3rd International Conference on Advances in Electrical, Electronics, Information, Communication and Bio-Informatics (AEEICB17)

Application and performance comparison of variants

of the firefly algorithm to the economic load dispatch
problem
Fatma Sayed Moustafa Ahmed El-Rafei
Department of Mathematics Department of Engineering Physics and Mathematics
Faculty of Basic Sciences, German University in Cairo Faculty of Engineering, Ain Shams University
Egypt Cairo, Egypt
[email protected] [email protected]

N.M. Badra Almoataz Y. Abdelaziz

Department of Engineering Physics and Mathematics Department of Electrical Power and Machines
Faculty of Engineering, Ain Shams University Faculty of Engineering, Ain Shams University
Cairo, Egypt Cairo, Egypt
[email protected] [email protected]

Abstract—The Economic Load Dispatch problem is an computations to minimize objective function while satisfying
optimization problem which minimizes cost such that the load constraints. This task has proven to be costly and time
demand is met and the generating equality and inequality consuming. Hence, the proper choice of an optimization
constraints are satisfied. Previously, conventional techniques like technique is essential [1].
linear programming and lambda iteration were applied to solve
the economic dispatch problem given their simplicity.
To solve the ELD problem, many mathematical based
Nevertheless, they do not always converge to global optimum optimization techniques were used. The utilized deterministic
which gave rise to metaheuristic techniques such as evolutionary conventional methods include, but not limited to: Lambda
and bio-inspired swarm algorithms. Firefly algorithm is a swarm Iteration, Newton method, Linear Programming (LP), and
based recent metaheuristic that has a high convergence rate and Quadratic Programming (QP) [2-5]. These approaches are
short execution time compared to other metaheuristic techniques easy to use, have a simple mathematical model, and rapid
when solving the economic load dispatch problem. Given that the search. However, they are local optimizers and cannot
firefly algorithm has its shortcoming of getting trapped in local guarantee global optimum. Modern intelligent techniques have
optima, many researchers have proposed modifications and been proposed to overcome the deficiencies of these
hybrids that improve the performance of firefly algorithm to
obtain optimum global solutions rapidly and efficiently. In this
conventional methods. These intelligent techniques are general
paper, three of these recent enhancements were adopted to solve purpose, flexible, and depend on randomness and diversity.
the economic dispatch problem of six generating units. The They search the solution space more meticulously and reach
performance of these variants was compared and analyzed. The global optimal solutions, thus, overcoming the limitations of
results show high efficiency in achieving optimal results in less conventional methods. Artificial Intelligence techniques such
time than the original firefly algorithm. as the Differential Evolution (DE), Genetic Algorithm (GA),
Simulated Annealing (SA), Particle Swarm Optimization
Keywords— Economic load dispatch, Firefly algorithm, (PSO), Ant Colony optimization (ACO), and Firefly
Memetic firefly, Modified firefly, and Variable step size firefly
Algorithm (FFA) were used to solve the ELD problem [6-9].

I. Introduction FFA is a novel nature inspired promising metaheuristic

The main objective of electric power generating systems is method proposed by Yang [10]. FFA is a simple, robust, and
to provide customers with reliable, economic, and high quality evolving intelligent technique. Many researchers applied it to
power supply. Due to the high competition in the industry of solve ELD problems showing the FFA advantages of
generating and supplying electricity and the large increase in obtaining accurate optimal solutions with higher convergence
power demand and fuel cost, the Economic Load Dispatch rate and less computational time compared to other techniques
(ELD) plays a critical role in the efficient and economical like GA, PSO, SA, DE, and others [11-12]. Although FFA has
operation of scheduled generating units. The ELD basically shown its supremacy, it can get trapped in local optima, it
finds the optimum generation scheduling by minimizing fuel cannot memorize earlier obtained better solutions, and its
cost while considering the operational equality and inequality parameters are decided initially and remained unchanged
constraints and matching the load demand of the generating during the iterations. To overcome such shortcomings, some
units. ELD is a highly non-convex nonlinear constrained researchers have suggested modifications and enhancements
optimization problem that requires intensive mathematical to the original firefly algorithm creating new variants of the

