i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 8 8 3 3 e8 8 4 2

Available online at www.sciencedirect.com

ScienceDirect

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

Parameter identification and state-of-charge

estimation for lithium-polymer battery cells using
enhanced sunflower optimization algorithm

Ragab A. El-Sehiemy a, M.A. Hamida b,*, T. Mesbahi c

a
Department of Electrical Engineering, Faculty of Engineering, Kafrelsheikh University, Egypt
b
Ecole Centrale de Nantes, LS2N UMR CNRS 6004, Nantes, France
c
ICube UMR CNRS 7357, INSA Strasbourg, Strasbourg, France

highlights

An Enhanced Sunflower Optimization Algorithm (ESFOA) is investigated.

ESFOA is used for parameter identification and SoC estimation of lithium ion battery.

Three cases are considered to simulate single and multi-objective frameworks.

Numerical simulations are employed to show the capability of the proposed ESFOA.

Assessing the ESFOA with water-cycle and whale optimization algorithms is carried out.

article info abstract

Article history: This paper is concerned with the investigation of accurate parameter identification method
Received 21 September 2019 and state of charge (SoC) estimation for Lion Lithium battery. The proposed identification
Received in revised form method is implemented using an accurate state space model obtained from electric
8 January 2020 equivalent circuit. The process of parameter identification is expressed as nonlinear
Accepted 12 January 2020 optimization problem. An Enhanced sunflower optimization algorithm (ESFOA) is
Available online 6 February 2020 employed to solve such problem. The search space is managed by applying the reduction
strategy. This strategy is accomplished with the sunflower optimization algorithm to
Keywords: enhance the solution quality. Three cases studied are considered as single and multi-
Lithium-polymer battery cells objective frameworks. In these cases, battery voltage or SoC or combined between them
Sunflower optimization algorithm as objective functions are optimized for the three cases studied. Numerical simulations as
Parameter identification well as experimental implementation are executed on 40 Ah Kokam Li-Ion Battery to prove
State of charge estimation the capability of the proposed parameter identification method. The ability of the proposed
Driving cycle ESFOA is accomplished with high accuracy is proven compared with Water-Cycle and
Reduction strategy Whale optimization algorithms for two driving cycle profiles. Added to that, high closeness
is achieved compared with the experimental measurements for battery parameters and
SoC. The solution quality improvement of the proposed ESFOA is noticed as it achieves the
lowest the fitness function levels (in the range 60e90%) of the cases studied compared with
the competitive optimization algorithms.
© 2020 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.

* Corresponding author.
E-mail addresses: [email protected] (R.A. El-Sehiemy), [email protected] (M.A. Hamida).
https://doi.org/10.1016/j.ijhydene.2020.01.067
0360-3199/© 2020 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.
8834 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 8 8 3 3 e8 8 4 2

Impedance-based estimation method, the battery SoC is

Introduction dependent on some of its impedance parameters which
boost researchers to identify this parameters for the SoC
Motivation estimation [15]. The dependence of the SoC on the
impedance parameters changes significantly over the bat-
The technology of Lithium-polymer cells is very attractive tery lifetime which represents the main drawbacks of this
in the domain of energy-storage devices [1,2]. This tech- method [16].
nology combines the following merits: very high energy and
Static battery characteristics estimation method, it con-
power densities allowing to use for plug-in hybrid electric sists to use the battery voltage, current and temperature
vehicles (PHEVs). One of the major development of this relationship when the battery load rests constant to esti-
technology is the lithium-ion (Li-ion) battery [3,4]. The mate the SoC [17].
battery management systems have a crucial role for the
Artificial intelligence estimation methods [18], the artificial
power efficiency of transportation systems. Their role en- intelligence methods have been applied for SoC estima-
sures the optimal energy consumption and to make energy tion, such as fuzzy logic (FL), neural network (NN) [19],
management system information to be available for the ANFIS, support vector machines (SVMs).
vehicular in normal and abnormal operating conditions [5].
Measurement techniques based estimation, several
The SoC estimation is an urgent prerequisite to achieve a approach have been proposed in this category one of them
good supervising and controlling of battery charging and uses additional sensors are used for the battery magnetic
discharging [6]. Moreover, accurate estimation of battery characteristic measurement which can be considered as an
parameters and SoC represents an urgent need for many indicator for the SoC [20].
reasons such as extending battery life, battery state of
Model-based estimation method, Its principle is the use of
charge controlling, improving the battery performances [7], the battery measurements such as the voltage, the current
optimizing the energy management, supervising the safety and the temperature to design such battery model. The
of the battery [8e10]. relationship of the battery SoC and OCV is used to incor-
porate the SoC into the battery model. In this model the
Literature review SoC is represented by a state equation. This model can be
used therefore to estimate the SoC using the measure-
The battery is a complex non-linear and non-stationary elec-
ments as inputs. Two model types are generally employed,
trochemical system. This non-linearity of the battery is due to
the electrical one is the famous model type for SoC esti-
the fact that the relationship between the current applied at
mation [6].
the input and the voltage recovered at the output cannot be
expressed by a linear relationship. The battery is non-
Among the previous estimation methods, the electrical
stationary because the characteristics of the internal elec-
battery model is widely used to identify the battery parame-
trical parameters of the battery are variable during a charge/
ters and to estimate the SoC. Based on this model, several
discharge cycle and during its life cycle [11]. The definition of
parameter estimation methods were developed such as ob-
the parameters value and equation type of a battery model is
servers in Ref. [21], classical identification methods [22], on-
based on the exploitation of the battery voltage curves
line estimation procedure was presented in Ref. [23]. The
(output) as a function of a current profile input). However, the
state observer based methods have been largely used in the
power or current profiles directly derived from measurements
literature allowing to an online identification of battery pa-
or estimated in relation to normalized driving cycles in elec-
rameters. The used methods are sliding-mode observer,
tric vehicle applications are relatively complex. There are
Luenberger observer [24], Kalman filter, adaptive observers
several approaches in the literature to simplify these profiles,
[25]. For these methods, the observability of battery models is
based on statistical counting and classification methods.
rarely studied. Moreover it not always easy to prove the sta-
Different battery models have been investigated in the liter-
bility of the proposed observer. Machine learning and artificial
ature according to the target accuracy degree and the practical
intelligence techniques such as neural network [26] have been
application such as electric vehicles. An efficient dynamic
used to estimate the SoC and the battery parameters, support
model that comprising two elements, resistance and capacitor
vector machine [27] and multi-variable adaptive regression
(RC) in its equivalent circuit was proposed [12]. Several works
splines [28]. In recent decades, continuous development is
have been proposed in the literature to address the parame-
noticed in the application of recent evolutionary optimization
ters identification and SoC estimation. These methods can be
techniques in variant engineering and science fields. These
classified as follows:
algorithms have been adopted to solve various engineering
optimization problems such as neural network algorithm [29]

