Proceedings of the 2023 IEEE 11th International TuCC.

2
Conference on Systems and Control, Sousse, Tunisia,
December 18-20, 2023

Identification of equivalent circuit model parameters

for a Li-ion battery cell
Filip Rukavinaa , Dorijan Lekoa , Matija Matijašićb , Ivan Bralićb , Juan M. Ugaldeb , and Mario Vašaka
a
Laboratory for Renewable Energy Systems,
University of Zagreb Faculty of Electrical Engineering and Computing, Zagreb, Croatia
Email: {filip.rukavina, dorijan.leko, mario.vasak}@fer.hr

b
Battery Systems Engineering, Rimac Technology d.o.o., Sveta Nedelja, Croatia
2023 IEEE 11th International Conference on Systems and Control (ICSC) | 979-8-3503-0488-6/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICSC58660.2023.10449745

Email: {matija.matijasic, ivan.bralic, juan.ugalde}@rimac-technology.com

Abstract—Batteries raise issues in electric vehicles because model of the battery to compute control actions. Naturally,
of their limited range and lifespan. They can however be model accuracy is crucial for the correct operation of the
extended by implementing smart, accurate and computationally control algorithm [5].
inexpensive control algorithms. Herein, a lithium-ion battery cell
is modelled with an equivalent circuit model which is complex This work focuses on modelling a Li-ion battery cell as it is
enough to track the output voltage of the cell while remaining the most common technology in EVs. An accurate model of a
simple enough to be used in a control algorithm. This work battery can be achieved by using Doyle-Fuller-Newman (DFN)
presents the identification procedure of equivalent circuit model model [6], or single particle model (SPM) [7], but such models
parameters for a lithium-ion battery cell by applying high are hard to parametrize based on readily available data, and
currents to the cell. Consequently, high accuracy was achieved
for the whole range of possible currents. The parameters were computationally too expensive to run in real-time on embedded
identified on a dataset with high currents, while the model was systems found in EVs. A more suitable model for real-time
evaluated on a different dataset acquired from a real electric implementation of a battery-related control algorithm is the
vehicle usage data. Metrics from both the identification and the equivalent circuit model (ECM) composed of purely electrical
validation datasets are similar meaning that the model generalises components. The most common examples of ECMs found in
well.
Index Terms—parameter identification, optimization, equiva- literature [5], [8]–[12] are composed of an open circuit voltage
lent circuit, modelling, Li-ion, battery, EV (OCV), a resistor, and a number of RC elements connected in
series.
I. I NTRODUCTION In [5], authors developed a procedure to identify state of
In the current environment of energy efficiency, zero carbon charge (SoC) - OCV curve from just one charging/discharging
emissions, and environment protection, electric vehicles (EVs) cycle and explored various methods to approximate the SoC-
are seen as the future of transportation. They are strongly OCV curve. Cipin et al. [8] compared the first- and the second-
advocated by policy makers such as European Union (EU), order models and discussed the main problems with offline
which brought a directive on their use [1]. EVs have plenty parameter identification being: nonlinear dependence of the
advantages compared to traditional internal combustion ve- OCV and the internal resistance on the SoC of the battery,
hicles, but their biggest drawback is their battery autonomy, dependence of the battery parameters on temperature, and for
which is one of the critical factors for their integration [2]. some chemistries, hysteresis behaviour of the OCV during
There are several strategies and approaches to increase the charging and discharging. In [9], the focus was on obtaining
share of EVs, where most of them focus on decreasing cost, data for the model parameter identification by applying con-
increasing efficiency, and increasing lifespan of their batteries tinuous pulsating excitation to the battery cell. The authors
[3]. This can be achieved by implementing new technologies, showed that such method can greatly reduce the identification
but adequate control techniques and battery management sys- time while ensuring accuracy. Wu et al. [10] identified the
tems (BMSs) can be used to improve both new and existing parameters using offline hybrid pulse power characterization
battery technologies [4]. One of such techniques is model (HPPC) test cycle and went further to develop an online
predictive control (MPC) which relies on a mathematical model identification procedure to increase accuracy of the
offline model by implementing forgetting factor. Similarly to
This work was co-financed by the European Union from the European the previous research, authors of [11] developed first- and
Regional Development Fund via Operative Programme Competitiveness and second-order RC models under various charge states, but they
Cohesion 2014-2020 for Croatia within the project Dynamic Predictive
Health Protection of an Electric Vehicle Battery (EVBattPredtect, contract did it for a Li-sulphur battery. They concluded that both
no. KK.01.1.1.07.0029). models performed proficiently with respect to constant current

