¼ X imax
stored when termination criteria is satisfied. ELSE IF X inew < X imin
X inew ¼ X imin :
END
Table 3
Best power output for three generator system without considering transmission
END IF fEnd of Learner Phaseg
losses (PD = 800 MW). END FOR
Set k = k + 1
Unit power output TLBO Classical PSO (34)
END WHILE
P1 (MW) 369.9383 369.9355
P2 (MW) 315.5174 315.5187
P3 (MW) 114.5443 114.5438
Total generation cost ($/h) 7738.5 7738.51
Post process results and visualization.

Results and discussions

Table 4
Comparison with different methods for three generator system without considering
transmission losses. The applicability of the TLBO algorithm for practical application
has been tested in three test cases. Case 1 consists of three unit
Sl. No. Load demand Conventional GA method TLBO
systems [34], case 2 consists of thirteen unit system [35] and case
(MW) method ($/h) (34) ($/h) (34) method ($/h)
3 consists of forty unit system [35]. The programs are developed
1. 585 5821.45 5827.5 5821.4
using MATLAB 7.01 and the system configuration is Pentium IV
2. 700 6838.41 6877.2 6838.4
3. 800 7738.51 7756.8 7738.5
processor with 3.2 GHz speed and 1 GB RAM. Computational
results are based on 30 trials.
 S. Banerjee et al. / Electrical Power and Energy Systems 73 (2015) 456–464 461

Table 5 7738.51
Best power output for three generator system with valve point loading without
considering transmission losses (PD = 850 MW).
7738.509
Unit power output TLBO
P1 (MW) 394.5243 7738.508
P2 (MW) 56.2764
P3 (MW) 399.1993
Total generation cost($/h) 8280.9 7738.507

Fuel Cost
7738.506

5821.4393
7738.505

5821.4393 7738.504

5821.4393 7738.503
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Cycle
Fuel Cost

5821.4393 Fig. 4. Convergence characteristic of 3-generator system for 800 MW.

5821.4392

8460
5821.4392

8440

5821.4392
0 500 1000 1500 2000 8420
Cycle
8400
Fig. 2. Convergence characteristic of 3-generator system for 585 MW.
Fuel Cost

8380

8360

8340

8320
6838.435

8300

6838.43 8280
0 100 200 300 400 500 600 700 800 900 1000
Cycle
Fuel Cost

6838.425 Fig. 5. Convergence characteristic of 3-generator system for 850 MW with consid-
ering valve point loading effect.

6838.42

6838.415 Table 6
Best power output for thirteen generator system without considering transmission
losses (PD = 1800 MW).

6838.41 Unit power output TLBO

0 200 400 600 800 1000 1200 1400 1600 1800 2000
P1 (MW) 364.9932
iteration
P2 (MW) 227.9523
P3 (MW) 217.4649
Fig. 3. Convergence characteristic of 3-generator system for 700 MW.
P4 (MW) 95.2258
P5 (MW) 106.6728
P6 (MW) 123.5435
Setting of TLBO parameter P7 (MW) 112.5300
P8 (MW) 144.2271
Similar to other optimization techniques, parameters such as P9 (MW) 126.0757
P10 (MW) 60.2360
population size are to be determined before its implementation.
P11 (MW) 48.4754
It is to be determined that an intermediate value for the population P12 (MW) 91.3640
size gives an increase in efficiency and a higher converged score for P13 (MW) 81.2393
the same number of generations. Following parameters are most fit Total generation cost($/h) 18141.6
for the TLBO algorithm.
462 S. Banerjee et al. / Electrical Power and Energy Systems 73 (2015) 456–464

Table 7 4
x 10
Best power output for thirteen generator system with valve point loading without 1.8018
considering transmission losses (PD = 1800 MW).