Ampere-hour counting method which has economical and
and particle swarm optimization (PSO) [30], have been
technical benefits, low current sensors costs and low
employed in optimization and applied inverse methods due to
power consumption [13].
their excellent global search abilities. The estimation algo-

Open-circuit voltage (OCV) estimation method, the OCV
rithms and on-line parameter identification methods are
and SoC relationship is exploited for all battery technolo-
needed to guarantee the accuracy of the model-based SOC
gies SoC estimation [14]. This relationship changes slightly
estimation with nonlinear battery models. On the other hand,
over battery lifetime, which represents one of the most
numerous the parameters and SoC estimation based on
advantages of this method.
optimization algorithms such as particle swarm optimization
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 8 8 3 3 e8 8 4 2 8835

[30], Hybrid Particle swarm-Nelder-Mead algorithm [31] and

some other classical optimization algorithms have been pro-
posed to estimate the machine parameters based generally on
algebraic models. In Ref. [32] grey wolf optimization algorithm
has been compared with PSO and genetic algorithm tech-
niques for only parameters estimation of lithium battery
based on algebraic model. Several efforts with the application
of recent optimization algorithms were developed for solving
energy applications such as parameter estimation of solar Fig. 1 e Battery equivalent circuit.
cells such as [33], technical, economical and environmental
scheduling of energy resources [34,35], electrical based inte-
gration of renewable energy into stand-alone systems [36],
and for energy management of hybrid renewable energy Lithium ion battery modeling
resource as presented in Refs. [37,38] In 2018, the SFOA is one
of the meta-heuristic algorithms that was firstly proposed by The common used equivalent circuit model is constructed
Gomes and others [39] in their work about identification of with n times RC elements denoted by nRC model [41], normally
damage on laminated composite plates. Another application up to 3 RC elements are used [42]. In this paper, a reduced
of SFOA for parameter estimation of three diode models of model with n ¼ 1 i. e single RC model is considered as shown
solar cell in Ref. [40]. SFOA stimulated by the sunflowers in Fig. 1. The battery model is modeled by using RC equivalent
movement towards the sunlight [39]. The SFOA is stimulated circuit as shown in Fig. 1. The relaxation effect is represented
by the inverse-square law of radiation intensity, where the by RC group. This model avoids the complexity raised for the
amount of heat is proportional to the squared distance be- problem of parameters identification. The following sub-
tween sunflowers and the sun. sections describe the mathematical and state space model of
the battery. Equation (1) presents a linear relationship to
Paper contribution describe the static characteristic of the battery under pre-
determined conditions of temperature and age, between the
In this paper, an enhanced version of sunflower optimization Voc and SoC is assumed as [43] as:
algorithm is investigated for parameter identification and SoC
estimation. The salient features of the proposed method can VOC ¼ f ðSOCÞ ¼ b0 þ b1 SOC: (1)
be summarized as follows: The parameters identification and SoC estimation problem
is formulated as state space model. Based on Fig. 1 and Eq. (1),
1. Using the non-linear electrical circuit for modeling lithium the state space model for the battery dynamics is represented
battery with state space representation. in the following form:
2. Developing the ESFOA based on reduction strategy to
enhance the search space in order to obtain the optimal X_ ¼ AX þ Bu (2)
parameters and to estimate SoC of lithium ion battery.
3. Three cases are considered to simulate single and multi- y ¼ CX þ Du þ b0 (3)
objective frameworks. The objectives measure the close-
where,
ness between the estimated and experimental recorded for 2 3
2 3 1
the terminal battery voltage and state of charge.
 0 0 6Q 7
SOC 6 R7
4. A comparison study is employed to show the capability of X¼ ; A¼4 1 5; B ¼ 6 7; C ¼ ½b1 1; D ¼ R0 : VT .
Vrc 0  4 1 5
the proposed ESFOA. RC
5. Assessing the sunflower optimization algorithm with C
and IL are the battery terminal voltage and current, respec-
water-cycle and whale optimization algorithms.
tively. R0 is the internal resistance. Qr is assumed to be the
nominal battery capacity. In the current work, the terminal
voltage (VT ) and current (IL ) values are assumed to be acces-
Paper structure
sible from the state space dynamic model in Fig. 1. Both the
temperature and ageing of the battery effects are not consid-
The following sections of the current work are organized as
ered in this work. According to the system Eqs. (2) and (3), the
follows: Lithium Ion Battery modeling presents to the state
battery parameters are identified for state of charge estima-
space dynamic modeling of the lithium ion battery. The esti-
tion as well as computing the terminal battery voltage.
mation problems of battery parameters and state of charge
are formulated as optimization problem with single and
multiobjective frameworks in Problem formulation.
Sunflower solution methodology presents the sunflower
Problem formulation
optimization algorithm for solving the predefined problem in
Parameterization model development
Problem formulation. Design of experiment presents the
setup of experimental works and the related bench marking
The dynamic model of the battery described in the previous
driving cycle. Conclusion concludes the current research
section is used to identify the battery parameters to
findings.
8836 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 8 8 3 3 e8 8 4 2

