332
2, JUNE 2014
Mathematical Modeling of Li-Ion Battery Using
Genetic Algorithm Approach for V2G Applications
Kannan Thirugnanam, Student Member, IEEE, Ezhil Reena Joy T. P., Student Member, IEEE,
Mukesh Singh, Student Member, IEEE, and Praveen Kumar, Member, IEEE
Abstract— This paper presents an electric circuit-based battery
and a capacity fade model suitable for electric vehicles (EVs) in
vehicle-to-grid applications. The circuit parameters of the battery
model (BM) are extracted using genetic algorithm-based optimiza-
tion method. A control algorithm has been developed for the bat-
tery, which calculates the processed energy, charge or discharge
rate, and state of charge limits of the battery in order to satisfy the
future requirements of EVs. A complete capacity fade analysis has
been carried out to quantify the capacity loss with respect to pro-
cessed energy and cycling. The BM is tested by simulation and its
characteristics such as charge and discharge voltage, available and
stored energy, battery power, and its capacity loss are extracted.
The propriety of the proposed model is validated by superimposing
the results with four typical manufacturers’ data. The battery pro-
files of different manufacturers’ like EIG, Sony, Panasonic, and
Sanyo have been taken and their characteristics are compared
with proposed models. The obtained battery characteristics are
in close agreement with the measured (manufacturers’ catalogue)
characteristics.
Index Terms—Batteries, capacity loss, electric vehicles (EVs),
genetic algorithm (GA), vehicle-to-grid (V2G).
I. INTRODUCTION
ITH the increased penetration of electric vehicles (EVs)
W in the market, batteries are considered as one of the im-
portant components in an automobile industry [1], [2]. Among
various available batteries, Lithium-ion batteries are more suit-
able for EVs due to its long cycle life, less self-discharge rate,
high-energy density, high efficiency, low maintenance, and safe
use [3]. EVs’ batteries are charged from the distribution node
and can be discharged, while having excess amount of energy
in it [4]. This concept is referred as grid-to-vehicle (G2V) and
vehicle-to-grid (V2G) technology [5]. If EVs’ batteries are in-
tended to perform G2V and V2G operations, it would get sub-
jected to varying node voltage conditions [4]. Meanwhile, dur-
ing acceleration and regenerative breaking conditions, the EVs’
batteries would get discharged and charged frequently. This
affects the internal circuit parameters due to change in state-of-
charge (SOC) or depth-of-discharge (DOD), charge rate (Cr ) or
Manuscript received February 11, 2013; revised July 6, 2013 and November
1, 2013; accepted December 23, 2013. Date of publication January 20, 2014;
date of current version May 15, 2014. Paper no. TEC-00083-2013.
The authors are with the Department of Electrical and Electronic En-
gineering, Indian Institute of Technology Guwahati, Guwahati 781039, In-
dia (e-mail: kannan@iitg.ernet.in; ezhil@iitg.ernet.in; s.mukesh@iitg.ernet.in;
praveen_kumar@iitg.ernet.in).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TEC.2014.2298460
0885-8969 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 29, NO.
discharge rate (Dr ) of the battery. On the other hand, the ca-
pacity of EVs’ batteries would decrease due to frequent charg-
ing/discharging process at different Cr /Dr [6], [7]. Therefore, a
precise model of the EV battery is very much required to predict
the performance of EVs in real-time V2G interaction. Capacity
fade study is mainly required to predict batteries life time, which
has been neglected in most battery models (BMs) [8]–[10]. The
capacity fade is proportional to charge or discharge rate and
temperature of the battery [6], [7], [11].
An accurate BM and capacity fade model (CFM) have three
advantages as follows:
1) It provides precise battery data for the simulation of EVs
to predict the performance of the system.
2) To analyze the changes in circuit parameters according to
SOC variations, for providing stable control in the system.
3) It is possible to analyze the capacity loss and hence to
predict the lifetime of the battery.
In the literature, there are four types of BMs such as the ex-
perimental, electrochemical, mathematical, and electric circuit
model [5], [12]–[25]. The electrochemical models are the most
accurate models, but they require complex nonlinear differen-
tial equations and detailed knowledge of the chemical reactions
of the batteries [12], [26]. Experimental models require experi-
mentation to determine the internal parameters of the battery [5],
[13]–[15]. However, experimental and electrochemical models
are not well suited to represent the cell dynamics for the purpose
of SOC estimation of battery packs [16], [17]. The mathematical
models are based on stochastic approaches to predict the effi-
ciency and runtime of the batteries [18]–[20]. Due to high com-
plexity and intensive computations, the aforementioned models
are difficult to use in real-time power management and circuit
simulations to predict the performance of the systems [21]. The
electric circuit-based BM can be used to represent the electrical
characteristics of the EVs’ batteries [16], [21]–[25]. The simple
common electric circuit model has a voltage source in series
with an internal resistance [22]. However, this model does not
take the account of battery SOC, Cr , and Dr . There is another
model based on an open-circuit voltage in series with the resis-
tance and parallel RC circuit called as Warburg impedance [25].
The parameters of this model are identified using the compli-
cated impedance spectroscopy method [23], [24], [27]. From
the literature, it is identified that the accurate prediction of inter-
nal battery parameters is a challenging task due to its nonlinear
effects. In fact, the battery parameters extraction is a kind of
optimization problem. Many soft computing and optimization
techniques have been developed for battery parameter estima-
tion [28]–[31]. The genetic algorithm (GA) is one of the best
THIRUGNANAM et al.: MATHEMATICAL MODELING OF LI-ION BATTERY USING GENETIC ALGORITHM APPROACH FOR V2G APPLICATIONS 333

