Ain Shams Engineering Journal (2015) 6, 1187–1194

Ain Shams University

Ain Shams Engineering Journal

www.elsevier.com/locate/asej
www.sciencedirect.com

ELECTRICAL ENGINEERING

Comparative study of PEM fuel cell parameter

extraction using Genetic Algorithm
N. Rajasekar, Basil Jacob, Karthik Balasubramanian, K. Priya
K. Sangeetha, T. Sudhakar Babu *

Solar Energy Research Center, School of Electrical Engineering, VIT University, Vellore, Tamil Nadu 632014, India

Received 11 November 2014; revised 6 May 2015; accepted 18 May 2015

Available online 6 August 2015

KEYWORDS Abstract In this paper, a new formulation based on the derivative of maximum power with respect
Fuel cell parameter extrac- to current is proposed for fuel cell parameter extraction. Compared to conventional extraction
tion; methods, the proposed method has various features such as faster convergence, higher efficiency
Proton exchange membrane and accuracy. In this work, the problem of parameter extraction is formulated as an optimization
fuel cell (PEMFC); task and it is solved using simple GA. Explanations about the new formulation and its implemen-
Genetic Algorithm (GA) tation using GA are elaborated. To validate the candidature of the proposed formulation, perfor-
mance analysis is done with two sets of initial value ranges and the results are compared with the
existing curve fitting method in terms of accuracy, convergence characteristics and objective func-
tion value. The results show that the proposed formulation employing GA is capable of extracting
fuel cell parameters accurately with lesser computational steps and time.
Ó 2015 Faculty of Engineering, Ain Shams University. Production and hosting by Elsevier B.V. This is an
open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction harness the existing renewable energy resources. Fuel cell sys-
tems are prominent energy sources gaining momentum in
Increased power demand, continuous depletion of fossil fuels recent past due to its high efficiency, reliability, durability
and emission of greenhouse gases urge researchers and practi- and ease of application in diverse fields.
tioners to focus on new technologies that can efficiently Fuel cell is an electrochemical device that converts hydrogen
fuel into electric power. Fuel cell system comprises of two
electrodes namely anode and cathode. Hydrogen is allowed
* Corresponding author. Tel.: +91 9490113830.
to pass through the positive electrode and oxygen through
E-mail addresses: [email protected] (N. Rajasekar),
the negative electrode, with an electrolyte between them facili-
[email protected] (B. Jacob), [email protected]
tating the exchange of electric charges. The flow of electrons
(K. Balasubramanian), [email protected] (K. Priya), sangukrish21@
gmail.com (K. Sangeetha), [email protected], sudhakar. through the external circuit produces electric power [1].
[email protected] (T. Sudhakar Babu). Due to its robustness and higher efficiency, fuel cells have
Peer review under responsibility of Ain Shams University. been extensively utilized in commercial, industrial and residen-
tial applications (such as primary and backup power genera-
tion) [2]. Fuel cells can be categorized based on the type of
electrolyte used and the start-up time required. For instance,
Production and hosting by Elsevier
the start-up time for Proton Exchange Membrane (PEM) fuel
http://dx.doi.org/10.1016/j.asej.2015.05.007
2090-4479 Ó 2015 Faculty of Engineering, Ain Shams University. Production and hosting by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1188 N. Rajasekar et al.