estimate state of charge. To optimize the parameters of the Constraints

Li-ion battery model, an objective function has to be set for
fitting the estimated output voltages with the corresponding The previous objective functions, Eqs. (5)e(7), are solved by
recorded voltage in the experimental tests. The proposed using ESFOA subject to the following constraints:
objective function is based on the minimization of Squares
Error Sum (SSE) between the experimental data and the Rmin  R  Rmin (8)
model-based simulation results. The general representation
of the proposed optimization problem takes the following R0min  R0  R0min (9)
form:
Cmin  C  Cmin (10)
minu Fi ðuÞ; i ¼ 1; 2:nobf (4)
b0min  b0  b0min (11)
umin  u  umax
b1min  b1  b1min (12)
where, Fi ðuÞ is the objective function i, u is the estimated pa-
rameters of the battery,N is the number of the estimated and Qrmin  Qr  Qrmin (13)
umin , umax are the parameter bounds of control variable's vector
where min and max operators refer the minimum and the
u that are reflected the accepted research domain, nobf is the
maximum values of he identified parameters.
number of objective functions. The identification of the model
parameters is based on the exploitation of the battery exper-
imental data.
Sunflower solution methodology
Objective functions
ESFOA concept and mathematical representation
In the proposed method, three cases studied are considered.
The concept of sunflower optimization algorithm (ESFOA)
These cases can be defined as follows:
simulates the motion of the sun flowers toward the sunlight. It
is dependent on pollination intensity between the closeness
1. Minimizing the normalized deviation between the esti-
sunflowers. ESFOA is considered as a new optimization algo-
mated and experimental battery voltage F1 ðuÞ with Equa-
rithm that is dependent on the inverse square law radiation
tion (5) which is considered as the primary objective
[39,40,44] as:
function of the considered problem as
Sp
Sr ¼ (14)
X V b
bat  Vex
2 4pd2
F1 ðuÞ ¼ max ; (5)
Vbat where Sr is the solar radiation intensity, Sp is defined as the
sun power, d is the distance between the sun and the sun-
where Vb bat is the value of the estimated battery model voltage.
flower. Sunflower ðsi Þ is moved toward the sun, and its di-
However, Vbat is the recorded experimental battery voltage.
rection is determined by equation
2. Minimizing the normalized deviation between the esti-
mated and experimental state F2 ðuÞ of charge with aid of ! X  Xi
Si ¼ ; i ¼ 1; 2; …; np : (15)
Equation (6). This equation is considered as the primary k X  Xi k
objective function instead of 5 in the previous case as where Xi is the current position of sunflower i, X is the best
X S OC
b  soc2 solution among np sunflowers. The norm operator is denoted
F2 ðuÞ ¼ ; (6) by kk. The updating equation for sunflower direction toward to
socmax
the sun is represented by:
b is the value of the estimated battery state of
where S OC
charge. Xiþ1 ¼ Xi þ di Si (16)
3. Providing a multi-objective function MoF for parameter where Xiþ1 is the updated position of sunflowers.
estimation of the battery by combining Equations (5) and (6).
This objective function aims to minimize simultaneously di ¼ l*Pi ð k Xi þ Xi1 k ÞjjXi þ Xi1 jj (17)
the deviation between the estimated and experimental where di is the step of sunflower toward the direction of the
battery voltage and state of charge. The common multi- sun. Where l is the sunflower inertial displacement.
objective function of the battery model identification is Pi ðjjXi þXi?1 jjÞ refers to the probability of two adjacent sun-
given by: flowers (i and i  1) pollination. The step transition is limited
by Eq. (18) as:
n  b  n  b 2 
2
X V bat  Vex X 
MoFðuÞ ¼ w1  þ w2   S OC  soc  (7)
Vmax  socmax  Xmax  Xmin
i¼1 bat i¼1 dmax ¼ (18)
2Npop
where, w1 and w2 represent the weighting factors that reflect
the degree importance of F1 and F2 . where Xmin and Xmax are the boundaries values, Npop is the
population size of sunflowers. Applying the reduction strategy
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 8 8 3 3 e8 8 4 2 8837