Unit power output TLBO NN-EPSO (35) 1.8016

P1 (MW) 448.7988 490.0000

1.8014
P2 (MW) 224.6004 189.0000
P3 (MW) 149.6106 214.0000
P4 (MW) 109.8659 160.0000 1.8012

Fuel Cost
P5 (MW) 109.8664 90.0000
P6 (MW) 109.8891 120.0000 1.801
P7 (MW) 109.8607 103.0000
P8 (MW) 109.8962 88.0000
1.8008
P9 (MW) 109.9019 104.0000
P10 (MW) 77.3953 13.0000
P11 (MW) 77.4043 58.0000 1.8006
P12 (MW) 92.4209 66.0000
P13 (MW) 70.4896 55.0000 1.8004
Total generation cost ($/h) 18,115 18442.59

1.8002
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Cycle
Table 8
Best power output for thirteen generator system with valve point loading without
Fig. 6. Convergence characteristic of 13-generator system for 1800 MW.
considering transmission losses (PD = 2520 MW).

Unit power output TLBO GA [35] SA [35] GA-SA EP-SQP

[35] [35] 4
x 10
P1 (MW) 623.5641 628.32 668.40 628.23 628.3136 1.88
P2 (MW) 299.2522 356.49 359.78 299.22 299.0524
P3 (MW) 299.2019 359.43 358.20 299.17 299.0474
1.87
P4 (MW) 159.7330 159.73 104.28 159.12 159.6399
P5 (MW) 159.7350 109.86 60.36 159.95 159.6560
P6 (MW) 159.7242 159.73 110.64 158.85 158.4831 1.86
P7 (MW) 160.3826 159.63 162.12 157.26 159.6749
P8 (MW) 159.4098 159.73 163.03 159.93 159.7265
1.85
Fuel Cost

P9 (MW) 159.3962 159.73 161.52 159.86 159.6653

P10 (MW) 77.3997 77.31 117.09 110.78 114.0334
P11 (MW) 77.4040 75.00 75.00 75.00 75.0000
1.84
P12 (MW) 92.3988 60.00 60.00 60.00 60.0000
P13 (MW) 92.3985 55.00 119.58 92.62 87.5884
Total generation cost 24,197 24398.23 24970.91 24275.71 24266.44 1.83
($/h)

1.82

Case 1: Three unit systems

1.81
0 200 400 600 800 1000 1200 1400 1600 1800 2000
In this case a three unit system is solved for ELD using the pro- Cycle
posed TLBO algorithm. The generation cost coefficients and power
Fig. 7. Convergence characteristic of 13-generator system for 1800 MW by
generation limits are taken from [34]. Here transmission losses are
considering valve point loading effect.
neglected. The total production cost obtained for the three unit
systems of 585 MW, 700 MW and 800 MW without losses is
5821.4 $/h, 6838.4 $/h and 7738.5 $/h respectively. The details of
between number of iterations vs. cost in $/h for load of 585 MW,
the power dispatch of each unit are given in Tables 1–3. Table 5
700 MW and 800 MW respectively. Fig. 5 shows convergence char-
gives best power output with valve point loading without consid-
acteristic of 3-generator system for 850 MW with considering
ering transmission losses for 850 MW. It can be seen that the
valve point loading effect.
power output of the units in each iteration satisfies the generation.
Further, the sum of power generation of each unit for each iteration
equals the load demand. Table 4 provides comparison of the total Case 2: Thirteen unit system
cost obtained using TLBO algorithm with that of other techniques
for without losses. It is clearly seen that the proposed method In this case a thirteen unit system is solved for ELD using the
shows better result than PSO, GA [34] while satisfying all the con- proposed TLBO algorithm. The generation cost coefficients and
straints considered. Thereby, it is clear that the new proposed algo- power generation limits are taken from [35]. Here transmission
rithm is efficient and cheap (in terms of generating cost) than the losses are neglected. Power generation limits and valve point load-
other algorithms. In other words, the proposed algorithm is cap- ing are also included. The corresponding dispatch of units without
able of giving a more optimum solution. Figs. 2–4 shows the graphs considering transmission losses but with valve point loading is

Table 9
Comparison with different methods for thirteen generator system without considering transmission losses with valve point loading effect.