cell is 1 s whereas for Solid Oxide (SO) fuel cell it is 10 min. 2. PEM fuel cell model
Under normal operating conditions, a simple fuel cell typically
produces output voltage in the range of 0.5–0.9 V. Since this Fuel cell is an electrochemical device that converts the stored
low voltage becomes insufficient for real-time applications, a energy of fuel directly into electricity with the help of an oxi-
stack of fuel cells are arranged in series so that the required dant [10]. It consists of a cathode and an anode with a
voltage can be attained [3].The most commonly used fuel cell proton-conducting membrane which serves as the electrolyte.
is the PEM fuel cell. It is popular due to its distinctive features The electrochemical equations occurring in electrodes of a
such as no waste is generated, high efficiency, low operating PEM fuel cell can be described as follows [11].
temperature and pressure [4]. Anode side:
It is appreciable to build an effective fuel cell model before
proceeding into the installation part of the system so that the H2 ! 2Hþ þ 2e ð1Þ
design and testing can be made much easier [5]. Modelling of Cathode side:
fuel cell characteristics has drawn considerable attention of
researchers over the last decade as it helps in the better under- 1
2Hþ þ O2 þ 2e ! H2 O ð2Þ
standing of the phenomena occurring within the cell. It is a 2
major concern for researchers and practitioners to model fuel Overall electrochemical reaction:
cell characteristics accurately since its behaviour largely
1
depends upon the predicted characteristics as well as its model H2 þ O2 ! H2 O ð3Þ
2
parameter values. The values of all the model parameters
required for fuel cell modelling are not provided in the manu- The electrochemical model of fuel cell is proposed by
facturer’s datasheet. Therefore, it facilitates the need to accu- Amphlett et al. [12]. The output voltage equation governing
rately compute the values of the unknown parameters by a single fuel cell is
employing any suitable optimization technique. VFC ¼ ENernst  Vact  Vohmic  Vcon ð4Þ
From the past literature, it can be inferred that the previous
attempts followed curve fitting method for fuel cell modelling ENernst can be calculated using the following formula
[6–9,15,16]. In this procedure, model parameters are computed ENernst ¼ 1:229  0:85  103 ðT  298:15Þ
by considering the sum of square of errors between computed
and measured values on predefined points on the fuel cell char- þ 4:3085  105  T½lnðPH2 þ 0:5 lnðPO2 Þ ð5Þ
acteristics. The main shortcomings of this method are as fol- where PH2 and PO2 are partial pressures (atm) of hydrogen and
lows: it consumes more time, is strenuous, error is large and oxygen, respectively. T is the fuel cell absolute temperature
is complex. Hence, in order to overcome these drawbacks a (K).
new formulation is proposed to accurately determine the The activation over potential can be expressed in a para-
model parameters. The proposed formulation is based on the metric form as follows,
fact that derivative of power with respect to current at
Maximum Power Point (MPP) is zero. This formulation Vact ¼ e1 þ e2 T þ e3 T lnðCO2 Þ þ e4 T lnðiÞ ð6Þ
henceforth is named as Maximum Power Point (MPP) method where the terms ei are semi-empirical coefficients, i is the cell
and it applies simple Genetic Algorithm for solving the fuel cell current (A), CO2 is the concentration of oxygen in the catalytic
parameter identification problem. interface of the cathode (mol cm3).
In the proposed formulation, only 3 points on the I-V char-
PO2
acteristics namely Voc ; Vmpp and Isc are utilized to estimate the CO2 ¼ 498
model parameters unlike the curve fitting method which con- 5:08  106 eð T Þ
siders many points on the I-V characteristics to estimate the The ohmic voltage drop is given by the expression [13]
model parameters. This formulation is accurate, fast converg-
Vohmic ¼ iðRm þ Rc Þ ð7Þ
ing and simple since it requires the computation of only 3
points. And,
For validation and detailed analysis of the novel formula- qm l
tion, extensive evaluation and comparison are done between Rm ¼
A
the conventional curve fitting method and the proposed
h    T 2  i 2:5 i
MPP method. In order to estimate the goodness of proposed
181:6 1 þ 0:03 Ai þ 0:062 303 A
MPP method and for fair comparison, constraints such as fit- qm ¼   i   T303
ness function value (best, worst, mean and standard devia- k  0:634  3 A exp 4:18 T
tion), convergence characteristics, matching of the computed
where Rm and Rc are the equivalent inner membrane resistance
and expected I-V characteristics and statistical tools like stan-
and outer contact resistance respectively. l is the thickness of
dard deviation, absolute error and histograms are taken into
the PEM (cm), which serves as the electrolyte of the cell, A
account.
is the active cell area (cm2), qm is the membrane specific resis-
The remaining part of this paper is subdivided into 5 sec-
tivity and k is an adjustable parameter.
tions. Section 2 expounds the PEM fuel cell model. Section 3
The concentration voltage drop is given by,
explains the formulation of the objective function. Section 4  
details the optimization technique used (i.e.) Genetic I
Vcon ¼ b ln 1  ð8Þ
Algorithm (GA). Section 5 deals with results and discussions. Imax
Comparative study of PEM fuel cell parameter extraction 1189