979-8-3503-0488-6/23/$31.00 ©2023 IEEE 671

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on June 14,2024 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.
 discharge. Yutthanava et al. [12] also used second-order RC variables of the model are diffuse voltages u1 and u2 defined
model identified by using HPPC experiments. in continuous time domain with ODEs in (1).
From the previous research it can be concluded that the d 1 1
second-order RC model is accurate enough to model a lithium u1 (t) = − u1 (t) + i(t) (1a)
dt R1 C1 C1
based battery cell for control purposes, therefore it is used
in this work as well. The biggest novelty in this approach d 1 1
u2 (t) = − u2 (t) + i(t) (1b)
is the dataset applied for parameter identification. Usually, dt R2 C2 C2
when model parameters are identified on a dataset with low
The continuous model is then discretised with zero-order
currents, the model performs poorly when high currents are
hold (ZOH) method giving (2),
applied. Therefore, during the data acquisition for parameter  
1 1
identification the battery cell was subjected to currents higher u1 (k + 1) = e− T1 Ts u1 (k) + R1 1 − e− T1 Ts i(k) (2a)
than the manufacturer’s recommendations both in charging and
discharging, in order to capture accurate behaviour for any 1
 1

current applied to the cell. Furthermore, the identified model u2 (k + 1) = e− T2 Ts u2 (k) + R2 1 − e− T2 Ts i(k) (2b)
is evaluated on a dataset not used for parameter identification where T1 = R1 C1 and T2 = R2 C2 , while k represents time
which was obtained from real-world EV usage. The identified instance tk of measurements u(k) = u(tk ) and i(k) = i(tk ).
model can be then used in a battery management system Ts represents sampling time or rather time difference between
(BMS) to ensure battery longevity while maximizing the two consecutive time instances in which the measurements
current ranges the battery can be charged/discharged with. were taken i.e. Ts = Ts (k) = tk+1 −tk . This kind of modelling
The paper is organized as follows. The ECM is defined and allows for parameter identification from measurements with
described in Section II. Then, the optimization procedure for varying sampling time, which can greatly lower data storage
parameter identification is presented in Section III. In Section requirements. For identification purposes it is assumed that the
IV, the battery cell and the measurement data are presented current i(k) is constant on the period [tk , tk+1 ). Estimation of
along with the evaluation of the identified model on both voltage u(k) is done with (3), where û(k) denotes estimated
identification and validation data. Lastly, the whole work is voltage rather than actual voltage u(k).
concluded in Section V where options for future work are
discussed. û(k) = uOCV − u1 (k) − u2 (k) − R0 i(k) (3)

All ECM parameters are dependent on SoC and the

II. E QUIVALENT CIRCUIT MODEL current direction [13], i.e. p = f (SoC, sgn(i)) , ∀p ∈
{uOCV , R0 , R1 , T1 , R2 , T2 }. Therefore, every parameter p is
The ECM of a lithium based battery cell is depicted in defined with two lookup tables, one for charging pchg (SoC)
Fig. 1. It consists of a source of an open circuit voltage (OCV) and one for discharging pdch (SoC). Each lookup table con-
uOCV , an internal resistance R0 , and two RC circuits which tains m parameter values corresponding to m predefined SoC
makes this model a second-order RC model. Every RC circuit 
data points SoC1DP , SoC2DP , ..., SoCm DP
. To get parameter
y ∈ {1, 2} is composed of a resistance Ry and capacitance values for the whole range of SoC, linear interpolation between
Cy , having a diffuse voltage uy across it. Positive value of the two adjacent SoC data points is used.
current represents discharging while negative value represents Consequently, the value of SoC ∈ [0, 1] must be estimated
charging of the battery cell. which is done with the direct measurement method not based
on the battery model [14], given in (4),
R1 R2
Ts
R0 i SoC(k + 1) = SoC(k) − i(k) (4)
C
C1 C2 where C represents the battery cell capacity in [As]. Even
+ though the nominal cell capacity is provided by the manufac-
+ u OCV + u1 – + u2 – u
turer, its actual value might differ slightly. Also, the battery cell
parameters might be identified after some use which degrades
– the capacity. Therefore, the value of C is identified along with
– the other parameters.