and robust kinds of probability search algorithm which has been

used in this paper [32].
There are several other experimental works found in the litera-
ture, which have focused on capacity fade analysis of a particular
manufacturers’ battery [6], [7], [33]–[35]. However, perform-
ing experiments to determine the capacity fade of a battery for
different Cr and Dr is a tedious as well as time-consuming pro-
cess [11]. Moreover, these studies are confined to a particular
type of battery. The results would differ for other battery types
with different Cr and Dr , which cannot be performed experi-
mentally every time. Therefore, there is a requirement of BM
and CFM which can be used for any type of battery by reducing
testing time and optimizing battery parameters for different Cr
and Dr . Based on these aspects, the main motivations of this
paper are: Fig. 1. Nonlinear BM.
1) To develop a simple circuit-based BM and CFM, which
can be used for any type of battery.
2) To extract the parameters of the model using GA approach. R2 = (a8 + a9 x + a10 x2 )e−a 1 1 y + (a12 + a13 x + a14 x2 ) (2)
3) To validate the developed model with four types of battery C = −(a15 + a16 x + a17 x2 )e−a 1 8 y + (a19 + a20 x + a21 x2 )
manufacturers’ catalogue.
This paper describes an electric circuit-based BM and CFM (3)
which represent different type of battery manufacturers’ Cr 2 −a 2 5 y 2
V0 = (a22 + a23 x + a24 x )e + (a26 + a27 y + a28 y
and Dr characteristics. The parameters of the BM are repre-
sented by a polynomial equation, which is optimized using GA + a29 y 3 ) − a30 x + a31 x2 . (4)
approach. A control algorithm has been developed inside the
battery, which calculates the processed energy, Cr /Dr , current In the aforementioned set of equations, R1 , R2 , C, and V0
SOC (SOCcr ) and DOD (DODcr ) constraints of the battery. Sim- are represented in terms of polynomial equations with 31 coef-
ulations are performed with the developed BM and CFM. The ficients from a1 to a31 . The detailed extraction method to find
validation of the models have been carried out by comparing these polynomial coefficients has been explained in Section III.
the simulated results with the real-time battery data obtained The battery parameters for charging process can be derived by
from four manufacturers’ data sheets such as EIG [36], Sony replacing x and y with Cr and SOCcr , while for discharging
US18650 [11], [37], Panasonic [38], and Sanyo [39]. process with Dr and (1 − DODcr ). Here, (1 − DODcr ) is cho-
This paper is organized as follows. Section II describes the sen as V0 decreases with the increase in DODcr for discharging
proposed BM. The details of battery parameter extraction using voltage. These equations are used for calculating the battery
GA approach are discussed in Section III. Section IV describes terminal voltage for charging or discharging with different Cr
the CFM. The model validation have been done in Section V by and Dr . The terminal charge or discharge voltage of the bat-
comparing the proposed model result with manufactures’ data, tery (VcCi or VdCj ) changes with respect to different capacity of
and the conclusion is given in Section VI. the battery, SOCcr /DODcr levels and Cr /Dr . The parameters of
nonlinear relation of VcCi /VdCj can also be represented in terms of
II. BATTERY MODEL polynomial equations, where i and j denotes ith and jth calcu-
lated value of the charging and discharging voltage, respectively.
An accurate, intuitive, and simple electric circuit-based BM
Therefore, under the constant current, the battery terminal volt-
is developed using an open-circuit voltage (V0 ) with a series
age for charging and discharging scenario with respect to time
resistance (R1 ), and the parallel combination of resistance–
is as follows:
capacitance (R2 C) as shown in Fig. 1. The parameters of the    
battery depends on multivariable functions like SOCcr , DODcr , Qr tc
VcCi = + Ic R2 exp −
Cr , Dr , temperature, and cycle number which has a nonlinear C R2 C
behavior. The parameters V0 , R1 in series with parallel R2 C + V0 − (Ic (R1 + R2 )) (5)
characterizes the usable capacity, instantaneous voltage drop,    
and self-discharge energy. The R2 C network in the model is Qr td
VdCj = + Id R2 exp −
similar to Thevenins’ model which simulates the transient re- C R2 C
sponse. The battery parameters R1 , R2 , V0 , and C have an expo- + V0 − (Id (R1 + R2 )) (6)
nential nature [40] and are represented as polynomial functions.
The general polynomial equations for denoting the parameters where Qr is the remaining capacity of the battery; tc , Ic , td ,
are as follows: and Id are charge time, charge current, discharge time, and dis-
charge current, respectively. Thus, (5) and (6) can accurately
R1 = (a1 + a2 x + a3 x2 )e−a 4 y + (a5 + a6 x + a7 x2 ) (1) represent the behavior of any battery types, if the parameters
334
Fig. 2. Functional flowchart for C r and SOC c r .
are well defined. These equations capture the nonlinear behav-
ior of the battery which depends uniquely on the actual battery
charge/discharge voltage. Hence, this model yields accurate re-
sults and also represents the actual behavior of the battery. The
main feature of this model is that the parameters can be eas-
ily detected and can be compared with different types of battery
manufacturers’ data. Fig. 1 shows the representation of the elec-
tric circuit-based BM with its nonlinear equations.
A. Charge/Discharge Rate and SOC Calculations
The charge or discharge rate algorithm is used to determine
the amount of energy stored or extracted from the EV battery.
The Cr and SOCcr of the battery varies depending on the present
condition of the battery.
Fig. 2 explains the calculation of Cr and SOCcr for the charg-
ing scenario. The control algorithm developed inside the battery
checks the battery status and then calculates the current charge
rate (Crcrt ) of the EV battery. It has also taken into account the
user defined Cr limit (Crlmt ) and initial battery SOC (SOCini ).
The Crcrt and Drcrt of the battery can be expressed as given in
(7). This is calculated based on the current status of the bat-
tery, which is the ratio of the current and remaining capacity of
the battery. The algorithm chooses the minimum of charge rate
based on the Crlmt and Crcrt to regulate the charge current of
the battery. Similar type of the control algorithm is used for the
discharging scenario
Ic Id
Cr = Crcrt = ; Dr = Drcrt = . (7)
Qr Qr
The SOCcr and DODcr can be calculated as
 
Ic Δtc
SOCcr = SOCini + ; DODcr = DODini
Qr 3600
 
Id Δtd
+ .
Qr 3600
Here, SOCini is the initial SOC of the battery. The SOCm ax
and DODm ax are maximum user defined charge and discharge
limits. If SOCcr and DODcr of the battery reaches SOCm ax
and DODm ax , then the control algorithm should not allowed
to charge or discharge the battery to prevent overcharging or
discharging. The control algorithm used to charge/discharge the
EVs’ batteries from/to grid have been reported in [41].
IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 29, NO. 2, JUNE 2014
B. Battery Power and Processed Energy
The battery power for charging (Pc ) and discharging (Pd )
scenario is given as
Pc = VcCi Ic ; Pd = V C
d j Id . (9)
The amount of stored energy (Estor ) during the charging process
depends on the increase in VcCi and SOCcr , which is given as
Estor = VcCi Qr SOCcr . (10)
The processed energy (PEc ) for the charging scenario is given
as