Sl. No. Load demand (MW) GA ($/h) [35] SA ($/h) [35] NN-EPSO ($/h) [35] GA-SA ($/h) [35] EP-SQP ($/h) [35] TLBO ($/h)
1. 1800 – – 18442.59 – – 18,115
2. 2520 24398.23 24970.91 – 24275.71 24266.44 24,197
 S. Banerjee et al. / Electrical Power and Energy Systems 73 (2015) 456–464 463

4 5
x 10 x 10
2.49 1.34

2.48 1.335

2.47 1.33

2.46 1.325
Fuel Cost

Fuel Cost
2.45 1.32

2.44 1.315

1.31
2.43

1.305
2.42

1.3
2.41
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Cycle 1.295
0 100 200 300 400 500 600 700 800 900 1000
Iteration
Fig. 8. Convergence characteristic of 13-generator system for 2520 MW by
considering valve point loading effect.
Fig. 9. Convergence characteristic of 40-generator system for 10,500 MW by
considering valve point loading effect.

Table 10 best power output with valve point loading without considering
Best power output for forty generator system with valve point loading without transmission losses for 1800 MW. It can be seen from Table 7 that
considering transmission losses (PD = 10500 MW). fuel cost obtained from TLBO is less which is compared with
Unit power output TLBO NN-EPSO (35) NN-EPSO [35]. Table 8 gives best power output for 2250 MW and
P1 (MW) 36.1161 114.0000
it is compared with GA-SA [35], EP-SQP [35]. Table 9 provides com-
P2 (MW) 37.9455 114.0000 parison of other different methods without considering transmis-
P3 (MW) 61.8403 120.0000 sion losses but with valve point loading effect. It can be noticed
P4 (MW) 93.4369 190.0000 that the total cost is very much less in case of TLBO. Likewise the
P5 (MW) 83.3052 97.0000
computation time is also less. Fig. 6 shows the graph between
P6 (MW) 120.2602 140.0000
P7 (MW) 290.4140 300.0000 no. of iterations and cost in $/h for load of 1800 MW without con-
P8 (MW) 200.0000 300.0000 sidering transmission losses but with valve point loading effect.
P9 (MW) 293.7905 300.0000 Figs. 7 and 8 shows convergence characteristic for 1800 and
P10 (MW) 210.5287 300.0000
2520 MW respectively with considering valve point loading effect.
P11 (MW) 337.4764 375.0000
P12 (MW) 249.7551 375.0000
P13 (MW) 380.7705 500.0000 Case 3: Forty unit system
P14 (MW) 125.2402 500.0000
P15 (MW) 487.4984 500.0000 In this case a forty unit system is solved for ELD using the pro-
P16 (MW) 500.0000 500.0000
posed TLBO algorithm. The generation cost coefficients and power
P17 (MW) 319.7599 402.6000
P18 (MW) 237.2392 225.0000 generation limits are taken from [35]. In this case transmission
P19 (MW) 516.5296 508.0000 losses are not included and only power generation limits are
P20 (MW) 524.5736 458.0000 included. Valve point loading effect is also included. The corre-
P21 (MW) 540.1990 356.0000 sponding dispatch of units is shown in Table 10. The power output
P22 (MW) 549.3921 394.0000
P23 (MW) 550.0000 355.0000
for each unit satisfies the generation limit constraints. It can be
P24 (MW) 522.9545 525.0000 noticed that the total cost is very much less in case of TLBO.
P25 (MW) 532.1005 310.0000 Likewise the computation time is also less. It can be seen from
P26 (MW) 542.7990 448.0000 Table 10 that fuel cost obtained from TLBO is less compared with
P27 (MW) 56.7790 72.0000
NN-EPSO [35]. Fig. 9 shows convergence characteristic for
P28 (MW) 23.8696 131.0000
P29 (MW) 12.7165 75.0000 10,500 MW by considering valve point loading effect without con-
P30 (MW) 86.0264 67.0000 sidering transmission losses.
P31 (MW) 190.0000 151.0000
P32 (MW) 190.0000 112.0000
P33 (MW) 190.0000 139.0000
Conclusion
P34 (MW) 192.4549 90.0000
P35 (MW) 189.1622 129.0000 The teaching learning based optimization method has been suc-
P36 (MW) 195.0759 104.0000 cessfully implemented to solve economic load dispatch by consid-
P37 (MW) 109.6457 36.0000
ering valve point loading effect. Here transmission losses and other
P38 (MW) 110.0000 89.0000
P39 (MW) 109.3120 104.0000 constraints are not included. The proposed method is efficiently
P40 (MW) 501.2304 550.0000 and effectively applied on three different test systems. The com-
Total generation cost ($/h) 12,996 130328.325 parison of the results with other existing methods reported in
the literature shows the superiority of the proposed method.
Here, TLBO is able to explore the solution space for obtaining the
shown in Table 6. The power output for each unit satisfies the gen- global optimum solution. So, it can be concluded that TLBO tech-
eration limit constraints. Further, the sum of power generation of nique is a promising method for solving ELD in power system
each unit for each iteration equals the load demand. Table 7 shows operation.
464 S. Banerjee et al. / Electrical Power and Energy Systems 73 (2015) 456–464