III. PARAMETER IDENTIFICATION PROCESS

Fig. 1. Equivalent circuit model of the battery cell. The battery cell whose parameters are being identified is
Samsung INR21700-48X [15] whose nominal parameters are
Measured values are charging/discharging current i and shown in Table I.
voltage at the battery cell terminals u, which also represent The overall identification process of ECM parameters is
input and output variables of the model, respectively. State carried out with these steps:

979-8-3503-0488-6/23/$31.00 ©2023 IEEE 672

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on June 14,2024 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.
 TABLE I 4.2
N OMINAL PARAMETERS OF S AMSUNG INR21700-48X BATTERY CELL 4
[15].
3.8

Voltage uOCV [V]

Specification Items 21700-48X
3.6 discharging
charging
Typical Energy (4.2V, 0.2C discharge) 17.47 Wh 3.4
Typical Capacity (4.2V, 0.33C charge) 4800 mAh 3.2
Energy Density (Typical) 717 Wh/l
3
General Energy Density (Typical) 256 Wh/kg
2.8
IR @ 25°C (AC / DC) 13 mΩ / 22 mΩ
Weight (Max) 69.5 g 2.6
2.5
Nominal Voltage 3.64 V 0 0.2 0.4 0.6 0.8 1
SoC
Charging Voltage 4.2 V
Standard Charging Current 0.33 C Fig. 2. SoC-OCV curve of the battery cell at 25°C.
Charge
Max Charging Current
2.0 C (9.6 A)
(not continuous)
Max Charging Current (continuous) 1.0 C (4.8 A) original data points, as in (5),
chg chg
Discharging End Voltage 2.5 V error = ∥udch dch
OCV − ûOCV ∥2 + ∥uOCV − ûOCV ∥2 (5)
Standard Discharging Current 0.2 C
Discharge where uOCV represents values obtained from the battery cell
Max Discharging Current
3.0 C (14.4 A) manufacturer, while ûOCV represents approximated values.
(not continuous)
Max Discharging Current (continuous) 1.0 C (4.8 A) Exponents dch and chg correspond to discharging and charg-
Cycle Life (1.0C 4.15V charge 70% ing curves, respectively.
Life
/ 1.0C 3V discharge) @ 3000 cycle The number of SoC data points chosen for the identification
process is 22 as an increase in the number of data points does
not reduce the approximation error significantly. Errors for all
1) The SoC-OCV curves obtained from the cell’s manufac- numbers of SoC data points is displayed in Fig. 3.
turer are simplified to lower computational complexity,
2
but with minimal approximation error.
OCV aproximation error

2) The dataset used for parameter identification, current i 1.5

and voltage u, is acquired through measurements.
3) Optimization problem is defined in terms of optimization 1
variables, cost function, and constraints. The optimiza-
tion problem is then solved by employing a suitable 0.5
algorithm.
4) Resulting model parameters are evaluated on the valida- 0
tion dataset.
0 5 10 15 20 25 30
Number of SoC data points, m
A. SoC-OCV curves
Fig. 3. Approximation error of SoC-OCV curves depending on the number
of SoC data points chosen.
OCV could be identified with the rest of the parameters,
but it could also be its own process since there are several
approaches to identify it as it is pointed out in [16]. However, B. Data acquisition
SoC-OCV curves were provided from the manufacturer and The data used for parameter identification process are taken
they are shown in Fig. 2. in laboratory conditions, but the currents applied to the cell
The OCV curves in Fig. 2 are defined with a lookup table have been measured during intense driving of an EV and
having uOCV values for every 0.005 of SoC meaning that charging when needed with high current. Such sequence is
every curve is defined with m = 201 data points. Conse- repeated to obtain a current profile which corresponds to driv-
quently, every parameter of the ECM would have 402 values ing 5000 km and lasts around 14 days. The measured current
which would make the optimization process computationally and voltage are presented in Fig. 4. During the experiment, the
too expensive. Therefore, an approximation of the SoC-OCV battery cell was immersed in a dielectric oil kept at constant
curves is made for different numbers of SoC data points temperature of 25°C. Although the conditions in the EV are
ranging from 2 to 30, i.e. for m ∈ {2, 3, ..., 30}. Values of different than laboratory conditions, the battery compartment
SoC data points SoC DP are chosen by minimizing the error in the EV has its own cooling system to keep the batteries at
between the actual values and approximated values for all 201 an optimal temperature.