PEc = Estor . (11)
The available energy (Eavail ) in the battery during the discharg-
ing process decreases VdCi with the increase in DODcr , which
can be calculated using the following equation:
Eavail = VdCj Qr DODcr . (12)
The processed energy (PEd ) for discharging scenario is given as

PEd = Eavail . (13)
The total processed energy (Etotal ) of the battery in a cycle is
calculated using the following equation:

Etotal = (PEc + PEd ). (14)
Equations (9)–(14) represent the real-time performance of the
battery during the charging and discharging process. Simu-
lations are done based on these equations for the developed
BM and have been validated with the manufacturers’ catalogue
which is discussed in Section V.
III. PARAMETER EXTRACTION USING GA APPROACH
A. Genetic Algorithm
GA is a technique for stochastic search, learning, and opti-
mization. GA imitates the process of biological evolution, in-
cluding selection, crossover, and mutation based on the princi-
ple; good individuals survive and breed good individuals [31].
Battery parameters are extracted using the GA-based optimiza-
tion technique. Different types of manufacturers’ data of Li-ion
batteries are considered for the extraction purpose. The main
objective of the GA is to optimize the battery parameters poly-
nomial coefficients (a1 − a31 ) to evaluate the equations given
from (1) to (6). The advantages of using the GA to optimize the
battery coefficients are given below.
1) The purpose of using the GA for battery parameter ex-
(8)
traction problem lies in the fact that it requires only man-
ufacturers Cr and Dr characteristics and gives consis-
tent polynomial coefficient of BM during relatively fewer
iterations.
2) The GA is more flexible in extracting the battery parame-
ters with any initial values, while other numerical methods
are incapable of obtaining satisfactory solution.
3) The algorithm is easy to understand and can be optimized
using the fitness function.
THIRUGNANAM et al.: MATHEMATICAL MODELING OF LI-ION BATTERY USING GENETIC ALGORITHM APPROACH FOR V2G APPLICATIONS 335

4) With its optimizing capability, GA steer the fitness func- on the calculated fitness value, the best list gets updated and the
tion to be more representative and yields an accurate solu- lowest fitness is discarded. If F (x) has to be maximized, then
tion set even if the initial values are far from the solutions. f (x) should be minimized which is mathematically expressed
5) The GA encodes an optimum value based on a simple as
chromosome. 1
6) By its evolutionary character, the GA does not suffer from F (x) = . (17)
1 + f (x)
convergence problem and can be dealt with any number
of coefficients in a defined boundary limits. The fitness function is followed by checking its optimal condi-
tion to know whether the solution set has reached the best fitness.
B. Parameter Extraction Process Using the GA Afterwards, the best fitness function is taken to be the parent for
next generation based on the Roulette-Wheel method [42]. The
The main objective is to find the battery parameters for the mathematical formulation of f (x) for charging and discharging
developed BM, which should exactly match the manufacturers’ scenario is given as
data. The GA assigns random values and applies standard muta- ⎧ n
tion and crossover operators without any impact on the periodic ⎪   
⎪
⎪  VcM − VcC 
or continuous nature of the solution. The details of parameters ⎪
⎨ i i
i=0
extraction problem are given below. f (x) = m   (18)
⎪
⎪   M
1) Parameter Specifications: The charge or discharge rate ⎪ C 
⎪
⎩  Vd j − Vd j .
characteristics of the BM depends on polynomial coefficients j =0
(a1 − a31 ) which are given in (1)–(4). To perform GA-based
4) Selection: The individual having the highest fitness value
optimization, the population size (Sn ) has been considered as a
is allowed to remain in the next generation as an elite individual.
matrix size of 300 × 31. The length of the string is taken as 15.
The selection operator is used to choose individual for evolution.
The crossover (Pc ) and mutation probability (Pm ) are taken to
The Roulette-wheel method is used to select the best individuals
be 0.85 and 0.05, respectively. The measured battery terminal
based on the high value of F (x). The bigger f (x) value has
voltage for charging (VcM i
) and discharging (VdMi ) scenarios are
worse quality of the solution set. The selection of the fitness
obtained from manufacturers’ catalogue to optimize the poly-
values for the next iteration can be obtained from (19) to (21).
nomial coefficients.
Here, Favg is the average fitness function value, Eco is expected
2) Generation of Initial Solution Set: The GA process be-
count, and Pes is the probability of each string selected
gins with the generation of an initial random solution set to