978-1-5090-5434-3©2017 IEEE
 3rd International Conference on Advances in Electrical, Electronics, Information, Communication and Bio-Informatics (AEEICB17)
firefly algorithm [13]. Modified versions of the FFA were Pi= Power output from generator (i)
implemented to solve the ELD problem and have shown their Pimin = Minimum permitted power output by generator (i)
efficiency [14-16]. Variants of firefly algorithm such as Pimax = Maximum permitted power output by generator (i)
modified FFA (MFA) was applied on mechanical engineering
design problems, memetic FFA (MFFA) was applied to solve III.Firefly Algorithm
combinatorial problems and variable step size FFA (VSSFA)
was used to solve sixteen benchmark functions and the results Firefly Algorithm is a new population based swarm
demonstrated their optimizing capabilities [17-19]. To the best intelligence metaheuristic method that is developed by Yang
of our knowledge, these methods were not applied to the ELD in 2008 [10]. It is inspired by the bioluminescence behavior of
problem. In this paper, these three variants are applied to solve fireflies at night. The luminosity of fireflies attracts
the nonlinear and non-convex ELD problem. A comparative prospective mates and scare off predators in its neighborhood.
study was carried out on the solution of ELD problem using The three main rules to construct the main algorithm are:
those recent variants and the classical firefly algorithm for 1) Fireflies are attracted to the brighter ones independent of
different test cases. Efficiency was evaluated by comparing their gender
best solutions obtained in terms of execution time, fuel cost 2) The objective function’s values determine the brightness of
and power loss. fireflies
3) Brightness and attractiveness are directly proportional and
II. Problem Formulation
they are inversely proportional to distance, such that:
1
I∝β∝
r
The objective of the nonlinear ELD optimization problem
The factors affecting the algorithm are light intensity,
is to minimize cost while satisfying the load demand and other
attractiveness, distance, and movement and are given by:
operational system equality and inequality constraints [1]. 2
Light intensity I r =I0 e-γr (7)
A. Objective function- cost function Attractiveness β r =β0 e-γr
2
(8)
Ng where I and β are the initial light intensity and initial
Minimize FT =F Pi = ∑i=1 Fi (Pi ) $/hr (1) 0 0
brightness, respectively.
F (P )=a +b P +c P 2 $/hr (2) This distance between firefly i and firefly j is represented as:
i i i i i i i
2
rij = xi -xj = ∑dk=1 xik -xjk (9)
where
Fт: Total Quadratic cost function; it could be also a cubic function The formula that controls the fireflies’ movement is given by:
-γrij 2 t t
Pi: Real power generated xt+1 t
i =xi +β0 e xj -xi +α rand-0.5 (10)
Ng: Number of generation busses where t is the number of current iteration, α∈ 0,1 is the
ai, bi, ci : Fuel cost coefficients for ith unit randomization parameter and γ∈[0,∞) is the coefficient of
absorption. The first term is the current position of the firefly
B. Constraints i, the second term is due to attraction towards a brighter firefly
The objective function must be minimized while considering the j and the last term represents the random walk of the firefly.
following constraints:
A. Variants of Firefly Algorithm
1) Equality constraint- Energy balance equation
Since FFA was developed, it has become a popular
N
∑i=1g Pi =PD +PL (3) optimizer. In recent studies, major improvements and changes
were applied to the FFA to enhance its performance. In this
N N
PL = ∑i=1g ∑j=1g Pi Bij Pj (4) paper, the three variants of FFA described in the following
where subsections are applied to solve the ELD problem and
PD = Load demand comparisons were made with each other and the original FFA.
PL= Power transmission losses B. Modified Firefly Algorithm
Bij= Loss coefficients (constants)
Pi, Pj = Active power injection at the ith and jth generators Three new modifications were suggested in [17]: adding
In some cases, power losses are neglected and the active memory, newborn fireflies and updating formula.
power balance equation becomes:
Ng 1) Adding memory
∑i=1 Pi =PD (5)
This modification is done via 2 approaches. The first approach
2) Inequality constraint- Generating limits is to transfer the best solutions obtained (m₁ high rank
Generated active power should lie between minimum and fireflies) in each iteration to the next iteration. The second
maximum operational values approach is to make a copy of m₂ high rank fireflies in each
Pi min ≤Pi ≤Pi max (6) iteration then replacing the worst solutions obtained (m₂ low
where

978-1-5090-5434-3©2017 IEEE
 3rd International Conference on Advances in Electrical, Electronics, Information, Communication and Bio-Informatics (AEEICB17)
rank fireflies) in each new iteration by the high rank ones
copied from the previous iteration.