979-8-3503-0488-6/23/$31.00 ©2023 IEEE 673

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on June 14,2024 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.
 80 4) the parameter values form a convex function across
Cell current, i
60 SoC domain, i.e. their second derivative must be non-
Current [A]

40 negative, as in (8) ∀p ∈ {R0 , R1 , T1 , R2 , T2 }, both for

20 discharging and charging.
0 p(SoCsDP ) − p(SoCs−1
DP
) p(SoCs+1DP
) − p(SoCsDP )
DP DP
≤ DP
,
-20 SoCs − SoCs−1 SoCs+1 − SoCsDP (8)
2 4 6 8 10
∀s ∈ {2, ..., 21}
Time [s] #105

4.5
The values of ECM parameters differ greatly from one
Cell voltage, u
another, meaning that their values can be a few orders
4
Voltage [V]

of magnitude apart which can cause numerical instabil-

3.5 ity in the optimization. Therefore, every parameter p ∈
{C, R0 , R1 , T1 , R2 , T2 } is defined as in (9),
3
p = pfix popt (9)
2.5
2 4 6 8 10
where pfix is a predefined expected value of the parameter p,
Time [s] #105
while popt is a part that is being optimized inside the opti-
Fig. 4. Measured voltage and current for parameter identification. mization vector x. This way, all the variables being optimized
will be of the same order of magnitude.

C. Optimization IV. M ODEL EVALUATION

The parameters of the ECM are identified by minimizing After the optimization, the resulting parameter values are
the cost function (6), presented in Fig. 5 and Fig. 6.
N −1 2
0.2
X
J= u(k) − û(k) (6)
discharging, R0dch
k=0 charging, R0chg
Resistance R0 [+]

0.15
where û(k) is the estimated voltage at time instance k obtained
from the ECM i.e. expressions (2, 3, 4). As the estimated 0.1
voltage is a nonlinear function, nonlinear optimizer fmincon
from Matlab was used. However, the constraints used are 0.05
linear, so the overall optimization problem is defined in (7),
arg min J(x) 0
x 0 0.2 0.4 0.6 0.8 1
s. t. Aineq x ≤ bineq (7) SoC
xmin ≤ x ≤ xmax Fig. 5. Resulting parameter R0 after optimization.
where x is the optimization vector comprised of:
• the ECM parameters for discharging and charging
Resistance R1 [+]

Resistance R2 [+]

0.1 0.1
pdch , pchg , ∀p ∈ {R0 , R1 , T1 , R2 , T2 }; discharging, R1dch discharging, R2dch
charging, R1chg charging, R2chg
• capacity of the battery cell C; 0.05 0.05

• initial conditions SoC(0), u1 (0), and u2 (0).

0 0
Separation of discharging and charging parameter values 0 0.5 1 0 0.5 1
is done by adding exponents dch and chg to parameters. SoC SoC
Constraints being applied in the form (7) ensure that: 300 discharging, T1dch 150
discharging, T2dch
Time T1 [s]

Time T2 [s]

1) there is a clear separation between slow and fast RC charging, T1chg

200 100 charging, T2chg
circuits or rather their time 2 T2 (SoC) ≤
 parameters, i.e.DP
T1 (SoC), ∀SoC ∈ SoC1DP , ..., SoC22 for both 100 50

discharging and charging; 0 0

0 0.5 1 0 0.5 1
2) the resistances of RC circuits are lower than the internal SoC SoC

 ≤ R0 (SoC) andDPR2 (SoC) ≤

resistance, i.e. R1 (SoC)
Fig. 6. Resulting parameters R1 , R2 , T1 , and T2 after optimization.
R0 (SoC), ∀SoC ∈ SoC1DP , ..., SoC22 for both
discharging and charging; The resulting parameters are validated on a separate dataset
3) the battery cell capacity does not exceed plausible value, visible in Fig. 7 which was acquired in the same way as
i.e. C ≤ 4.9 Ah; the identification set: in laboratory conditions with constant

979-8-3503-0488-6/23/$31.00 ©2023 IEEE 674

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on June 14,2024 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.
 temperature, and the currents applied to the cell were measured for SoC ∈ [0.4, 1] and used second order polynomials as
during driving of an EV and multiple charging intervals. functions for model parameters’ SoC-dependence. Neverthe-
However, in this case the driving was normal without currents less, the model developed in [17] is similar enough to be
high as in the identification dataset. The capacity of the battery compared with the model in this work. Therefore, having
cell amounts to 4809.4 mAh which is slightly higher than its MAPE of 0.39% and 0.37% for identification and validation
nominal value, but it is still a good fit since the actual capacity datasets shows that the model developed here has pretty good
can be slightly higher than the nominal one. performance.
A detail from the validation dataset is presented in Fig. 8.
40
The detail shows current applied to the cell, measured voltage
u against estimated voltage û, model error u − û, slow u1
Current [A]