the problem. The entire solution set (300 × 31) is discretized f (x)
and the variable is defined with a discrete binary set of strings Favg = (19)
Sn
(am i ,1 . . . am i ,n ). Each string is a combination of different rep-
F (x)
resentation of substrings (am 1 ,n . . . am i ,n ). Let, Sn be the ran- Eco = (20)
dom solution set of battery coefficient having m columns and n Favg
rows and is given in (15). Here, am i ,n denotes nth coefficient Eco
of ith solution set Pes = . (21)
Sn
⎛ ⎞
am 1 ,1 . . . am 1 ,n Thus, the fitness values are arranged based on the maximum
⎜ .. ⎟ .
Sn = ⎝ ... ..
. . ⎠ (15) probability. The best solution set values are converted from real
am i ,1 ··· am i ,n value to binary string of the solution set. The string manip-
ulation has been carried out using genetic operators such as
Each nth coefficients of ith string in the solution set has a reproduction, crossover, and mutation.
15-bit number of a total of 465 bits. The 15-bit number in a 5) Reproduction: This process rejects many worst F (x) val-
single coefficient is denoted as ues and selects best values whose fitness is above Favg . It picks
the best fitness value from the current population and forms a
am i ,n = (x1 x2 . . . xl ) (16)
mating pool. Multiple copies are inserted in the mating pool in a
where x1 and xl are the lower and upper bit element in a single probabilistic manner. In order to choose the strings for next gen-
coefficient. The binary numbers in the strings are converted to eration, a random numbers between 0’s and 1’s are generated.
real value, which will be manipulated with different genetic The cumulative probability is calculated based on the individual
operators. string probabilities. A string is chosen from the random number
3) Fitness Function: The next requirement after generation and is compared with the previous best solution set based on the
of the random solution set is the measure of the quality of the so- cumulative probability. The string with higher fitness is selected
lution set. This can be achieved by establishing a fitness function and is copied to the mating pool.
(F (x)), which is rated by each solution according to its fitness. 6) Crossover: After the next generation of the solution set,
The function that has to be maximized (F (x) = 1) is called as the crossover genetic operators will be applied for the selected
the fitness function. The difference between measured (VcM i
or solution set. During crossover, new strings are created by ex-
VdMj ) and the calculated (VcCi or VdCj ) characteristics for charging changing information among strings in the mating pool. A
or discharging case is determined as given in (5) or (6). Based couple of parent (P1 and P2 ) strings are selected to perform
336
Fig. 3. Flowchart for parameter extraction using GA approach.
Fig. 4. New offsprings (O 1 and O 2 ) generated from the parents (P 1 and P 2 ).
crossover to get two new offspring strings. If the selected par-
ents have the probability of Pco ≤ 0.85, it performs crossover
or else it is copied to the next generation.
To illustrate this process, two parents P1 and P2 are picked
up from the mating pool. The crossover point is chosen to be
9 bit. Two offsprings O1 and O2 are created by exchanging
information among string in the matting pool as shown in Fig. 4.
When reproduction happened between two intelligent genes,
the probability to produce an intelligent gene is high. Good
parent strings are combined to form a better child string, if an
appropriate site is chosen. But good children strings may or
may not be produced. This is not considered much because bad
IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 29, NO. 2, JUNE 2014
Fig. 5. Mutation process for the n th offspring.
TABLE I
POLYNOMIAL COEFFICIENTS (PC) a 1 TO a 3 1 FOR THE CHARGING
AND DISCHARGING SCENARIO
children strings would be discarded in the future mating pool
based on their fitness.
7) Mutation: Mutation selects a particular offspring and ran-
domly performs changes for one or more characters in the string
by changing 0 to 1 or vice versa. The offspring selection is based
on Pm which is evaluated from fitness. If the selected number
is less than Pm , it performs mutation by altering the chosen
string to create a better child string; otherwise, it keeps the bit
unaltered which is used for next generation. Fig. 5 shows the
mutation process of the selected offspring.
After several generations, the algorithm converges the best
individuals to represent the optimal solution. Fig. 3 shows the
functional flowchart to obtain the polynomial coefficients of the
battery. The obtained polynomial coefficients for charging and
discharging scenario are given in Table I.
IV. CAPACITY FADE MODEL
The stored or available energy of the battery decreases due
to frequent charging or discharging at elevated temperatures.
In this paper, the capacity loss is calculated based on variable
charge or discharge rate at an ambient temperature. The charge
and discharge characteristics for a cycle are illustrated in Fig. 6.
A cycle is defined as an interval during which the battery charges
till SOCm ax and discharges up to DODm ax . When the battery
experiences different charge and discharge cycles, it fails to meet
its performance thus reducing its life expectancy. Therefore, it
is required to study the capacity fade to quantify the capacity
loss with respect to number of cycles.
Capacity fade of the battery can be quantified using the Arrhe-
nius equation for constant charge or discharge rate [6]. However,
THIRUGNANAM et al.: MATHEMATICAL MODELING OF LI-ION BATTERY USING GENETIC ALGORITHM APPROACH FOR V2G APPLICATIONS 337

Fig. 6. Battery performance characteristics for a cycle.

TABLE II
PREEXPONENTIAL AND ADJUSTABLE FACTOR FOR SONY US18650 BATTERY

Pre-exponential factor Adjustable factor

A B C D
0.09051 0.003873 1.295 1.442e-5

in real-time grid support, the charge or discharge rate of the bat-

tery would vary with respect to peak power demand and node
voltage variations. This causes the capacity loss of the battery.
To account this loss, a mathematical model is developed in
this paper to predict the capacity fade at different charge and dis-
charge rate. The total capacity fade can be quantified and written
in terms of nominal capacity (Q), Cr , Dr , PEc , PEd , preexpo-
nential (A, B) and adjustable (C, D) factors, gas constant (R),
and temperature (T ) which is given in (22). In (22), QC l k denotes
the calculated capacity loss and the remaining capacity of the
battery can be expressed as given in (23)
 
C PEc Qr Cr SOCcr
QClk = A exp −
RT
  Fig. 7. Functional flowchart for the capacity loss.
D PEd Qr Dr DODcr
+ B exp − (22)
RT
Qr = Q − QC CFM and have been validated with different manufacturers’
lk . (23)
catalogue.
The preexponential and adjustable factors are calculated Fig. 8 shows the detailed block diagram used to validate
based on the trial and error method. The capacity loss char- the proposed battery and CFM. The battery manufacturers’
acteristics are optimized by the fitness function (F  (x)). F  (x) charge/discharge rate characteristics have been extracted using
is maximized, when the difference between the measured (QM lk ) the mathematical equations described in Section II. Fig. 8 has
and calculated (QC lk ) capacity loss is minimized. The mathe- two loops: the inner loop computes the charging process and
matical representation of F  (x) is given in (24), where n is the the outer loop computes the discharging process. The control
number of measured and calculated values algorithm decides the amount of energy stored into the battery
1 or extracted from the battery based on the current energy sta-
F  (x) =   . (24) tus and user defined Cr /Dr limit of the battery. The inputs of
1 + nk=1 QM l k − Ql k
C
the control algorithm are: charge/discharge current, SOC/DOD
The functional flowchart for the capacity loss calculation is limits and polynomial coefficients for charging/discharging sce-
shown in Fig. 7 has explained the complete process to calculate nario (a1 − a31 ). Based on these inputs, the BM calculates
the capacity fade of the battery, where cnt denotes the number SOCcr /DODcr , charge/discharge voltage (VcCi /VdCj ), current
of cycles involved in the calculation process. Table II shows the (Ic /Id ), power (Pc /Pd ), and energy (PEc /PEd ). The capac-
optimal values of preexponential and adjustable factors for Sony ity loss is calculated using energy of the battery, preexponential
US18650 batteries. Simulations are done to verify the developed factor, gas constant, temperature, and adjustable factor.
338 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 29, NO. 2, JUNE 2014

Fig. 8. Block diagram of the battery and CFM.