2) Newborn fireflies
In each iteration, k of the low rank fireflies are replaced by
newly generated ones using the same random initialization
procedure used to initialize the population of fireflies.

3) Updating formula
The fireflies approached the brighter firefly in a step wise
strategy. Each firefly moves towards each brighter firefly one
at a time. So the firefly ends up moving in an irregular path to
update its location. Fig. 2. Schematic diagram that depicts the representative point P which is the
Fireflies 1 to 6 are sorted according to their objective average position of fireflies brighter than firefly 6. The update of position is
done in one step.
function, 1 being the brightest and 6 having the least
brightness. During the first iteration, firefly 6 moves towards C. Memetic Firefly Algorithm
firefly 1 and updates its position. In the second iteration,
firefly 6 moves towards firefly 2 and updates its position for Memetic algorithm (MA) balances between exploration
the second time. The previous step is repeated for the third, and exploitation of search space. The exploration helps in
fourth and fifth iterations, in each time firefly 6 moves along a discovering new search space (random walk) and the
different path approaching the firefly brighter than itself as exploitation focuses on the vicinity of promising solutions and
shown in Fig. 1. For each firefly j with ( j-1) brighter fireflies, hence reaching optimum solutions more efficiently. Thus, it
the firefly has to make (j-1) steps to update its position. Such combines the benefits of population based and local search
approach is inefficient, time consuming and decreases the techniques. Memetic firefly algorithm (MFFA) enjoys the
performance of the algorithm. advantages of FFA over other techniques and balance of MA.
Therefore, the suggested approach is to get a point to represent MFFA is designed to solve high dimensional problems and
the overall distribution of brighter fireflies. Many multiple strategies are introduced. Parameters α and β are fine-
representations could be possible, but the one implemented is tuned to avoid the inefficient search by high exploration and
the average position of the brighter fireflies. Therefore, the premature convergence by high exploitation [18].
fireflies update their position in one step as shown in Fig. 2. α is adjusted dynamically by:

The representative point Pi is the average of the coordinates of 1

1
t
fireflies brighter than firefly i is given below: α=α * U-L (14)
9000
where α initially is set to 0.2, t is the number of current
1
Pi = ∑i-1
k=1 xk (11) iteration, U is the upper bound of the search space and L is
i-1
the lower bound of the search space.
ri = Pi -xi (12)
and thus changing the updating formula with a scaled random β changes in the range [0.2,1] using this formula:
movement to be: 2
t t β=βmin +(β0 -βmin )e-γrij (15)
xt+1 t
i =xi +β Pi -xi +αε (13)
-γri 2 where βmin =0.2 and β0 =1
with =β0 e , ε= rand-0.5 * U-L
and thus changing the updating formula with a scaled random
where U and L are the upper and lower bounds of the search movement to be:
space, respectively.
xt+1 t t t
i =xi +β xj -xi +α rand-0.5 * U-L (16)
where U and L are the upper and lower bounds of search
space, respectively.
D. Variable Step Size Firefly Algorithm
In the original FFA, the parameters are set fixed. The step size α
controlling randomness is fixed affecting the balance between
diversification and intensification. To reach the optimum solution
effectively, parameter α should decrease dynamically with the
iterations. Since at the beginning of the iteration process, a large α is
needed to allow for the exploration of new search space and to
Fig. 1. Schematic diagram that represents the step wise path that firefly 6 prevent getting stuck in local optima. It is suitable however to have a
move along to update its position approaching the brighter fireflies 1-5. The smaller α as the iterations go on to have less exploration and more
update of position is done in 5 steps.

