IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 69, NO.

1, JANUARY 2020 235

General Parameter Identification Procedure and

Comparative Study of Li-Ion Battery Models
Henry Miniguano , Student Member, IEEE, Andrés Barrado , Senior Member, IEEE,
Antonio Lázaro , Member, IEEE, Pablo Zumel , Member, IEEE, and Cristina Fernández , Member, IEEE

Abstract—Accurate and robust battery models are required models is a trade-off between model complexity, parameter
for the proper design and operation of battery-powered systems. estimation effort, and accuracy.
However, the parametric identification of these models requires In general, battery models can be classified into three cate-
extensive and sophisticated methods to achieve enough accuracy.
This article shows a general and straightforward procedure, based gories: electrochemical, mathematical, and electrical. Electro-
on Simulink and Simscape of Matlab, to build and parameterize chemical models consist of a set of partial differential-algebraic
Li-ion battery models. The model parameters are identified with the equations along with a wide range of limitations [5], [6]. The
Optimization Toolbox of Matlab, by means of an iterative process to estimation of the electrochemical model is very accurate [7].
minimize the sum of the squared errors. In addition, this procedure However, simulation with these models can be time-consuming,
is applied to a selection of five different models available in the
literature for electric vehicle applications, obtaining a comparative since it requires to solve a complex set of equations to identify
study between them. Also, the performance of each battery model the electrochemical properties [8]. Mathematical models are an
is evaluated through two current profiles from two driven profiles alternative based on empirical formulas approaches [9], [10].
known as the Urban Driving Cycle (ECE-15 or UDC) and the These models are empirical functions with parameters derived
Hybrid Pulse Power Characterization (HPPC). The experimental from datasheet or from static measurements and use a fewer
results obtained from a Li-ion polymer battery have been compared
with the data provided by the models, confirming the effectiveness number of equations. However, these models cannot represent
of the proposed procedure, and also, the application field of each the dynamic information correctly and provide inaccurate results
model as a function of the required accuracy. in the order of 5%–20% error. Finally, circuit-based or electri-
Index Terms—ECE15, HPPC, Li-ion battery models, Matlab, cal models are able to accurately exhibit the current-voltage
parameter estimation, Simscape, Simulink. characteristics of the batteries while maintaining simulation
efficiency. The accuracy of these electrical models lies between
I. INTRODUCTION electrochemical and mathematical models [11]. These models
N RECENT years, rechargeable batteries are playing an are based on resistors and capacitors connected to controlled
I essential role as energy storage and as power sources for
some electrical systems such as communications systems, re-
voltages sources, and they can represent the fast dynamic of the
battery with a high level of accuracy [12].
newable power systems, electric vehicles, electric buses, etc. In the literature, several papers develop a comparative study
[1], [2]. Lithium-ion batteries are being an essential technology among different models. So, [13] shows a systematic review of
in powertrain electrification of hybrid electric vehicles (HEV) ten models, using nonlinear identification technique with the
and pure electric vehicles (EV), which allow reducing fuel dual Extended Kalman Filter (dual-EKF). This paper concludes
consumption and decreasing greenhouse gas emissions [3]. This the RC model provides the best dynamic response and accuracy.
kind of battery presents some performances very suitable for In addition, [14] examines eleven electrical circuits models in
transportation applications, such as high energy/power density which parameters are estimated with a genetic algorithm (GA),
ratio, slow self-discharge, high cycle life and lack of hysteresis and it concludes that the higher and constant RC networks
effect in some of them [4]. have better robustness considering parameter variations and
In the state-of-the-art, numerous battery models have been sensor errors.
reported for system-level modeling. Choosing between those The main contribution of this paper is the proposal of a general
and straightforward offline-procedure in order to estimate the
Manuscript received February 14, 2019; revised June 30, 2019 and October 2, battery model parameters. The identification method is based on
2019; accepted October 27, 2019. Date of publication November 11, 2019; date the interface of MATLAB System Identification Toolbox, which
of current version January 15, 2020. This work was supported by the Ministry uses numerical optimization to minimize the error between the
of Economy and Competitiveness and FEDER funds through the research
project “Storage and Energy Management for Hybrid Electric Vehicles based measured and the estimated battery voltage. In order to verify the
on Fuel Cell, Battery, and Supercapacitors”—ELECTRICAR-AG-(DPI2014- identification procedure, a comparative analysis of the most used
53685-C2-1-R). The review of this article was coordinated by Prof. M. Preindl. battery models for electric vehicle applications is performed.
(Corresponding author: Henry Miniguano.)
The authors are with the Power Electronics Systems Group, Universidad The selected models can represent the most important battery
Carlos III de Madrid, Leganés 28911, Spain (e-mail: [email protected]; effects that affect the control and battery.
[email protected]; [email protected]; [email protected]; The next sections are organized as follows: Section II de-
[email protected]).
Digital Object Identifier 10.1109/TVT.2019.2952970 scribes the parameter estimation procedure for battery models