TABLE III
SPECIFICATIONS OF BATTERIES

Fig. 9. Comparison of the proposed model result with EIG battery at 1C r and
1D r .
V. MODEL VALIDATION
The battery and CFMs explained in Sections II and IV have
been compared with manufacturers’ catalogue to validate its
performance. An extensive simulation study has been performed
with the proposed models. Comparison with four real-time bat-
teries such as EIG, Sony US18650, Panasonic, and Sanyo is pro-
vided to show the superiority of the models. The specifications
of these batteries are given in Table III. Due to unavailability of
the manufacturer data, the models are compared with low volt-
age and ampere-hour ratings. The results have been discussed
in two parts. First part, the validation of the BM has been car-
ried out with four battery manufactures’ data. The capacity fade
results are analyzed in the second part.
The performance of batteries from different manufacturers
is analyzed separately. Comparison has been done for calcu-
lated and measured characteristics of different batteries. The
dotted line represents the measured data from the battery man-
ufacturers’ catalogue and solid line denotes the proposed BM
characteristics. In all the graphs, superscript C and M denotes
the calculated values from the proposed model and measured
value from the manufacturers’ catalogue, respectively.

A. Performance Characteristics of the BM

Case (i)—EIG Battery: Fig. 9 shows the comparison of battery
characteristics with 2.5-V, 8-Ah EIG battery at 1Cr /1Dr .
The charging and discharging voltage with respect to SOCcr
and DODcr are shown in Fig. 9(a) and (b). The simulations are
observed for different SOCcr and DODcr values from 0–100%
and 100%–0%. The end of charge value (EOCV) is 4.3V and
end of discharge value (EODV) is 1.3V. It is observed from
Fig. 9(a) and (b), the end of charge and discharge voltages are
higher and lower with respect to the increase in SOCcr and
DODcr . The initial portion of the curve is linear in fashion and
Fig. 10. Comparison of the proposed model result with EIG battery at 4C r
and 4D r .
after certain period it has varied. The comparison of charging
(Pc ) and discharging (Pd ) power with respect to SOCcr and
DODcr is shown in Fig. 9(c) and (d). Fig. 9(e) and (f) shows the
comparison of charging (Ec ) and discharging (Ed ) energy with
respect to SOCcr and DODcr . It is clear from these plots that
energy and power values are maximum at the end of SOCcr and
DODcr value.
THIRUGNANAM et al.: MATHEMATICAL MODELING OF LI-ION BATTERY USING GENETIC ALGORITHM APPROACH FOR V2G APPLICATIONS 339

Fig. 11. Comparison of the proposed model result with Sony US18650 battery Fig. 12. Comparison of the proposed model result with Sony US18650 battery
at 1C r and 1D r . at 2C r and 2D r .

Fig. 10 shows the comparison results of the battery perfor-

mance at 4Cr /4Dr . The EOCV and EODV for this case are 2.75
and 1.5 V, respectively. As can be seen in Figs. 9 (a) and 10 (a),
for constant current charging, the charging voltage is equal to
the battery terminal voltage. However, in the discharging case,
it has an additional voltage drop. The comparison of Pc , Pd ,
Estor , Eavail with respect to SOCcr and DODcr at 4Cr and 4Dr
are shown in Fig. 10(c)–(f). It can be observed that the energy
extracted from the battery is inversely proportional to Dr and
directly proportional to the battery voltage and current. Notably
for high Cr /Dr , the energy stored/extracted to/from the battery
can be done in a short period of time, while for low Cr /Dr
it takes long time. This affects the net energy stored/extracted
from/to the battery. The time taken to charge or discharge the
battery at 4Cr or 4Dr takes only 0.333 hr, while for 1Cr or 1Dr
it requires 1 hr.
Case (ii)—Sony US18650 Battery: Fig. 11 shows the compar-
ison of battery characteristics with 3.7-V, 1.4-Ah Sony US18650
battery. Fig. 11(a) and (b) shows the simulated characteristics
of the terminal voltage plotted with SOCcr and DODcr . It can
be seen from these figures, EOCV and EODV for Sony battery
are 4.1 and 2.7 V, respectively. The results of the battery power
Fig. 13. Comparison of the proposed model result with Panasonic battery at
and energy shown in Fig. 11(c)–(f) for charging and discharging 1C r and 1D r .
cases are in good agreement. It is observed from Fig. 10 (e) that
the maximum stored energy is 17.46 Wh; however, in Fig. 10(f),
the initial extracted energy is only 17.32 Wh due to the capacity comparison of the battery power (Pc , Pd ) and energy (Estor ,
loss in the battery. Fig. 12 shows the comparison of performance Eavail ) with the manufacturers’ data is shown in Fig. 12(c)–(f).
characteristics of Sony US18650 battery with 2Cr and 2Dr at As it can be seen from Fig. 12(e)–(f), the amount of energy
2.8A. stored into the battery and extracted (available) from the battery
The terminal voltage curves for charging and discharging sce- has reduced with respect to SOCcr and DODcr .
nario are shown in Fig. 12(a) and (b). The EOCV and EODV Case (iii)—Panasonic Battery: Fig. 13 shows the comparison
observed in this case are 4.42 and 2.5 V, respectively. The results with 4.5-V, 0.83-Ah Panasonic battery at 1Cr and 1Dr .
340 IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 29, NO. 2, JUNE 2014

Fig. 14. Comparison of the proposed model result with Panasonic battery at
Fig. 15. Comparison of proposed model result with SANYO battery at
0.9398C r and 1.8795D r .
1C r /1D r .