978-1-5090-5434-3©2017 IEEE
 3rd International Conference on Advances in Electrical, Electronics, Information, Communication and Bio-Informatics (AEEICB17)
exploitation. This facilitates the convergence towards the global 1.40 0.17 0.15 0.19 0.26 0.22
optimum solution. 0.17 0.60 0.13 0.16 0.15 0.20
-4 0.15 0.13 0.65 0.17 0.24 0.19
Variable Step Size Firefly Algorithm (VSSFA) adopts a strategy B=10 ×
0.19 0.16 0.17 0.71 0.30 0.25
for adjusting the step α with the iterations [19]. The formula for the 0.26 0.15 0.24 0.30 0.69 0.32
step size α is a function in the kth iteration and MaxGeneration which 0.22 0.20 0.19 0.25 0.32 0.85
is the total number of iterations. It is given by the following formula:
A. Transmission losses not included
0.4
α k = 0.005* k-MaxGeneration
(17) Tables III to V demonstrate best, mean, and standard deviation for 6
1+exp units lossless system for 600, 700 and 800 MW. Fuel cost and time of
execution were compared in tables VI and VII.
IV. Test and Results
TABLE III. RELIABILITY ANALYSIS FOR POWER DEMAND =600MW
FFA and its variants were implemented to solve ELD FFA MFA VSSFA MFFA
Best 31489 31447 31576 31481
problem with 3 and 6 thermal units on MATLAB® Version
Mean 31842.75 31452.95 31945.7 31620.6
2014a on an Intel® Core™ i7-3770, 3.4GHz, and 8GB RAM SD 243.84008 2.9285348 244.08931 95.848784
32-bit OS computer. However, because of space limitations
and similarity in results, only the six units system is presented. TABLE IV. RELIABILITY ANALYSIS FOR POWER DEMAND =700MW
FFA MFA VSSFA MFFA
The parameters for FFA and its variants are given in Table Best 36075 36006 36036 36021
I. The fuel cost and loss coefficients for six units system are Mean 36353.7 36010.3 36212.2 36114.6
SD 152.7413 2.5152168 75.797931 44.446775
given in table II [20]. In each case study, a reliability analysis
of 20 independent runs was carried out on FFA and its variants TABLE V. RELIABILITY ANALYSIS FOR POWER DEMAND =800MW
to get best and mean cost, and the standard deviation (SD). FFA MFA VSSFA MFFA
Their efficiency was assessed by comparing execution time, Best 40739 40676 40701 40740
the best solution and power loss. Mean 40982.05 40681.3 40886.6 40950.3
SD 121.88065 2.6969769 77.10171 110.85605
TABLE I. PARAMETERS FOR FIREFLY ALGORITHM AND ITS VARIANTS
Values for different algorithms TABLE VI. COMPARISON OF COST FOR FIREFLY VARIANTS
Parameters
FFA MFA VSSFA MFFA Power demand FFA MFA VSSFA MFFA
MaxGeneration 150 150 150 150 600 31489 31447 31576 31481
n 25 25 25 25 700 36075 36006 36036 36021
α 0.2 0.2 Varies with Initially α=0.2 but 800 40739 40676 40701 40740
iterations decreases with
iterations TABLE VII. COMPARISON OF TIME FOR FIREFLY VARIANTS
γ 1 1 1 1 Power demand FFA MFA VSSFA MFFA
β0 1 1 1 1 600 0.519390 0.081707 0.572476 0.250614
βmin NA NA NA 0.2 700 0.545151 0.083877 0.598494 0.422341
800 0.545414 0.081805 0.540558 0.460001
m1 NA 0 NA NA
m2 NA 1 NA NA From the above tables, it can be shown that not only MFA
k NA 1 NA NA obtains the minimum cost for different power demands, but also has
a gain in speed that outperforms that of the original FFA and the
TABLE II. PARAMETERS OF 6 UNITS SYSTEM AND LOSS COEFFICIENT B other two variants.
Unit Pmin Pmax a b c
1 10 125 756.79886 38.53973 0.1524 B. Transmission losses included
2 10 150 451.32513 46.15916 0.10587
3 35 225 1049.9977 40.39655 0.02803
Tables VIII to X demonstrate best, mean, and standard deviation
4 35 210 1243.5311 38.30553 0.03546 for 6 units loss included system for 600, 700 and 800 MW. Fuel cost,
5 130 325 1658.5596 36.32782 0.02111 time of execution and power loss were compared in tables XI to XIII.
6 125 315 1356.6592 38.27041 0.01799
TABLE VIII. RELIABILITY ANALYSIS FOR POWER DEMAND =600MW
FFA MFA VSSFA MFFA
Best 32122 32098 32159 32109
Mean 32373.45 32103.95 32364.75 32274.45
SD 159.54325 4.7069378 159.48894 103.44512