Cell current, i
20 and fast u2 diffuse voltages, and SoC across time period of
1000 s. Since the whole validation dataset is about 3.5 × 105
0 s long, only a detail of 1000 s is shown in order to have
higher resolution. The detail chosen shows model performance
0.5 1 1.5 2 2.5 3 3.5 during driving which includes regenerative breaking (negative
Time [s] #105
values of i). It can be seen that the model estimates measured
4.5 voltage well during dynamic current changes, and the model
error is symetrical around 0. Lastly, it can be seen that the
4
Voltage [V]

fast diffuse voltage u2 has significantly faster dynamics than

3.5 the slow diffuse voltage u1 , which shows that the separation
3 between slow and fast diffuse voltages was successful.
Cell voltage, u
2.5 V. C ONCLUSION AND FUTURE WORK
0.5 1 1.5 2 2.5 3 3.5
Overall, the ECM with its identified parameters shows
Time [s] #105
good performance. The identification dataset was acquired by
Fig. 7. Measured current and voltage for parameter validation. applying high currents to the battery cell which in turn gave
good results on the validation dataset which does not have
Errors on both the identification and validation dataset are such high currents. Consequently, for the considered cell, the
presented in Table II. Errors considered are root mean squared ECM current amplitude sensitivity might be relatively low.
error (RMSE), mean absolute error (MAE), mean absolute Therefore, carrying on the parameter identification solely on
percentage error (MAPE), minimum of u − û difference, and high currents would be beneficial as both low and high perfor-
maximum of u − û difference. RMSE, MAE, and MAPE are mance cases could be covered while reducing the necessary
quite similar between the datasets meaning that the model experimental data.
did not overfit on the identification dataset, i.e. it generalises The approach presented can be extended by using the same
well. The minimum of u − û difference is significantly higher methodology over a wider temperature range, ideally with
by absolute value on the identification dataset which can be relatively high and low (under 0°C conditions) temperatures,
attributed to the dataset itself as the SoC dropped below where the model accuracy often decays. By doing so, the
0.5 only a few times during the extensive data acquisition, dependence on the battery cell temperature would be obtained.
therefore it was harder to identify parameters for that region The model parameters would be defined as 2-D lookup tables
of small SoC. The maximum of u − û difference is fairly having SoC as one dimension and cell temperature as the other
similar across the datasets and it amounts to roughly 6% of dimension.
the measured value which incurred at low SoCs. Similarly, the identification dataset could be extended to
ensure that the whole range of SoC is equally represented.
TABLE II Indeed, the identified ECM has the largest error for low SoC
M ETRICS FOR EVALUATION OF THE IDENTIFIED MODEL PARAMETERS .
values since the identification dataset is mostly acquired for
Identification dataset Validation dataset SoC values higher than 0.5.
RMSE [V] 0.02186 0.01753 R EFERENCES
MAE [V] 0.01412 0.01414 [1] European Parliament and the Council, “Directive (EU) 2019/1161 of
the European Parliament and of the Council of 20 June 2019 amending
MAPE [%] 0.39 0.37
Directive 2009/33/EC on the promotion of clean and energy-efficient
min (u − û) [V] -0.60961 -0.18245 road transport vehicles,” 2019.
[2] J. A. Sanguesa, V. Torres-Sanz, P. Garrido, F. J. Martinez, and J. M.
max (u − û) [V] 0.21000 0.22895
Marquez-Barja, “A review on electric vehicles: Technologies and chal-
lenges,” Smart Cities, vol. 4, no. 1, pp. 372–404, 2021.
A similar model was developed in [17], where authors [3] B. Lebrouhi, Y. Khattari, B. Lamrani, M. Maaroufi, Y. Zeraouli, and
T. Kousksou, “Key challenges for a large-scale development of battery
reported MAPE of 0.54% when using Coulomb counting for electric vehicles: A comprehensive review,” Journal of Energy Storage,
SoC estimation. The authors identified model parameters only vol. 44, p. 103273, 2021.