where b is the parametric coefficient (V) that depends on the where

type of cell and its state of operation. I is the actual cell current ððk  0:634Þ þ Jp ð0:6k  0:038Þ þ J2:5 2 2
p DT ð0:217k  0:1375  0:465Jp Þ  0:9Jp Þ
C¼
(A), Imax is the maximum possible value of I. ðk  0:634  3Jp Þ
2

For the PEM fuel cell, the relationship between

PH2 ; PO2 ; PH2 O , and CO2 can be expressed as [14]: Impp
Jp ¼ ð18Þ
logðPH2 O Þ ¼ 2:95  102 ðT  273:15Þ  9:18105 ðT  273:15Þ2 A
T
þ 1:44  107 ðT  273:15Þ3  2:18 ð9Þ DT ¼ ð19Þ
303
0:79
PN 2 ¼ PO ð10Þ
0:21 2  
0:291ði=AÞ 4. Optimization technique
PO2 ¼ Pc  RHc PH2 O  PN2 exp ð11Þ
T0:832
"   1 # 4.1. Genetic Algorithm
1:635ði=AÞ RHa PH2 O
PH2 ¼ 0:5RHa PH2 O exp   1 ð12Þ
T1:334 Pa
Genetic Algorithm is an evolutionary computational technique
where PH2 O is the saturation pressure of water vapour (atm) which follows Darwin’s principle of ‘‘survival of the fittest’’.
PN2 is the partial pressure of N2 at the cathode gas flow chan- The application of this concept for optimization was put for-
nel (atm). Pa , Pc are the anode and cathode inlet pressures ward by Holland in 1975 which was later modified by different
(atm), RHa and RHc are the relative humidity’s of the vapour authors. The essence of GA constitutes the encoding of opti-
in anode and cathode respectively. mization function arrays comprising of bits to emulate chro-
mosomes and different operations are performed on
3. Problem formulation chromosomes by various genetic operators [6,17].
The important processes employed in Genetic Algorithm
are reproduction, selection, crossover and mutation. A set of
From the equations mentioned above, it is evident that many
probable solutions entitled as population is made to undergo
parameters are involved for describing the fuel cell operation
GA operations and their values are frequently modified in
and most of them are unknown. Since they are not given in
order to converge to the best solution. In every step, the parents
the manufacturer’s datasheet, the calculation of these parame-
are chosen from the current population to produce offspring
ters should be carried out by applying some optimization pro-
which is nothing but a new set of chromosomes. The flow chart
cedure. The specific set of parameters required for fuel cell
emulating the process involved in GA is shown in Fig. 1.
modelling are e1 ; e2 ; e3 ; e4 ; b; Rc and k. For effective problem
solving capability, it is important to define a clear objective
4.2. Application of GA towards parameter estimation
function. At the same time, proper definition of objective func-
tion is crucial for extracting accurate parameter values as it
ensures that the obtained model results perfectly synchronizes The fuel cell parameter identification problem aims at finding
with the actual fuel cell characteristics. Hence, in this work an optimal set of parameters that exactly reproduces the actual
objective function is formulated by applying the fact that the fuel cell characteristics. The optimization technique applied for
derivative of power with respect to current is equal to zero the above problem must have the capacity to explore the
at Maximum Power Point (MPP). i.e. dP dI
¼ 0. The fuel cell
DC power output is given by: Start