The terminal voltage characteristics with respect to SOCcr and

DODcr at 1Cr and 1Dr are shown in Fig. 13(a) and (b). The
EOCV and the EODV of this case are 4.3 and 3 V, respectively.
Fig. 13(c)–(f) shows the performance characteristics of the bat-
tery power and energy. In all the cases, the measured data agree
with the proposed data.
The comparison of Panasonic battery with the proposed
model is shown in Fig. 14. Fig. 14(a) and (b) shows the com-
parison of terminal voltage characteristics with 0.9398Cr at
0.78A and 1.8795Dr at 1.56A. The charge and discharge volt-
age characteristics are shown in Fig. 14(a) and (b). The EOCV
and EODV observed in this case are 4.3 and 3 V, respectively.
Fig. 14(c) and (d) shows the performance of the battery power.
The amount of stored and available energy is shown in Fig. 14(e)
and (f).
Case (iv)—Sanyo Battery: The battery used in this case for
comparison is Sanyo which has a nominal capacity 1.5 Ah and
a voltage of 3.7 V. Here, the battery is charged and discharged
with a constant current of 1.5 A. Fig. 15(a) and (b) shows the
terminal voltage curves of charge and discharge characteristics.
As can be seen from the plots, the EOCV and EODV in this Fig. 16. Comparison of proposed model result with SANYO battery at
case are 4.3 and 2.5 V. In this figure, the variation of SOCcr and 2C r /2D r .
DODcr value has been reflected in the charging and discharging
voltage. The performance of power and energy characteristics the proposed BM characteristics exactly fit the manufacturers’
is shown in Fig. 15(c)–(f). characteristics, and in some cases it has an error less than 5%.
Fig. 16 shows the comparison of measured and calculated
characteristics of Sanyo battery of 2Cr /2Dr at 3 A. The EOCV
and EODV observed in this case are 4.02 and 2.5 V, respec- B. Capacity Fade Analysis
tively. Fig. 16(c) and (f) shows the characteristics of the charging The validation of capacity fade study has also been carried
and discharging battery power and energy (Pc , Pd , Eavail and out by comparing the proposed model with four manufacturers’
Estor ) at 2Cr and 2Dr . Thus, it is observed from Figs. 9 to 16, data. The capacity loss data are taken from the battery catalogues
THIRUGNANAM et al.: MATHEMATICAL MODELING OF LI-ION BATTERY USING GENETIC ALGORITHM APPROACH FOR V2G APPLICATIONS
Fig. 17. Capacity loss characteristics of EIG battery.
Fig. 18. Capacity loss characteristics of SONY US18650 battery.
provided by the battery manufacturers. However, due to limited
availability of measured data, the comparison has been done
only for specified charge and discharge rate. Figs. 17 to 20 report
the results obtained from capacity fade study. The variation of
the capacity loss is shown with respect to cycle number and
total processed energy. The remaining capacity of the battery
has been given in each case to show the net processed energy.
Case (i)—EIG Battery: Fig. 17 shows the comparison of ca-
pacity fade characteristics of the EIG battery. Fig. 17 (a) shows
341
Fig. 19. Capacity loss characteristics of PANASONIC battery.
Fig. 20. Capacity loss characteristics of Sanyo battery.
the capacity loss as a function of cycle number for the EIG
battery.
In Fig. 17 (b), the capacity loss is observed as a function of
total processed energy. As it can be seen in Fig. 17(a) and (b), the
capacity loss as a function of cycle number and processed energy
has a rising behavior. Fig. 17(c) and (d) shows the required
and available energy plots with respect to the cycle number.
The observation is done till 3000 cycles to check the increase
and degradation of charge and discharge energy. These results
342
TABLE IV
RESULTS OBTAINED FROM THE BATTERY AND CFM
indicate that the charge and discharge rate plays a strong role
in determining the capacity fade of batteries. Fig. 17(e) and
(f) shows the remaining capacity curve as a function of cycle
number and total processed energy.
Case (ii)—Sony US18650 Battery: The measured and calcu-
lated capacity fade characteristics of Sony US18650 are shown
in Fig. 18. In this case, the capacity fade study has been observed
for 300 cycles. The capacity fade as a function of the cycle
number and total processed energy are shown in Fig. 18(a) and
(b). Fig. 18(c) and (d) shows the characteristics of charge and
discharge energy as a function of cycle number, while the char-
acteristics of remaining capacity and total processed energy are
shown in Fig. 18(e) and (f). It is noticed from the plots that the
energy and remaining capacity of the battery with the increase
in cycle has a falling behavior.
Case (iii)—Panasonic Battery: Fig. 19 shows the measured
and calculated characteristics of capacity fade for Panasonic
battery. Here, the study has been done till 500 cycles. Fig. 19(a)
and (b) shows the capacity loss as a function of cycle number
and energy. The capacity loss as a function of total processed
energy is shown in Fig. 19 (b). The characteristics of required
and available energy as a function of cycle number are shown
in Fig. 19(c) and (d). It can be seen that the energy stored or
extracted from or to the battery has decreased with the increase
in cycle number due to the capacity loss. Fig. 19(e) and (f) shows
the remaining capacity curve as a function of cycle number and
total processed energy. As like previous cases, the energy and
capacity loss characteristics have falling behavior with increase
in cycle number and processed energy.
Case (iv)—Sanyo Battery: Fig. 20 shows the comparison of
capacity fade profiles of Sanyo battery. The capacity loss as a
function of cycle number and total processed energy are shown
in Fig. 20(a) and (b). As can be seen from these plots, the char-
acteristics have rising behavior with respect to cycle and total
processed energy. Fig. 20(c) and (d) shows the characteristics of