References [20] Hemamalini S, Simon Sishaj P. Dynamic economic dispatch using artificial
immune system for units with valve-point effect. Electr Power Energy Syst
2011;33:868–74.
[1] Saadat Hadi. Power system analysis. New Delhi: Tata McGraw Hill Publishing
[21] Nima Amjady, Hossein Sharifzadeh. Solution of non-convex economic dispatch
Company; 2002.
problem considering valve loading effect by a new modified differential
[2] Basu M. Artificial bee colony optimization for multi-area economic dispatch.
evolution algorithm. Electr Power Energy Syst 2010;32:893–903.
Electr Power Energy Syst 2013;49:181–7.
[22] Subbaraj P, Rengaraj R, Salivahanan S, Senthilkumar TR. Parallel particle
[3] Liang ZX, Glover JD. A zoom feature for a dynamic programming solution to
swarm optimization with modified stochastic acceleration factors for solving
economic dispatch including transmission losses. IEEE Trans Power Syst
large scale economic dispatch problem. Electri Power Energy Syst 2010;32:
1992;7(2):544–50.
1014–23.
[4] Lee Jia-Chu, Lin Whei-Min, Liao Gwo-Ching, Tsao Ta-Peng. Quantum genetic
[23] Rajesh Kumar, Devendra Sharma, Abhinav Sadu. A hybrid multi-agent based
algorithm for dynamic economic dispatch with valve-point effects and
particle swarm optimization algorithm for economic power dispatch. Electr
including wind power system. Electr Power Energy Syst 2011;33:189–97.
Power Energy Syst 2011;33:115–23.
[5] Saravuth Pothiya, Issarachai Ngamroo, Waree Kongprawechnon. Ant colony
[24] Chandram K, Subrahmanyam N, Sydulu M. Equal embedded algorithm for
optimisation for economic dispatch problem with non-smooth cost functions.
economic load dispatch problem with transmission losses. Electr Power
Electr Power Energy Syst 2010;32:478–87.
Energy Syst 2011;33:500–7.
[6] Arul R, Ravi G, Velusami S. Chaotic self-adaptive differential harmony search
[25] Ahmed Yousuf Saber. Economic dispatch using particle swarm optimization
algorithm based dynamic economic dispatch. Electr Power Energy Syst 2013;
with bacterial foraging effect. Electr Power Energy Syst 2012;34:38–46.
50:85–96.
[26] Vaisakh K, Praveena P, Rao S, Mohana Rama, Kala Meah. Solving dynamic
[7] Christober Asir Rajan C. A solution to the economic dispatch using EP based SA
economic dispatch problem with security constraints using bacterial foraging
algorithm on large scale power system. Electr Power Energy Syst 2010;32:
PSO–DE algorithm. Electr Power Energy Syst 2012;39:56–67.
583–91.
[27] Mohammadi-Ivatloo B, Rabiee A, Soroudi A, Ehsan M. Iteration PSO with time
[8] Hota PK, Barisal AK, Chakrabarti R. Economic emission load dispatch through
varying acceleration coefficients for solving non-convex economic dispatch
fuzzy based bacterial foraging algorithm. Electr Power Energy Syst
problems. Electr Power Energy Syst 2012;42:508–16.
2010;32:794–803.
[28] Ling Wang, Ling-po Li. An effective differential harmony search algorithm for
[9] Worawat Sa-ngiamvibool, Saravuth Pothiya, Issarachai Ngamroo. Multiple
the solving non-convex economic load dispatch problems. Electr Power Energy
tabu search algorithm for economic dispatch problem considering valve-point
Syst 2013;44:832–43.
effects. Electr Power Energy Syst 2011;33:846–54.
[29] Basu M. Hybridization of bee colony optimization and sequential quadratic
[10] Immanuel Selvakumar A. Enhanced cross-entropy method for dynamic
programming for dynamic economic dispatch. Electr Power Energy Syst