P ¼ VI ð13Þ
Random generation of
Taking differentiation of above equation we get, Initial population
dP dV
¼VþI ð14Þ
dI dI
Applying MPP condition to the above equation, RHS of Evaluation of
Fitness Function
equation is set to zero
dV V
þ ¼0 ð15Þ Mate selection
dI I
Thus, the objective function for fuel cell parameter extrac-
tion can be expressed as
  Crossover and Mutation
dV Vmpp
J ¼   þ ð16Þ
dI ðVmpp ;Impp Þ Impp
  No Termination
In the above equation the value of dV
dI ðVmpp ;Impp Þ
can be obtained Satisfied
from the basic voltage equation. Yes
2 0 0 11 3
     
dV T b l181:6
  4
¼  e4 þ @
þ Rc þ @
A A C5
 dI  Impp Imax Impp Stop
ðVmpp ;Impp Þ Aexp 4:18ðT303Þ
T

ð17Þ Figure 1 Flow chart of GA.

1190 N. Rajasekar et al.

solution space and arrive at optimal solution set in shorter with the aid of Roulette wheel selection method and multi
time. The various steps involved when GA is applied to the point crossover is employed.
present problem are described as follows:
4.4. Importance of range selection
Step 1: Parameter setting
Fix the GA parameters namely population size, cross over The proper selection of GA parameters alone does not guaran-
rate, mutation rate and termination criteria. tee faithful results as its convergence and the obtainment of
Step 2: Create an initial population faithful results also depends on the chosen initial population.
In this step, chromosomes i.e., individuals of the population Further, the range of fuel cell model parameters namely
are created. In this work, binary coded GA is used with e1 ; e2 ; e3 ; e4 ; b; Rc and k also plays a significant role in the proper
chromosome length of 70. Generate an initial population modelling of fuel cell system. Unfortunately, there is a lack of
with each chromosome (i.e., parameter values) representing proper knowledge about the model parameter values as well
a potential solution to the problem. as their range. This flaw can be concealed up to a certain extent
Step 3: Evaluation of objective function by allowing large number of iterations with entirely different
Fitness i.e., goodness of a solution, of each individual gen- random initial values so that the desired results are obtained.
erated in previous step is evaluated. In the present work, the However, increased number of iterations can yield poor results.
fitness value assigned to each chromosome is same as the Hence, to refine the search process in a proper manner as well as
objective function value; since it is a minimization function. to guarantee sufficient class of accuracy between the computed
After objective function evaluation i.e., Eq. (16), each chro- and actual characteristics, the model parameters are subjected
mosome is assigned a sector proportional to its fitness value to a definite range and the results obtained are analysed.
in a virtual roulette wheel. Roulette wheel selection proce- Further, the range should be constrained by considering the
dure is adopted to select the parents for the next generation. physical meaning of the parameter as well so that it becomes
Step 4: Cross over and mutation operation realizable. There are two possibilities in the range selection pro-
The parents selected performing the above steps are allowed cess 1. Wide range and 2. Narrow range. If the chosen range is
to undergo cross over, i.e., bits of chromosomes are wide, the search space is large and the applied optimization
exchanged, multipoint cross over is followed in this work technique takes longer time to converge to the best possible
for better efficiency. The offspring’s generated via cross solution. Moreover, there is a great possibility that the method
over are mutated. Bits of chromosomes are toggled based may converge to any of the local best solution which resembles
on the value of mutation probability. the best one and further deteriorates the efficiency of search
Step 5: Creation of new population process. To avoid the drawbacks of wide range selection, a nar-
In this step, old population is replaced with the newly gen- row range can be chosen. The issue of settling to local best value
erated population. and higher convergence time can be expelled with this perfor-
Step 6: Termination criteria mance, but care should be taken such that the best value should
Check whether a termination criterion is met. If it is satis- also be contained in the prescribed range. Thus, it is evident
fied, stop and print the result; else go to step 3. that proper selection of parameter ranges assures the expected
class of accuracy.
4.3. GA parameter selection The ranges for model parameters employed in this work are
presented in Table 1 where Range 1 is a narrow range whereas
range 2 is a wider range. These ranges are helpful in identifying
For obtaining the exact model parameters applying GA,
the best objective function for the proposed problem.
proper selection of predefined GA parameters is crucial. If
the selection of the values is not accurate, the result obtained
will be erroneous which results in the mismatch between the 5. Results and discussion
computed fuel cell characteristics and the actual characteris-
tics. Further, in order to rule out the possibility of getting inac- To accentuate the supremacy of the new formulation, its per-
curate results completely the values of population size, formance is compared with that of the conventional curve fit-
crossover probability, mutation rate and maximum genera- ting method. To make a fair comparison, same GA method is
tions are judiciously adjusted in such a way that better results applied for both the approaches and the performance is evalu-
are obtained. Each of the values is modified via continuous ated. Further, the performance of the proposed MPP approach
tuning within a certain range and the best value is evolved. and conventional curve fitting approach is evaluated in terms
Thus, the best set of GA parameters arrived which are men- of fitness function value (best, worst, mean and standard devi-
tioned herein. The value of population size is 100, mutation ation), convergence characteristics, matching between com-
rate is 0.3 and cross over rate is 0.8 and the number of itera- puted and expected I-V characteristics, standard deviation,
tions to be performed is 250. Selection of parents is performed absolute error graph and histograms. The results obtained