979-8-3503-0488-6/23/$31.00 ©2023 IEEE 675

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on June 14,2024 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.
 40
Cell current, i
Current [A] 20

2.1 2.101 2.102 2.103 2.104 2.105 2.106 2.107 2.108 2.109
Time [s] #105

4
Voltage [V]

3.5

Measured voltage, u
3 Model voltage, u
^ 2.103
2.1 2.101 2.102 2.104 2.105 2.106 2.107 2.108 2.109
Time [s] #105
0.1
Model error, u ! u
^
Voltage [V]

-0.1
2.1 2.101 2.102 2.103 2.104 2.105 2.106 2.107 2.108 2.109
Time [s] #105
0.2 Slow di,use voltage, u1
Voltage [V]

Fast di,use voltage, u2

0.1

-0.1
2.1 2.101 2.102 2.103 2.104 2.105 2.106 2.107 2.108 2.109
Time [s] #105
1
State of Charge
SoC [-]

0.5

0
2.1 2.101 2.102 2.103 2.104 2.105 2.106 2.107 2.108 2.109
Time [s] #105

Fig. 8. A detail lasting 1000 s from the validation dataset showing current applied to the cell, measured and model output voltages, their difference, diffuse
voltages, and the state of charge.

[4] V. Vaideeswaran, S. Bhuvanesh, and M. Devasena, “Battery management and Automation (IFEEA), 2022, pp. 130–136.
systems for electric vehicles using lithium ion batteries,” in 2019 [11] Q. Zhang, S. Zhang, H. Yin, K. Ma, M. Xiu, and M. Wang, “Equivalent
Innovations in Power and Advanced Computing Technologies (i-PACT), circuit model and parameter identification of lithium-sulfur battery,” in
vol. 1, 2019, pp. 1–9. 2023 International Conference on Distributed Computing and Electrical
[5] D. Marušić and M. Vašak, “Efficient method of identifying a Li-ion Circuits and Electronics (ICDCECE), 2023, pp. 1–8.
battery model for an electric vehicle,” in 2022 IEEE 20th International [12] T. Yutthanava, S. Khomfoi, and T. Phophongviwat, “Modeling Li-ion
Power Electronics and Motion Control Conference (PEMC), 2022, pp. battery using measurement data,” in 2023 Third International Symposium
421–426. on Instrumentation, Control, Artificial Intelligence, and Robotics (ICA-
[6] M. Doyle, T. F. Fuller, and J. Newman, “Modeling of galvanostatic SYMP), 2023, pp. 131–134.
charge and discharge of the lithium/polymer/insertion cell,” Journal of [13] Q.-K. Wang, Y.-J. He, J.-N. Shen, X.-S. Hu, and Z.-F. Ma, “State
the Electrochemical Society, vol. 140, no. 6, p. 1526, 1993. of charge-dependent polynomial equivalent circuit modeling for elec-
[7] B. S. Haran, B. N. Popov, and R. E. White, “Determination of the hydro- trochemical impedance spectroscopy of lithium-ion batteries,” IEEE
gen diffusion coefficient in metal hydrides by impedance spectroscopy,” Transactions on Power Electronics, vol. 33, no. 10, pp. 8449–8460,
Journal of Power Sources, vol. 75, no. 1, pp. 56–63, 1998. 2018.
[8] R. Cipin, M. Toman, P. Prochazka, and I. Pazdera, “Identification of [14] W. Zhou, Y. Zheng, Z. Pan, and Q. Lu, “Review on the battery model
Li-ion battery model parameters,” in 2019 International Conference on and SOC estimation method,” Processes, vol. 9, no. 9, 2021.
Electrical Drives & Power Electronics (EDPE), 2019, pp. 225–229. [15] Samsung SDI Co., Ltd., “Specification of product for lithium-ion
[9] B. Sun, X. He, W. Zhang, H. Ruan, X. Su, and J. Jiang, “Study of rechargeable cell INR21700-48X,” 2019.
parameters identification method of Li-ion battery model for EV power [16] Q.-Q. Yu, R. Xiong, L.-Y. Wang, and C. Lin, “A comparative study on
profile based on transient characteristics data,” IEEE Transactions on open circuit voltage models for lithium-ion batteries,” Chinese Journal
Intelligent Transportation Systems, vol. 22, no. 1, pp. 661–672, 2021. of Mechanical Engineering, vol. 31, no. 1, pp. 1–8, 2018.
[10] Y. Wu, H. Chen, L. Cao, J. Duan, X. Chen, and J. Zhai, “Research [17] S. J. Navas, G. C. González, F. Pino, and J. Guerra, “Modelling Li-ion
on online identification of lithium-ion battery equivalent circuit model batteries using equivalent circuits for renewable energy applications,”
parameters,” in 2022 9th International Forum on Electrical Engineering Energy Reports, vol. 9, pp. 4456–4465, 2023.

979-8-3503-0488-6/23/$31.00 ©2023 IEEE 676

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on June 14,2024 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.