required and available energy as a function of cycle number. The
values are observed for 500 cycles to check its performance. The
comparison of the remaining capacity curves as a function of
cycle number and total processed energy are shown in Fig. 20(e)
and (f).
It can be observed from Figs. 17 to 20 that the CFM char-
acteristics exactly fit with the different battery manufacturers’
curves. The typical simulation and measured results of different
battery types such as for EIG (till 3000 cycle), Sony (till 300
cycles), Panasonic (till 500 cycles), and Sanyo (till 500 cycles)
are shown in Table IV.
IEEE TRANSACTIONS ON ENERGY CONVERSION, VOL. 29, NO. 2, JUNE 2014
VI. CONCLUSION
In this paper, an accurate electric circuit-based battery and
capacity fade model have been proposed. The battery param-
eters are extracted using the GA-based optimization method.
Different performance characteristics like the charge voltage,
discharge voltage, power, energy, total processed energy and
capacity loss are analyzed. Four different battery manufactur-
ers’ data are used for validating the result. The proposed battery
and capacity models are simple and it accurately represents the
measured charge and discharge curves of the manufacturers’
datasheets. The simulation and measured results are in good
agreement.
REFERENCES
[1] E. Sortomme and M. El-Sharkawi, “Optimal combined bidding of vehicle-
to-grid ancillary services,” IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 70–
79, Mar. 2012.
[2] V. Agarwal, K. Uthaichana, R. DeCarlo, and L. Tsoukalas, “Development
and validation of a battery model useful for discharging and charging
power control and lifetime estimation,” IEEE Trans. Energy Convers.,
vol. 25, no. 3, pp. 821–835, Sep. 2010.
[3] C. Alaoui, “Solid-state thermal management for lithium-ion EV batteries,”
IEEE Trans. Veh. Tech., vol. 62, no. 1, pp. 98–107, Jan. 2013.
[4] M. Singh, P. Kumar, and I. Kar, “Implementation of vehicle to grid in-
frastructure using fuzzy logic controller,” IEEE Trans. Smart Grid, vol. 3,
no. 1, pp. 565–577, Mar. 2012.
[5] J. Dogger, B. Roossien, and F. Nieuwenhout, “Characterization of Li-
ion batteries for intelligent management of distributed grid-connected
storage,” IEEE Trans. Energy Convers., vol. 26, no. 1, pp. 256–263, Mar.
2011.
[6] P. Ramadass, B. Haran, R. White, and B. N. Popov, “Mathematical mod-
eling of the capacity fade of Li-ion cells,” J. Power Sources, vol. 123,
pp. 230–240, Mar. 2003.
[7] R. Spotnitz, “Simulation of capacity fade in lithium-ion batteries,” J Power
Sources, vol. 113, pp. 72–80, Jan. 2003.
[8] C. M. Shepherd, “Design of primary and secondary cells - part 2. an
equation describing battery discharge,” J. Electrochem. Soc., vol. 112,
pp. 657–664, Jul. 1965.
[9] O. Tremblay, L.-A. Dessaint, and A.-I. Dekkiche, “A generic battery model
for the dynamic simulation of hybrid electric vehicles,” IEEE Int. Veh.
Power Propuls. Conf., Sep. 2007, pp. 284–289.
[10] L. Gao, S. Liu, and R. Dougal, “Dynamic lithium-ion battery model for
system simulation,” IEEE Trans. Comp. Packag. Technol., vol. 25, no. 3,
pp. 495–505, Sep. 2002.
[11] G. Ning, B. Haran, and B. N. Popov, “Capacity fade study of lithium-
ion batteries cycled at high discharge rates,” J. Power Sources, vol. 117,
pp. 160–169, Dec. 2003.
[12] R. Rao, S. Vrudhula, and D. Rakhmatov, “Battery modeling for energy
aware system design,” Computer, vol. 36, no. 12, pp. 77–87, Dec. 2003.
[13] S. Tian, M. Hong, and M. Ouyang, “An experimental study and non-
linear modeling of discharge I-V behavior of valve-regulated lead-acid
batteries,” IEEE Trans. Energy Convers., vol. 24, no. 2, pp. 452–458, Jun.
2009.
THIRUGNANAM et al.: MATHEMATICAL MODELING OF LI-ION BATTERY USING GENETIC ALGORITHM APPROACH FOR V2G APPLICATIONS
[14] G. Capizzi, F. Bonanno, and G. Tina, “Recurrent neural network-based
modeling and simulation of lead-acid batteries charge-discharge,” IEEE
Trans. Energy Convers., vol. 26, no. 2, pp. 435–443, Jun. 2011.
[15] Y.-H. Liu, C.-H. Hsieh, and Y.-F. Luo, “Search for an optimal five-step
charging pattern for Li-ion batteries using consecutive orthogonal arrays,”
IEEE Trans. Energy Convers., vol. 26, no. 2, pp. 654–661, Jun. 2011.
[16] T. Kim and W. Qiao, “A hybrid battery model capable of capturing dy-
namic circuit characteristics and nonlinear capacity effects,” IEEE Trans.
Energy Convers., vol. 26, no. 4, pp. 1172–1180, Dec. 2011.
[17] M. Jongerden and B. Haverkort, “Which battery model to use?,” IET
Softw., vol. 3, no. 6, pp. 445–457, Dec. 2009.
[18] H. Duval, “Computer model of the lead/acid starter battery in automo-
biles,” J. Power Sources, vol. 53, no. 2, pp. 351–357, Feb. 1995.
[19] P. M. Gomadam, J. W. Weidner, R. A. Dougal, and R. E. White, “Math-
ematical modeling of lithium-ion and nickel battery systems,” J. Power
Sources, vol. 110, no. 2, pp. 267–284, Aug. 2002.
[20] Z. Salameh, M. Casacca, and W. A. Lynch, “A mathematical model for
lead-acid batteries,” IEEE Trans. Energy Convers., vol. 7, no. 1, pp. 93–98,
Mar. 1992.
[21] M. Chen and G. Rincon-Mora, “Accurate electrical battery model capable
of predicting runtime and I-V performance,” IEEE Trans. Energy Conv.,
vol. 21, no. 2, pp. 504–511, Jun. 2006.
[22] M. Durr, A. Cruden, S. Gair, and J. McDonald, “Dynamic model of a lead
acid battery for use in a domestic fuel cell system,” J. Power Sources,
vol. 161, no. 2, pp. 1400–1411, Oct. 2006.
[23] H. Blanke, O. Bohlen, S. Buller, R. W. D. Doncker, B. Fricke,
A. Hammouche, D. Linzen, M. Thele, and D. U. Sauer, “Impedance mea-
surements on lead-acid batteries for state-of-charge, state-of-health and
cranking capability prognosis in electric and hybrid electric vehicles,” J.
Power Sources, vol. 144, no. 2, pp. 418–425, Jun. 2005.
[24] F. Huet, “A review of impedance measurements for determination of