Table 1 Different parameter ranges for performance evaluation.

e1 e2 e3 e4 b Rc k
Range 1, Min 0.95 2e3 6e5 1.1e4 0.02 1e4 16
Range 1, Max 0.944 4e3 8.5e5 1.88e4 0.06 8e4 24
Range 2, Min 0.9 1.5e3 5.5e5 1.05e4 0.01 0.5e4 10
Range 2, Max 0.87 4.5e3 9e5 1.93e4 0.07 8.5e4 24
Comparative study of PEM fuel cell parameter extraction 1191

Table 2 Best, worst, mean and the standard deviation values of both the approaches for 150 runs.
Approach Best Worst Mean Standard deviation
Range 1-GA Proposed method 0.003366 0.003749 0.003532 0.0000793
Range 1-GA Point by point method 0.5147 1.44003 0.7511 0.209617
Range 2-GA Proposed method 0.001259 0.00199 0.001544 0.00013793
Range 2-GA Point by point method 0.35283 2.46632 1.122 0.384907

Objective function value Vs No. of iterations 60

0.004
50
0.0035

Frequency
Range 1 40
0.003 Range 2 30
Fitness Value

0.0025 20

0.002 10

0.0015 0
0.0013 0.0014 0.0015 0.0016 0.0017 0.0018 0.0019 0.002
0.001 Objective Function Value

0.0005 Figure 5 Histogram of the objective function value for Range 2

0 for proposed MPP method.
0 50 100 150 200 250 300
No.of Iterations 30
25
Figure 2 Variation of objective function for the proposed
Frequency

20
method for range 1 and range 2.
15
10
5
Objective Function Vs No. of iterations
10 0
9
Range 1
8
Range 2 Objective function value
7
Fitness Value

6 Figure 6 Histogram of the objective function value for Range 1

5 for conventional curve fitting method.
4
3 20
18
2 16
1 14
Frequency

0 12
0 50 100 150 200 250 10
No. of iterations 8
6
4
Figure 3 Variation of objective function by the conventional 2
curve fitting method for range 1 and range 2. 0