TABLE IX. RELIABILITY ANALYSIS FOR POWER DEMAND =700MW

FFA MFA VSSFA MFFA
Best 37004 36914 36960 36978
Mean 37317.3 36918.9 37165.5 37036.9
SD 186.88812 3.3229663 96.776193 35.982306

978-1-5090-5434-3©2017 IEEE
 3rd International Conference on Advances in Electrical, Electronics, Information, Communication and Bio-Informatics (AEEICB17)
TABLE X. RELIABILITY ANALYSIS FOR POWER DEMAND =800MW Engineering Research and Technology, vol. 2. no. 9, pp. 3147-3152,
FFA MFA VSSFA MFFA 2013.
Best 41939 41898 41976 41930 [3] F. Benhamida, I. Ziane, Y. Salhi, and B. Dehiba, “A quadratic
Mean 42240.35 41901.211 420079.85 41989.05 programming optimization for dynamic economic load dispatch:
SD 167.72574 2.3470773 68.854339 33.536273 Comparison with GAMS,” 3rd IEEE International Conference on
Systems and Control (ICSC), pp. 625-630, 2013.
TABLE XI. COMPARING COST FOR FIREFLY VARIANTS [4] C. E. Lin, S. T. Chen, and C. L. Huang, “A direct Newton-Raphson
Power demand FFA MFA VSSFA MFFA economic dispatch,” IEEE Trans. Power Syst., vol. 7, no. 3, pp. 1149-
600 32122 32098 32159 32109 1154, Aug. 1992.
700 37004 36914 36960 36978 [5] R.A. Jabr, A.H. Coonick, and B.J. Cory, “A homogeneous linear
800 41939 41898 41976 41930 programming algorithm for the security constrained economic dispatch
problem,” IEEE Trans. Power Syst., vol. 15, no. 3, pp. 930-936, 2000.
TABLE XII. COMPARING TIME FOR FIREFLY VARIANTS [6] C. Panigrahi, P. Chattopadhyay, R. Chakrabarti, and M. Basu,
Power demand FFA MFA VSSFA MFFA “Simulated annealing technique for dynamic economic dispatch,” Electr.
Power Compon. Syst., vol. 34, no. 5, pp. 577-586, 2006.
600 3.373328 0.495458 3.422343 1.127240
700 3.411830 0.512442 3.438569 2.675444 [7] N. Noman and H. Iba, “Differential evolution for economic load
800 3.326030 0.513960 3.352762 1.076264 dispatch problems,” Elect Power Syst. Res., vol. 78, no. 8, pp. 1322-
1331, 2008.
[8] S. Sahoo, K. M. Dash, and A. K. Barisal, “Solution of economic load
TABLE XIII. COMPARING LOSS FOR FIREFLY VARIANTS
dispatch by evolutionary optimization algorithms—A comparative
Power demand FFA MFA VSSFA MFFA study,” In Control, Instrumentation, Energy and Communication
600 14.7116 13.7368 13.9435 14.1097 (CIEC), pp. 259-263, 2014.
700 19.5620 18.6507 18.8286 18.4684
[9] F. S. Moustafa, N. M. Badra, and A. Y. Abdelaziz, “Solution of
800 24.5250 24.2210 24.7649 25.2604
economic load dispatch using recent swarm-based meta-heuristic
Even with transmission losses included, MFA maintains its algorithms: A survey,” International Electrical Engineering Journal, vol.
superiority in terms of obtaining minimum cost with the least 7, no.1, pp. 2136-2147, 2016.
execution time. It also minimizes the power loss as shown in the [10] X. S. Yang, “Firefly algorithms for multimodal optimization,” Proc.
above tables. SAGA, vol. LNCS 5792, pp. 169-178, 2009.
[11] R. Subramanian and K. Thanushkodi, “An efficient firefly algorithm to
V. Conclusion
solve economic dispatch problems,” International Journal of Soft
Computing and Engineering, vol. 2, no.1, pp. 52-55, 2013.
In previous studies, firefly algorithm was applied to solve [12] M. Younes, “A Novel Hybrid FFA-ACO Algorithm for Economic