aims to concentrate the search space and therefor enhance 1. Define the ESFOA parameters and building the state space
the overall solution quality. In this strategy, the search space model Eqs. (2) and (3).
is managed through adaptive variation of the upper and lower 2. Initialization process of the random position of nb flowers.
limits according to (19) and (20). Each sunflower represents the decision variables i:e the
target estimated parameters within their boundaries.
Xmin ¼ Xmin þ a*ðXmax  Xmin Þ (19)
3. Evaluation process of the initial positions to catch the best
solution S using Eqs. (5) and (6) for single objective cases
Xmax ¼ Xmax  a*ðXmax  Xmin Þ (20) and Eq. (7) for multi-objective case.
The factor a refers to the reduction coefficient that is 4. Employ the orientation of Npop sunflowers to the sun
applied in reduction strategy to manage the minimum and direction
maximum bounds. 5. While (k < Maxiteration)
1. Reduce the population size by m (%)
Proposed ESFOA-based solution methodology 2. Compute each plant step using state space model (2)e(3)
3. Re-evaluate the new position using Eqs. (5)e(7)
The procedure of sunflower solution methodology can be 4. For the global best individual update the sunflower second
mentioned as: level

Fig. 2 e (a) Laboratory test bench. (b) MMI interface for control and monitoring of the test bench.
8838 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 8 8 3 3 e8 8 4 2

(a) 50 classification, please refer to Refs. [45,46]. Before applying this

constraints classification, the common ARTEMIS driving cycle
Ic [A]

0
is chosen for identifying the parameters of the battery model
-50 by using a dynamics model of the urban electric vehicle
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
Time (s) (Bollore Bluecar).
(b)
Vbatt [V]

4
3.5
Driving cycle
expe
3 model
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 In the last few years, several normalized and non-normalized
Time (s) driving cycles have been recently proposed with the aim of
(c) 1 evaluating pollution emissions and fuel consumption of
gasoline-powered engines. One can cite for examples the
soc

0.5
UDC, ECE, NEDC, ARTEMIS, FTP75, NYCC and WLTC [47].
0
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 The representation of any driving cycle is generally based
Time (s) on speed-time sequences that represent the traffic conditions
Fig. 3 e Case 1 (a) Constant current profile with CC/CV and driving behavior in a specific area. For this study, the
charge protocol. (b) Experimental and modeling responses ARTEMIS driving cycle (Assessment and Reliability of Trans-
of battery voltage. (c) Experimental and modeling port Emission Models Inventory Systems) is selected. The
responses of battery SoC. latter is considered as a part of the main European project
which induced in the creation of a transient driving cycle
based on other European driving cycles [8]. It is divided into
6. Apply the reduction strategy to enhance the search space three phases (urban, rural and highway), which are repre-
and enhance the solution quality. The strategy is inte- sentative of real-world driving behavior. In this study, the
grated using Eqs. (19) and (20). In this paper the factor a ARTEMIS cycle chosen is represented by the urban and the
equals 0.08. rural cycle. Therefore, the distance covered is about 22 km in
7. For each plant compute the vector of the orientation 2075 s for an average speed of 38.4 km/h and a maximal speed
8. Check the stopping criteria of 111.5 km/h.
9. Print the identified variables
Laboratory test bench and battery technology

Design of experiment A test bench is used to carry out accelerated cycling aging tests
while trying to be as close as possible to the electric vehicle
In this context, we have laboratory experimental tests based application. The laboratory test bench is as shown in Fig. 2, the
on constraints classification in intensity levels. Its objective is bench includes two main components: a climatic chamber
to reduce cycling times while maintaining the most important and charge/discharge setup. The power that can be delivered
characteristic variables for test profiles, such as the maximum by this bench in continuous operation is 10 kW, by using a
current intensity and the amount of charge exchanged. The programmable power supply. The latter is connected to a
simplification adopted in this approach allows us to move reversible four-leg power converter (60V/600A). The FPGA
from a complex pulse current profile to a signal in the form of compact RIO NIcRIO-9012/9014 real-time control system with
a current step with well-defined current values. For more LabVIEW software is used for setting up the test procedure
details about the simplification of the real current profile, and for data acquisition [31,46]. Battery cells are placed in the
using the heuristic method based on constraints climatic chamber in order to perform the test at controlled
temperature conditions. These units can change the temper-
ature from 0 to 100  C and are able to change the temperatures
Table 1 e Characteristics Of Kokam Li-Ion Battery at a rate of 3 ± 0.02  C. The man-machine interface (MMI) in
Technologies [31]. Fig. 3 shows the front panel of RealTime (RT). It allows
Battery Kokam 40 HED entering the parameters related to control and acquisition, as
Nominal voltage (V) 3.7 well as the management of the safety of the bench (maximum
Internal resistance (m) 0.9 and minimum limits of the voltages and currents tolerated by
Capacity (Ah) 40 the components, maximum temperature limit … etc). This
Specific Energy (Wh/kg) 167 interface also allows us to visualize information on the state
Max current cha/dis (A) 40/40 of the bench (defects, type of defect, backup activation,
Weight (kg) 0.885

Table 2 e Sunflower optimization algorithm parameters.