Objective function value

90
Figure 7 Histogram of the objective function value for Range 2
80
70
for conventional curve fitting method.
Frequency

60
50 are explicitly discussed. Programs were developed in
40 MATLAB for GA with both methods. Simulations are per-
30 formed using Pentium 4 INTEL 4 GB RAM 2.4 GHz
20 processor.
10
GA is made to run 150 times for both the approaches and
0
0.0034 0.0035 0.0036 0.0032 comparison is made in terms of best, worst, mean and stan-
Objective function value dard deviation of objective function value obtained for all
the iterations is described in Table 2.
Figure 4 Histogram of the objective function value for Range 1 From Table 2, it is clearly evident that the proposed MPP
for proposed MPP method. method provides better results for all range of initial values.
1192 N. Rajasekar et al.

the effectiveness of the proposed formulation. The results con-

firm that the proposed approach is robust and it can be
employed for any chosen parameter range which exactly meets
the required standards.
Figs. 2 and 3 show the variation of fitness function value
against the number of iterations for best run of the proposed
MPP and conventional curve fitting approach respectively.
From the graphs, it is observed that there is a drastic reduction
in the objective function value when the new approach is
employed. Further, independent of the model parameter
range, the novel formulation converges to a very low fitness
value after settling which is in the order of 102 . However,
the objective function value is very close to unity when conven-
tional approach is applied. Moreover, the ultimate aim of the
Figure 8 Comparison of the proposed MPP, point by point
optimization technique is to achieve a very small objective
method and expected curves under 2.5/3 bar and 1.5/1.5 bar
function value which basically interprets that the error between
pressure.
extracted and actual characteristics is very low. This is
achieved in the proposed MPP method. Further the time taken
for GA optimization technique with the proposed MPP
method and point by point method is found to be 3.54 s and
8.62 s respectively. The difference in time taken is due to the
large computations involved in the point by point method.
Figs. 4 and 5 show the histograms for the frequency of fit-
ness function value for two ranges with the proposed MPP
method and Figs. 6 and 7 show for conventional curve fitting
approach. From the histograms, it is evident that the fre-
quency of getting best objective function values with less mag-
nitude is higher for the proposed method compared to that of
the conventional curve fitting method. However, due to larger
exploration area in range 2, as well as higher computational
data, the curve fitting method produces poor results in both
the cases. Further, the curve fitting procedure may end in local
Figure 9 Comparison of the proposed MPP, point by point optima in range 2 due to vast search space.
method and expected curves under 3/5 bar and 1/1 bar pressure. The extracted model parameters for both the approaches
are subsequently substituted in the MATLAB/SIMULINK
fuel cell model to plot the I-V characteristics of the fuel cell
Further, there is a remarkable improvement in the fitness func- stack. For comparison, experimental data are taken from man-
tion value obtained through the novel proposition compared ufacturer’s datasheet and compared with computed values for
to conventional curve fitting procedure which again proves the same fuel cell stack. The comparison between actual and

Absolute error curve

0.8

0.7 1/1 bar 343.15 K

1.5/1.5 bar 343.15 K
0.6 3/5 bar 353.15 K
2.5/3 bar 343.15 K
Absolute Error

0.5

0.4

0.3

0.2

0.1

0
4
3
20 25
Pressure in bar 2 10
15
1 0 5
Current in Amperes

Figure 10 Absolute error curve-proposed MPP method.

Comparative study of PEM fuel cell parameter extraction 1193

Absolute error curve

1.2 1/1 bar bar 343.15 K

1.5/1.5 bar 343.15 K
3/5 bar 353.15 K
1 2.5/3 bar 343.15 K

0.8
Absolute Error
0.6

0.4

0.2

0
4 25
3.5 20
3 15
2.5 10
2
1.5 5
Pressure in bar 1 0 Current in Amperes

Figure 11 Absolute error curve-conventional curve fitting method.