the state-of-charge or state-of-health of secondary batteries,” J. Power
Sources, vol. 70, no. 1, pp. 59–69, Jan. 1998.
[25] E. Kuhn, C. Forgez, P. Lagonotte, and G. Friedrich, “Modelling Ni-mH
battery using Cauer and Foster structures,” J. Power Sources, vol. 158,
no. 2, pp. 1490–1497, Aug. 2006.
[26] T. F. F. M. Doyle and J. Newman, “Modeling of galvanostatic charge and
discharge of the lithium/polymer/insertion cell,” J. Electrochem. Soc.,
vol. 140, no. 6, pp. 1526–1533, Jun. 1993.
[27] P. Mauracher and E. Karden, “Dynamic modelling of lead/acid batter-
ies using impedance spectroscopy for parameter identification,” J. Power
Sources, vol. 67, no. 1–2, pp. 69–84, Jul./Aug. 1997.
[28] Y.-S. Lee, W.-Y. Wang, and T.-Y. Kuo, “Soft computing for battery state-
of-charge (BSOC) estimation in battery string systems,” IEEE Trans. Ind.
Elect., vol. 55, no. 1, pp. 229–239, Jan. 2008.
[29] I.-H. Li, W.-Y. Wang, S.-F. Su, and Y.-S. Lee, “A merged fuzzy neural
network and its applications in battery state-of-charge estimation,” IEEE
Trans. Energy Convers., vol. 22, no. 3, pp. 697–708, Sep. 2007.
[30] M. Einhorn, F. Conte, C. Kral, and J. Fleig, “Comparison, selection, and
parameterization of electrical battery models for automotive applications,”
IEEE Trans. Power Electron., vol. 28, no. 3, 1429–1437, Mar. 2013.
[31] K. Deb, Optimization for Engineering Design: Algorithms and Examples.
New Delhi, India: Prentice-Hall, 1998.
[32] R. Escarela-Perez, T. Niewierowicz, and E. Campero-Littlewood, “Syn-
chronous machine parameters from frequency-response finite-element
simulations and genetic algorithms,” IEEE Trans. Energy Convers.,
vol. 16, no. 2, pp. 198–203, Jun. 2001.
[33] P. Ramadass, B. Haran, R. White, and B. N. Popov, “Capacity fade of Sony
18650 cells cycled at elevated temperatures Part I. cycling performance,”
J. Power Sources, vol. 112, pp. 606–613, Aug. 2002.
[34] B. Y. Liaw, R. G. Jungst, G. Nagasubramanian, H. L. Case, and
D. H. Doughty, “Modeling capacity fade in lithium ion cells,” J. Power
Sources, vol. 140, pp. 157–161, Aug. 2005.
[35] P. Ramadass, B. Haran, R. White, and B. N. Popov, “Capacity fade of
Sony 18650 cells cycled at elevated temperatures Part II. capacity fade
analysis,” J. Power Sources, vol. 112, pp. 614–620, Aug. 2002.
[36] EIG, “Battery Charge Rate Characteristics,” (Feb. 2000). [Online]. Avail-
able: http://www.eigbattery.com/
[37] Sony-US18650, “Lithium ion rechargeable batteries—Technical hand-
book,” (May 2001) [Online]. Available: http://www.sony.com.cn/
products/ed/battery/download.pdf
[38] Panasonic, “Lithium Ion Batteries—Individual Data Sheet,” (Apr. 2002).
[Online]. Available: http://www.batteryspace.com/prod-specs/4666.pdf
343
[39] Sanyo, “Lithium Ion Batteries—Specifications,” (Aug. 2003). [On-
line]. Available: http://www.manualslib.com/manual/148360/Sanyo-
Ur18500f.html and http://www.sanyoindia.com/index.aspx
[40] M. Chen and G. Rincon-Mora, “Dynamic lithium-ion battery model for
system simulation,” IEEE Trans. Energy Convers., vol. 21, no. 2, pp. 504–
Jun. 2006.
[41] T. Joy, K. Thirugnanam, and P. Kumar, “Bidirectional contactless charging
system using Li-ion battery model,” in Proc. IEEE 7th Int. Conf. Ind. Info.
Syst., Aug. 2012, pp. 1–6.
[42] L. Zhang, H. Chang, and R. Xu, “Equal-width partitioning roulette wheel
selection in genetic algorithm,” in Proc. Conf. Tech. Appl. Artif. Intell.,
Nov. 2012, pp. 62–67.
Kannan Thirugnanam (S’12) was born in 1983,
Tamil Nadu, India. He received the B.E. degree in
electronics and communication engineering from the
Government College of Engineering, Bargur, India,
in 2005, and the M.E. degree in power electronics and
drives from the Alagappa Chettiar College of Engi-
neering and Technology, Karaikudi, India, in 2009.
He is currently working toward the Ph.D. degree in
the Department of Electronics and Electrical Engi-
neering, Indian Institute of Technology Guwahati,
India.
His research interests include electric vehicles, mathematical modeling of
battery, power conversion, contactless system, and power electronics applica-
tions in power systems.
Ezhil Reena Joy T. P. (S’12) received the Master’s
degree in power electronics and drives from Karunya
University, Coimbatore, India. She is currently work-
ing toward the Ph.D. degree in the Department of
Electronics and Electrical Engineering, Indian Insti-
tute of Technology Guwahati, India.
Her research interest includes converters, invert-
ers, and power electronics applications in power
systems.
Mukesh Singh (S’12) received the B.Sc. degree
in electrical engineering from the Birsa Institute of
Technology, Sindri, India, in 2000, the M.Tech. de-
gree in power systems from the Walchand College
of Engineering, Sangli, India, in 2008, and the Ph.D.
degree from the Department of Electrical and Elec-
tronic Engineering, Indian Institute of Technology,
Guwahati, India.
He has worked as a Lecturer in reputed colleges
for nearly seven years. He is currently an Assistant
Professor in Maharashtra, India. His research inter-
ests include flexible ac transmission system, distributed generation, smart grid,
and vehicle to grid.
Praveen Kumar (M’11) received the B.Tech. degree
in electrical engineering from the National Institute of
Technology, Hamirpur, India, and the M.Tech. degree
in energy system from Indian Institute of Technology,
Delhi, India, in 1998 and 2000, respectively. He re-
ceived the Ph.D. degree from the Delft University of
Technology, Delft, The Netherlands.
He is currently an Assistant Professor at the In-
dian Institute of Technology, Guwahati, India. He has
worked as a Team Leader, Hybrid Systems in Drive
Trains Innovation at Netherlands. His research inter-
ests include optimization of electrical motors and drives, hybrid and electric
vehicles and smart grid.