Population Pollination Mortality survival Maximum
size rate rate rate iteration
50 0.1 0.005 0.9 40
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 8 8 3 3 e8 8 4 2 8839

number of cycles performed … etc) and to visualize the evo- (a) 50

lution of the electrical values and the temperatures of the

Ic [A]
0
components.
Different types of batteries have been recently designed for -50
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
electric and hybrid vehicles with the aim of improving the Time (s)
tradeoff between drive performance and high reliability [48]. (b) 4

Vbatt [V]
expe
As an example, a commercial 40 Ah Li-ion battery cell pro- 2 model
duced by Kokam manufacturer, under the references HED- 0
SLPB90216216 is used in our study. Table 1 gives the specifi- 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
cations of Li-ion battery cells, which present their most rele- Time (s)
vant specifications that can be found in the datasheet. (c) 1

soc
Fig. 6 (first figure) illustrates the cycling current profile 0.5
150 km ARTEMIS driving cycle (setpoint current) super- 0
imposed on the real current profile by using the tested battery, 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
Time (s)
Kokam Ko40HE. With the heuristic method based on con-
straints classification, the duration of cycles is reduced by 35% Fig. 4 e Case 2:(a) Constant current profile with CC/CV
of the real profiles. As for the recharge phase, it occurs when charge protocol. (b) Experimental and modeling responses
the desired depth of discharge is obtained. This recharging of battery SoC.
reduces the state of charge of the batteries to 100%. The
recharge follows the protocol defined by the manufacturer
Kokam. This profile consists of a CC/CV (Constant Current/
Constant Voltage) recharge, preceded by a rest phase. In this of the proposed ESFOA. Table 3 presented the boundaries of
study, the used battery has been designed for a range of the estimated battery parameters according the experimental
150 km. The cycling profiles are obtained by using the con- tests. The discharge with constant current and CC/CV charge
straints classification method and it was repeated seven times protocol is used to record the experimental voltage data of
to correspond to the 150 km distance and to obtain a depth of battery. In addition, the change of SoC during the test has
discharge of 80%. been recorded with a coulomb-accumulation method based
on the history of current, including self-discharging and cur-
Evaluation of the estimated results rent inefficiency on charge. As mentioned above three cases
are studied. In the first case, the primary objective function
In this section, the simulation results obtained by applying the aims at minimizing the deviation between the estimated and
proposed ESFOA are assessed with those recorded in the recorded battery voltages. According to Table 4, Figs. 3 and 6
experimental results and two other optimization algorithms shows the battery performance for two tested driving cycles.
called water-cycle (WCOA) and whale (WOA). The WCOA pa- It is cleared that: the high closeness between the estimated
rameters are customized from Ref. [49] while the parameters and measured variables are achieved for both battery voltages
of WOA are taken from Ref. [50]. Table 2 shows the parameters and the battery state of charge. The primary objective func-
tion reaches 5.23  105 while the second objective reaches
6.58  106. By the transition to Case 2, the minimizing of SOC
is considered as the primary objective function while the
Table 3 e Identified battery parameters. battery voltage is considered as secondary objective. In Case 2,
high closeness is noticed between the estimated and
R ðmUÞ Ci ðFÞ Ri ðUÞ b0 (V) b1 (V) Qr
measured state of charge for both tested driving cycles. The
LB 3 4000 1000 3 25 130000
simulation results of the estimated SoC is presented in Table
UB 4 5000 1300 5 40 170000

Table 4 e Identified battery parameters Case 1.

R Ci Ri b0 b1 Qr F1 F2 Fitness
WCOA 3.77 4396.46 1110.30 4.24 34.40 147320.27 5.23 105 6.58 106 5.23 105
WOA 3.38 4306.56 1001.49 3.72 31.75 135999.34 3.3 104 2.6 104 3.3 104
ESFOA 3.77 4396.46 1110.30 4.24 34.40 147320.27 5.23 105 6.58 106 5.23 105

Table 5 e Identified battery parameters Case 2.

R Ci Ri b0 b1 Qr F1 F2 Fitness
WCOA 3.65 4529.28 1101.74 3.72 30.35 147963.81 5.7 104 1.68 107 1.68 107
WOA 3.81 4586.68 1183.69 4.55 35.37 147963.81 6.2 104 5.82 107 5.82 107
ESFOA 3.77 4396.46 1110.30 4.55 34.40 147320.27 7.8 105 2.1 108 2.1 108
8840 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 8 8 3 3 e8 8 4 2

Table 6 e Identified battery parameters Case 3.

R Ci Ri b0 b1 Qr F1 F2 Fitness
WCOA 3.77 4396.46 1110.30 4.24 34.40 147320.27 6.23105 7.12 107 6.24 105
WOA 3.66 4515.54 1084.95 3.72 32.38 145804.00 3.5 104 7.39 106 3.6 104
ESFOA 3.66 4634.52 1107.23 4.13 32.71 1.4797 10 6.23 105 3.5 108 6.23 5

(a) 50
Ic [A]

-50
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
Time (s)
(b)
Vbatt [V]

4 expe
3.5 model
3
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
Time (s)
1
(c)
soc

0.5

0
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
Time (s)

Fig. 5 e Case 3 (a) Constant current profile with CC/CV Fig. 7 e Case 2:(a) Laboratory current profile with 150 km
charge protocol.(b) Experimental and modeling responses ARTEMIS driving cycle. . (b) Experimental and modeling
of battery voltage. (c) Experimental and modeling responses of battery SoC.
responses of battery SoC.