economic dispatch with valve-point effects. Electr Power Energy Syst 2011;
2013;44:591–6.
33:783–90.
[30] Reddy Srinivasa A, Vaisakh K. Shuffled differential evolution for economic
[11] Meng Ke, Wang Hong Gang, Dong ZhaoYang, Wong Kit Po. Quantum-inspired
dispatch with valve point loading effects. Electr Power Energy Syst
particle swarm optimization for valve-point economic load dispatch. IEEE
2013;46:342–52.
Trans Power Syst 2010;25(1).
[31] Jubril Abimbola M, Komolafe Olusola A, Alawode Kehinde O. Solving multi-
[12] Aniruddha Bhattacharya, Kumar Chattopadhyay Pranab. Biogeography-based
objective economic dispatch problem via semidefinite programming. IEEE
optimization for different economic load dispatch problems. IEEE Trans Power
Trans Power Syst 2013;28(3).
Syst 2010;25(2).
[32] Azizipanah-Abarghooee Rasoul, Niknam Taher, Gharibzadeh Masihallah,
[13] Aniruddha Bhattacharya, Pranab Kumar Chattopadhyay. Hybrid differential
Golestaneh Faranak. Robust, fast and optimal solution of practical economic
evolution with biogeography-based optimization for solution of economic
dispatch by a new enhanced gradient-based simplified swarm optimization
load dispatch. IEEE Trans Power Syst 2010;25(4).
algorithm. IET Gener, Trans Distrib 2013;7(January):620–35.
[14] Shaw B, Mukherjee V, Ghoshal SP. Seeker optimization algorithm: application
[33] Abbas Rabiee, Behnam Mohammadi-Ivatloo, Mohammad Moradi-Dalvand.
to the solution of economic load dispatch problems. IET Gener Transm Distrib
Fast dynamic economic power dispatch problems solution via optimality
2011;5(1):81–91.
condition decomposition. IEEE Trans Power Syst 2014;29(2).
[15] Chakraborty S, Senjyu T, Yona A, Saber AY, Funabashi T. Solving economic load
[34] Shubham Tiwari, Ankit kumar, Chaurasia GS. Economic load dispatch using
dispatch problem with valve-point effects using a hybrid quantum mechanics
particle swarm optimization. Int J Appl Innov Eng Manage 2013;2(4).
inspired particle swarm optimization. IET Gener Transm Distrib 2011;
[35] Pandian S, Vijaya Muthu, Thanushkodi K. An evolutionary programming based
5(10):1042–52.
efficient particle swarm optimization for economic dispatch problem with
[16] Taher Niknam, Faranak Golestaneh. Enhanced bee swarm optimization
valve-point loading. Eur J Sci Res 2011;52(3):385–97.
algorithm for dynamic economic dispatch. IEEE Syst J 2013;7(4).
[36] Behnam Mohammadi-ivatloo, Abbas Rabiee, Alireza Soroudi, Mehdi Ehsan.
[17] Lohokare MR, Panigrahi BK, Pattnaik SS, Devi S, Mohapatra A. Neighborhood
Imperialist competitive algorithm for solving non-convex dynamic economic
search-driven accelerated biogeography-based optimization for optimal load
power dispatch. Energy 2012;44:228–40.
dispatch. IEEE Trans Syst, Man, Cybern—Part C: Appl Rev 2012;42(5).
[37] Alsumait JS, Sykulski JK, Al-Othman AK. A hybrid GA–PS–SQP method to solve
[18] Jasmin EA, Imthias Ahamed TP, Jagathy Raj VP. Reinforcement learning
power system valve-point economic dispatch problems. Appl Energy 2010;87:
approaches to economic dispatch problem. Electr Power Energy Syst 2011;33:
1773–81.
836–45.
[38] Pourakbari-Kasmaei M, Rashidi-Nejad M. An effortless hybrid method to solve
[19] Basu M. Artificial immune system for dynamic economic dispatch. Electr
economic load dispatch problem in power systems. Energy Convers Manage
Power Energy Syst 2011;33:131–6.
2011;52:2854–60.