Power Dispatch,” Journal of Control Engineering And Applied
the economic dispatch problem and has succeeded in finding Informatics, vol. 15, no. 2, pp. 67-77, 2013.
more accurate optimal solutions than other metaheuristic [13] I. Fister, I. Fister Jr., X. S. Yang, and J. Brest, “A comprehensive review
algorithms. In this paper, recent variants of firefly algorithm: of firefly algorithms,” Swarm and Evolutionary Computation, vol. 13,
modified firefly, memetic firefly, and variable step size firefly no. 1, pp. 34-46, 2013.
algorithms were applied to solve the ELD problem. A [14] R. Liang, J. Wang, Y. Chen, and W. Tseng, “An enhanced firefly
thorough comparison was carried out between those variants algorithm to multi-objective optimal active/reactive power dispatch with
uncertainties consideration,” International Journal of Electrical Power &
in terms of execution time, optimal cost and losses (in case of Energy Systems, vol. 64, pp. 1088-1097, Jan. 2015
including transmission losses) when solving six thermal units [15] T. Niknam, R. Azizipanah-Abarghooee, and A. Roosta, “Reserve
systems. Constrained Dynamic Economic Dispatch: A New Fast Self-Adaptive
The results verify that the MFA is capable of yielding Modified Firefly Algorithm,” Systems Journal IEEE, vol. 6, pp. 635-
646, 2012, ISSN 1932-8184.
economical solutions with the least computational time and
[16] M. H. Sulaiman, H. Daniyal, and M. W. Mustafa, “Modified firefly
minimal losses compared to FFA, MFFA, and VSSFA. The algorithm in solving economic dispatch problems with practical
MFFA demonstrated its success and potential in finding high constraints,” IEEE International Conference on Power and Energy
quality solutions in a more reasonable time than the VSSFA (PECon 2012), Kota Kinabalu, Malaysia, 2012.
and FFA. Since the ELD problem is a real-time application, [17] M. J. Kazemzadeh-Parsi, “A modified firefly algorithm for engineering
the optimum results obtained by MFA with the significant design optimization problems,” Iranian Journal of Science and
Technology (IJST), Transactions of Mechanical Engineering, vol. 38,
gain in speed makes it a suitable choice to solve the ELD no. M2, pp 403-421, 2014.
problem over FFA, VSSFA, and MFFA. [18] I. Fister Jr., X.-S. Yang, I. Fister, and J. Brest, “Memetic firefly
algorithm for combinatorial optimization,” in Bioinspired Optimization
References Methods and their Applications (BIOMA 2012), 2012.
[19] S. Yu, S. Zhu, Y. Ma, and D. Mao “A variable step size firefly
algorithm for numerical optimization” Applied Mathematics and
[1] J. Zhu, Optimization of power system operation, 2nd ed., New York, Computation, vol. 263, pp. 214-220, 2015.
USA: Wiley, 2015.
[20] K. S. Reddy and M. D. Reddy, “Economic load dispatch using firefly
[2] A. Dhamanda, A. Dutt, S. Prakash, and A.K. Bhardwaj, “A traditional algorithm,” International journal of engineering research and
approach to solve economic load dispatch problem of thermal generating applications, vol. 2, pp. 2325-2330, July-August 2012.
unit using MATLAB programming,” International Journal of

978-1-5090-5434-3©2017 IEEE