Fig. 8 e Case 3 (a) Laboratory current profile with 150 km

Fig. 6 e Case 1 (a) Laboratory current profile with 150 km ARTEMIS driving cycle. (b) Experimental and modeling
ARTEMIS driving cycle. (b) Experimental and modeling responses of battery voltage. (c) Experimental and
responses of battery voltage. (c) Experimental and modeling responses of battery SoC.
modeling responses of battery SoC.

From Tables 4e6, it is concluded that the ESFOA leads to

5, Figs. 4 and 7. The primary objective function reaches 7.8  the lowest fitness functions (in the range 60e90%) and
105 while the secondary objective function (battery voltage) competitive parameter identification for the three cases
reaches 2.1  108. A compromise solution is obtained be- compared with the WCOA and WOA.
tween the considered two objective functions, battery voltage
and state of charge, in Case 3. The two objective functions are
simultaneously optimized for both tested driving cycles. The Conclusion
simulation results are presented in Table 6, Figs. 5 and 8. The
considered objective functions reaches 6.23  105 and 3.5  This paper has been proposed a n enhanced sunflower opti-
108, respectively. mization algorithm for solving the problem of parameter
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 8 8 3 3 e8 8 4 2 8841

identification and state of charge estimation of Lithium Poly- 10.1016/j.ijhydene.2019.01.190. URL, http://www.
mer Battery Cells. An electrical dynamic equivalent circuit sciencedirect.com/science/article/pii/S0360319919303490.
based on state space model has been presented. The problem [8] Liao Q, Mu M, Zhao S, Zhang L, Jiang T, Ye J, Shen X, Zhou G.
Performance assessment and classification of retired lithium
of parameter identification is formulated as non-linear opti-
ion battery from electric vehicles for energy storage. Int J
mization problem. Multi-objective sunflower optimization Hydrogen Energy 2017;42(30):18817e23.
algorithm is developed to simultaneously optimize two [9] Zhan L, Zhou X, Luo J, Ning X. Ion assisted anchoring sn
objective functions and to obtain the optimal parameter nanoparticles on nitrogen-doped graphene as an anode for
identification and estimate the state of charge. Both objectives lithium ion batteries. Int J Hydrogen Energy
aim at achieve the high closeness between experimental and 2019;44(45):24913e21. https://doi.org/10.1016/
estimated values for voltage and state of charge of the j.ijhydene.2019.07.153. URL, http://www.sciencedirect.com/
science/article/pii/S0360319919327296.
considered battery. This paper proves the ability of the pro-
[10] Bendjedia B, Rizoug N, Boukhnifer M, Bouchafaa F,
posed sunflower optimization algorithm as accurate method Benbouzid M. Influence of secondary source technologies
for obtaining the optimal battery parameters. The proposed and energy management strategies on energy storage
parameter identification procedure has been validated for system sizing for fuel cell electric vehicles. alternative
constant and varied current profiles, that is reflected real Energies for Sustainability Int J Hydrogen Energy
bench-marking driving cycles. The error levels among simu- 2018;43(25):11614e28. https://doi.org/10.1016/
j.ijhydene.2017.03.166. URL, http://www.sciencedirect.com/
lation results reflect high closeness between the estimated
science/article/pii/S0360319917311722.
and experimental data.
[11] Latroche M, Blanchard D, Cuevas F, Kharbachi AE,
Hauback BC, Jensen TR, de Jongh PE, Kim S, Nazer NS,
Ngene P, ichi Orimo S, Ravnsbæk DB, Yartys VA. Full-cell
hydride-based solid-state li batteries for energy storage. a
Acknowledgment
special issue on hydrogen-based Energy storage Int J
Hydrogen Energy 2019;44(15):7875e87.
Dr. Ragab A. El Sehiemy would like to acknowledge the sup- [12] Claude F, Becherif M, Ramadan H. Experimental validation
port provided by STDF-IFE collaboration through their funded for li-ion battery modeling using extended kalman filters. Int