extracted I–V characteristics for different conditions of pres- with the widely used curve fitting approach. The results
sure and temperature is done in detail. demonstrate that the new MPP formulation applying GA per-
A closer observation of the graphs Figs. 8 and 9 indicates forms better than curve fitting method in terms of accuracy,
that the computed and expected fuel cell characteristics are convergence characteristics, convergence speed and objective
exactly matching when the proposed approach is employed function value.
even under different test conditions however there is a notice-
able drift in the characteristics when point by point approach Acknowledgement
is executed. Hence, the proposed formulation always promises
to provide appropriate matching between the computed and The authors would like thank the Management, Vellore
the actual characteristics irrespective of the test conditions. Institute of Technology University, Vellore, India for
For further justification of the proposed method, the absolute providing support. This research work has been carried out
error is calculated by comparing the difference between computed in Solar Energy Research Center, School of Electrical
and actual I-V curve values. The calculation of the absolute error Engineering, VIT University, Vellore. The authors would like
from I-V curves is based on the following equation: to sincerely thank the reviewers for their valuable comments
  and recommendations to improve the quality of the paper.
Absolute error ¼ Vexperimental  Vcomputed 
The absolute error curves plotted for four different operating References
conditions (for 1/1 bar 343.15 K, 1.5/1.5 bar 343.15 K,
2.5/3 bar 343.15, and 3/5 bar 353.15 K) for both the [1] Yang Shipin, Wang Ning. A novel P systems based optimization
approaches are presented in Figs. 10 and 11 respectively. algorithm for parameter estimation of proton exchange mem-
It is noteworthy to mention that the absolute error value is brane fuel cell model. Int J Hydrogen Energy 2012;37:8465–76.
[2] Askarzadeh Alireza, Rezazadeh Alireza. A grouping-based global
considerably reduced when the new method is implemented
harmony search algorithm for modeling of proton exchange
and it is undoubtedly noticeable in Figs. 10 and 11. The max-
membrane fuel cell. Int J Hydrogen Energy 2011;36:5047–53.
imum error is less than 0.2 for the new proposition whereas it [3] Correa JM, Farret FA, Popov VA, Simoes MG. Sensitivity
rises up to a value close to unity in the conventional curve fit- analysis of the modelling parameters used in simulation of proton
ting approach. Moreover, the deviation in error value is very exchange membrane fuel cells. IEEE Trans Energy Convers 2005;
less even under different test conditions. This again gives 20:211–8.
strong evidence that the proposed MPP approach is robust [4] Kadjo AJ-J, Brault P, Caillard A, Coutanceau C, Garnier JP,
and reliable under all test conditions and it always promises Martemianov S. Improvement of proton exchange membrane fuel
to provide error value within a limited low range. cell electrical performance by optimization of operating param-
eters and electrodes preparation. J Power Sources 2007;172:
613–22.
6. Conclusions [5] Oliveira VB, Falcao DS, Rangel CM, Pinto AMFR. A compar-
ative study of approaches to direct methanol fuel cells modelling.
In this paper, a novel problem formulation based on derivative Int J Hydrogen Energy 2007;32:415–24.
[6] Ohenoja Markku, Leiviska Kauko. Validation of genetic algo-
of power with respect to current at maximum power point is
rithm results in a fuel cell model. Int J Hydrogen Energy 2010;35:
proposed. A simple Genetic Algorithm is used to solve
12618–25.
the above formulation. The seven model parameters [7] Askarzadeh Alireza, Rezazadah Alireza. Artificial immune sys-
e1 ; e2 ; e3 ; e4 ; b; Rc and k are extracted via GA procedure. For tem-based parameter extraction of proton exchange membrane
demonstration, results obtained with GA are benchmarked fuel cell. Electr Power Energy Syst 2011;33:933–8.
1194 N. Rajasekar et al.