project 31161. The authors would like to express their appre- J Hydrogen Energy 2017;42(40):25509e17.
ciation to Dr, Nassim Rizoug, S2ET-ESTACA-Laval, France for [13] Kutluay K, Cadirci Y, Ozkazanc YS, Cadirci I. A new online
providing technical assistance and battery experimental tests. state-of-charge estimation and monitoring system for sealed
lead-acid batteries in telecommunication power supplies.
IEEE Trans Ind Electron 2005;52(5):1315e27. https://doi.org/
10.1109/TIE.2005.855671.
references
[14] Dang X, Yan L, Jiang H, Wu X, Sun H. Open-circuit voltage-
based state of charge estimation of lithium-ion power
battery by combining controlled auto-regressive and
[1] Wu P, Bucknall R. Hybrid fuel cell and battery propulsion moving average modeling with feedforward-feedback
system modelling and multi-objective optimisation for a compensation method. Int J Electr Power Energy Syst
coastal ferry. Int J Hydrogen Energy 24 January 2017;90:27e36.
2020;45(4):3193e208. https://doi.org/10.1016/ [15] Watrin N, Roche R, Ostermann H, Blunier B, Miraoui A.
j.ijhydene.2019.11.152. URL, http://www.sciencedirect.com/ Multiphysical lithium-based battery model for use in state-
science/article/pii/S0360319919343757. of-charge determination. IEEE Trans Veh Technol
[2] Lin J, Yu L, Sun Q, Wang F, Cheng Y, Wang S, et al. 2012;61(8):3420e9.
Multiporous core-shell structured mno@n-doped carbon [16] Waag W, Kabitz S, Sauer DU. Experimental investigation of
towards high-performance lithium-ion batteries. Int J the lithium-ion battery impedance characteristic at various
Hydrogen Energy 13 January 2020;45(3):1837e45. https:// conditions and aging states and its influence on the
doi.org/10.1016/j.ijhydene.2019.11.083. URL, http://www. application [special Issue on Advances in sustainable biofuel
sciencedirect.com/science/article/pii/S036031991934265X. production and use - XIX International Symposium on
[3] Milocco R, Thomas J, Castro B. Robust polynomial approach Alcohol Fuels - ISAF] Appl Energy 2013;102:885e97.
for state of charge estimation in nimh batteries. Int J [17] Waag W, Sauer DU. Adaptive estimation of the electromotive
Hydrogen Energy 2012;37(13):10371e5. force of the lithium-ion battery after current interruption for
[4] Boudoudouh S, Maa ^ roufi M. Real-time battery state of charge an accurate state-of-charge and capacity determination.
estimation in smart grid application by multi agent system. Appl Energy 2013;111:416e27.
Int J Hydrogen Energy 2017;42(30):19487e95. https://doi.org/ [18] Li W, Jiao Z, Du L, Fan W, Zhu Y. An indirect rule prognosis
10.1016/j.ijhydene.2017.06.077. URL, http://www. for lithium-ion battery under vibration stress using elman
sciencedirect.com/science/article/pii/S0360319917323534. neural network. Int J Hydrogen Energy 2019;44(23):12270e6.
[5] Frost DF, Howey DA. Completely decentralized active https://doi.org/10.1016/j.ijhydene.2019.03.101. URL, http://
balancing battery management system. IEEE Trans Power www.sciencedirect.com/science/article/pii/
Electron 2017;33(1):729e38. S0360319919311073.
[6] Ramadan H, Becherif M, Claude F. Extended kalman filter for [19] Zhang S, Yang L, Zhao X, Qiang J. A gas optimization for
accurate state of charge estimation of lithium-based lithiumeion battery equalization based on soc estimation by
batteries: a comparative analysis. Int J Hydrogen Energy nn and flc. Int J Electr Power Energy Syst 2015;73:318e28.
2017;42(48):29033e46. [20] Zhang Z, Cheng X, Lu Z, Gu D. Soc estimation of lithium-ion
[7] Yue M, Jemei S, Gouriveau R, Zerhouni N. Review on health- battery pack considering balancing current. IEEE Trans
conscious energy management strategies for fuel cell hybrid Power Electron 2018;33(3):2216e26.
electric vehicles: degradation models and strategies. Int J [21] Meng J, Ricco M, Luo G, Swierczynski M, Stroe D-I, Stroe A-I,
Hydrogen Energy 2019;44(13):6844e61. https://doi.org/ Teodorescu R. An overview and comparison of online
8842 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 4 5 ( 2 0 2 0 ) 8 8 3 3 e8 8 4 2