[8] Zhang Wei, Wang Ning, Yang Shipin. Hybrid artificial bee colony Karthik Balasubramanian – Completed
algorithm for parameter estimation of proton exchange mem- B. Tech in Electrical and Electronics
brane fuel cell. Int J Hydrogen Energy 2013;38:5796–806. Engineering with Minor in Industrial
[9] Ye Meiying, Wang Xiaodong, Xu Yousheng. Parameter identi- Automation from VIT University, Vellore,
fication for proton exchange membrane fuel cell model using India, in 2014. He is currently working in
particle swarm optimization. Int J Hydrogen Energy 2009;34: Saipem India Projects Pvt. Ltd. as an Electrical
981–9. Engineer. His area of research interest is
[10] Ramakumar R. Fuel cells-an introduction. In: Proceedings of Power engineering, Fuel cell system design,
IEEE-PES summer meeting 15, 2001. p. 702–9. PV system design and applications of Power
[11] Rowe Andrew, Li Xianguo. Mathematical modelling of proton electronics in Renewable Energy systems.
exchange membrane fuel cells. J Power Sources 2001;102:82–96.
[12] Amphlett JC, Baumert RM, Mann RF, Peppley BA, Roberge PR.
Performance modelling of the Ballard mark IV solid polymer
electrolyte fuel cell. J Electrochem Soc 1995;142:1–15. K. Priya – Completed B.E degree from Anna
[13] Mann RF, Amphlett JC, Hooper MAI, Jensen HM, Peppley BA, University, M.Tech from V.I.T University,
Roberge PR. Development and application of a generalized Vellore, India. She is currently a full time
steady-state electrochemical model for a PEM fuel cell. J Power Research in the School of Electrical and
Sources 2000;86:173–80. Electronics Engineering, V.I.T. University.
[14] Mo Z-J, Zhu X-J, Wei L-Y, Cao G-Y. Parameter optimization for Her research interests are application of
a PEMFC model with a hybrid genetic algorithm. Int J Energy Optimization techniques for Electrical engi-
Res 2006;30:585–97. neering, development of novel nano compo-
[15] Askarzadeh A, Rezazadeh A. An innovative global harmony sites for HV applications.
search algorithm for parameter identification of a PEM fuel cell
model. IEEE Trans Ind Electron 2012;59:3473–80.
[16] Outeiro MT, Chibante R, Carvalho AS, de Almeida AT. A
parameter optimized model of a proton exchange membrane fuel
cell including temperature effects. J Power Sources 2008;185: Sangeetha K. – Received the Bachelor’s degree
952–60. in Electrical & Electronics Engineering from
[17] Rajasekar N, Krishna Kumar Neeraja, Venugopalan Rini. Mahatma Gandhi University, Kerala, in 2012.
Bacterial foraging algorithm based solar PV parameter estima- She is currently pursuing her Master’s
tion. Sol Energy 2013;97:255–65. degree in Power Electronics & Drives from
the School of Electrical Engineering, VIT
University, Vellore. Her current research
N. Rajasekar – He was born in Tamil Nadu, interest includes photovoltaic system design,
India. He obtained the B.E Degree from Maximum Power Point Tracking technology
Madras University, M.E degree in power along with the application of power electron-
electronics and Drives from Sastra University, ics in various renewable energy systems.
Thanjavur, Tamil Nadu, and Ph.D. degree
from NIT Trichy. He is presently working as
Professor in VIT University, Vellore, India. T. Sudhakar Babu – Completed B. Tech from
He has presented and published many JNTU University, Hyderabad, in 2009 and
research papers in various national and has completed M. Tech in Power Electronics
international conferences and journals. His and Drives form Anna University in 2011. He
research interests include application of power is currently pursuing Ph.D. in VIT University,
electronics in renewable energy sources and Vellore, India. His area of Interest is renew-
DC–DC converters. able energy sources and power electronics
applications in Renewable Energy systems.
Basil Jacob – Completed M. Tech in Power
Electronics and Drives from VIT University,
Vellore, India, in 2014. He is currently work-
ing as Lecturer in a Reputed University in
Kerala. His area of research interest is Power
electronics and Renewable Energy systems.