implementable soc estimation methods for lithium-ion [36] Little M, Thomson M, Infield D. Electrical integration of
battery. IEEE Trans Ind Appl 2017;54(2):1583e91. renewable energy into stand-alone power supplies
[22] Wei Z, Zou C, Leng F, Soong BH, Tseng K-J. Online model incorporating hydrogen storage. Int J Hydrogen Energy
identification and state-of-charge estimate for lithium-ion 2007;32(10e11):1582e8.
battery with a recursive total least squares-based observer. [37] Han Y, Chen W, Li Q, Yang H, Zare F, Zheng Y. Two-level
IEEE Trans Ind Electron 2017;65(2):1336e46. energy management strategy for pv-fuel cell-battery-based
[23] Zhang W, Wang L, Wang L, Liao C. An improved adaptive dc microgrid. Int J Hydrogen Energy 2019;44(35):19395e404.
estimator for state-of-charge estimation of lithium-ion [38] Lin R, Liu D, Xia S, Ma T, Dutruel B. Stack shut-down strategy
batteries. J Power Sources 2018;402:422e33. optimisation of proton exchange membrane fuel cell with
[24] Tang X, Liu B, Lv Z, Gao F. Observer based battery soc the segment stack technology. Int J Hydrogen Energy
estimation: using multi-gain-switching approach. Appl 2020;45(1):1030e44.
Energy 2017;204:1275e83. [39] Gomes GF, da Cunha SS, Ancelotti AC. A sunflower
[25] Zhang C, Wang LY, Li X, Chen W, Yin GG, Jiang J. Robust and optimization (SFO) algorithm applied to damage
adaptive estimation of state of charge for lithium-ion identification on laminated composite plates. Eng Comput
batteries. IEEE Trans Ind Electron 2015;62(8):4948e57. 2019. https://doi.org/10.1007/s00366-018-0620-8.
[26] Chaoui H, Ibe-Ekeocha CC. State of charge and state of health [40] Qais MH, Hasanien HM, Alghuwainem S. Identification of
estimation for lithium batteries using recurrent neural electrical parameters for three-diode photovoltaic model
networks. IEEE Trans Veh Technol 2017;66(10):8773e83. using analytical and sunflower optimization algorithm.
[27] Alvarez Anton JC, Garcia Nieto PJ, Blanco Viejo C, Vilan Applied Energy 2019. https://doi.org/10.1016/
Vilan JA. Support vector machines used to estimate the j.apenergy.2019.05.013.
battery state of charge. IEEE Trans Power Electron [41] Lai X, Gao W, Zheng Y, Ouyang M, Li J, Han X, Zhou L. A
2013;28(12):5919e26. comparative study of global optimization methods for
[28] Anton JCA, Nieto PJG, Gonzalo EG, Pe rez JCV, Vega MG, parameter identification of different equivalent circuit
Viejo CB. A new predictive model for the state-of-charge of a models for li-ion batteries. Electrochim Acta
high-power lithium-ion cell based on a pso-optimized 2019;295:1057e66.
multivariate adaptive regression spline approach. IEEE Trans [42] Orcioni S, Buccolini L, Ricci A, Conti M. Lithium-ion battery
Veh Technol 2015;65(6):4197e208. electrothermal model, parameter estimation, and simulation
[29] Song K, Li F, Hu X, He L, Niu W, Lu S, Zhang T. Multi-mode environment. Energies 2017;10(3):375.
energy management strategy for fuel cell electric vehicles [43] Rahimi-Eichi H, Baronti F, Chow M. Online adaptive
based on driving pattern identification using learning vector parameter identification and state-of-charge coestimation
quantization neural network algorithm. J Power Sources for lithium-polymer battery cells. IEEE Trans Ind Electron
2018;389:230e9. 2014;61(4):2053e61.
[30] Alvarez Anton JC, Garcia Nieto PJ, Garcia Gonzalo E, Viera [44] Shaheen MAM, Hasanien HM, Mekhamer SF, Talaat HEA.
Perez JC, Gonzalez Vega M, Blanco Viejo C. A new predictive Optimal power flow of power systems including distributed
model for the state-of-charge of a high-power lithium-ion generation units using sunflower optimization algorithm.
cell based on a pso-optimized multivariate adaptive IEEE Access 2019;7:109289e300. https://doi.org/10.1109/
regression spline approach. IEEE Trans Veh Technol ACCESS.2019.2933489.
2016;65(6):4197e208. [45] Rizoug N, Sadoun R, Mesbahi T, Bartholumeus P, LeMoigne P.
[31] Mesbahi T, Khenfri F, Rizoug N, Chaaban K, Bartholomeus P, Aging of high power li-ion cells during real use of electric
Moigne PL. Dynamical modeling of li-ion batteries for electric vehicles. IET Electr Syst Transp 2017;7(1):14e22. https://
vehicle applications based on hybrid particle swarm-Nelder- doi.org/10.1049/iet-est.2016.0012.
Mead (pso-nm) optimization algorithm. Elec Power Syst Res [46] Mesbahi T, Rizoug N, Bartholomeus P, Sadoun R, Khenfri F,
2016;131:195e204. Le Moigne P. Dynamic model of li-ion batteries incorporating
[32] Kumar N, Hussain I, Singh B, Panigrahi BK. Single sensor- electrothermal and ageing aspects for electric vehicle
based mppt of partially shaded pv system for battery applications. IEEE Trans Ind Electron 2018;65(2):1298e305.
charging by using cauchy and Gaussian sine cosine https://doi.org/10.1109/TIE.2017.2714118.
optimization. IEEE Trans Energy Convers 2017;32(3):983e92. [47] Torres J, Gonzalez R, Gimenez A, Lopez J. Energy
[33] Chenouard R, El-Sehiemy RA. An interval branch and bound management strategy for plug-in hybrid electric vehicles. a
global optimization algorithm for parameter estimation of comparative study. Appl Energy 2014;113:816e24.
three photovoltaic models. Energy Convers Manag [48] Affanni A, Bellini A, Franceschini G, Guglielmi P, Tassoni C.
2020;205:112400. https://doi.org/10.1016/ Battery choice and management for new-generation electric
j.enconman.2019.112400. URL, http://www.sciencedirect. vehicles. IEEE Trans Ind Electron 2005;52(5):1343e9.
com/science/article/pii/S0196890419314074. [49] El-Ela AAA, El-Sehiemy RA, Abbas AS. Optimal placement
[34] Sehiemy RAE, Selim F, Bentouati B, Abido M. A novel multi- and sizing of distributed generation and capacitor banks in
objective hybrid particle swarm and salp optimization distribution systems using water cycle algorithm. IEEE Sys J
algorithm for technical-economical-environmental 2018;12(4):3629e36.
operation in power systems. Energy 2020;193:116817. https:// [50] Algabalawy MA, Abdelaziz AY, Mekhamer SF, Aleem SHA.
doi.org/10.1016/j.energy.2019.116817. URL, http://www. Considerations on optimal design of hybrid power
sciencedirect.com/science/article/pii/S0360544219325125. generation systems using whale and sine cosine
[35] El-Ela AAA, El-Sehiemy RA, Ali ES, Kinawy A-M. optimization algorithms. J Elect Syst Inf Technol
Minimisation of voltage fluctuation resulted from renewable 2018;5(3):312e25.
energy sources uncertainty in distribution systems, IET
Generation. Transm Distrib 2019;13(12):2339e51.