0018-9545 © 2019 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.

Authorized licensed use limited to: Harcourt Butler Technical University Kanpur. Downloaded on August 24,2023 at 06:33:46 UTC from IEEE Xplore. Restrictions apply.
 236 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 69, NO. 1, JANUARY 2020

the state of charge (SOC); Ri and Ro are resistors, R1 C1 to Rn Cn

represent resistors and capacitors in parallel, and Rp Cp represent
resistors and capacitors in series. This equivalent circuit model
could be implemented in Simulink, Fig. 2b, or Simscape, Fig. 2c,
where the identification current profile is the input signal, and
the voltage and SOC are the outputs signals of the model.
In addition, the model could use different lookup tables,
variable voltage sources, nonlinear resistors, and nonlinear ca-
pacitors to represent the nonlinear behavior of the battery [23].
Fig. 1. Schematic diagram of the general parameter estimation.
Also, a detailed general identification procedure is depicted
in Fig. 2d.
The procedure for using the tool includes several and straight-
using Simulink and Simscape. In Section III, the experimental forward steps:
setup is described. Section IV presents a description of different
battery models and their main parameters. Section V shows the
models of experimental validation using the ECE15 and HPPC A. Parameter Estimation
dynamic driving cycle. Finally, in Section VI, the conclusions 1) Build the equivalent circuit model. It can be a block
are presented. diagram in Simulink or a circuit in Simscape.
2) Import experimental data for parameter estimation in the
II. PARAMETER ESTIMATION PROCEDURE tool Parameter Estimation of Simulink.
Battery models can be used in two ways: online and offline. 3) Choose the parameters to estimate and their limits.
Online models are used in real-time battery management sys- 4) Set up optimization options and parallel compute options
tems for the state of charge (SOC) estimation, with tracking to accelerate the numerical estimation.
algorithms on different parameters [15]–[17]. On the other hand, 5) Run the estimation task and get the parameters after the
an offline battery model is used for system-level design, to value needed iteration to optimize the cost function.
the battery state of health or for simulation [18], [19]. The 6) Verify the model response by comparing the experimental
accuracy of these models greatly depends on carefully identified results to the simulated data (obtained using the estimated
model parameters [20], [21]. parameters). If the error is not small enough, in this order,
However, from a practical point of view, this paper proposes or return to step 1 to change the Model ( 1 in Fig. 2c), or
to estimate and validate multiple parameters of the model at the change the Identification method and return to step 3 ( 2
same time, as it is made in the Optimization Toolbox of Matlab. in Fig. 2c), or modify the Identification current profile
This parameter identification proposal can be used in both types and return to step 2 ( 3 in Fig. 2c).
of applications, offline and online, with some differences in the
parameter estimation procedure. In this paper, this procedure is B. Validation
applied to offline modeling.
7) Import new application or verification current profile to
The identification of the model parameters p is performed by
the model, with the estimated parameters.
minimizing the cost function F(p) (1), which is the minimum
8) Obtain the voltage simulated response.
of the sum of the squared errors between the experimentally
9) Verify the model response by comparing the real volt-
measured output voltage Vm and the model output voltage Vs ,
age and the simulated voltage data. If the error is not
as shown in Fig. 1.
small enough, in this order, or return to step 1 to change
n
 the Model ( 1 in Fig. 2c), or change the Identification
F (p) = (Vm (ti ) − Vs (ti , p))2 (1) method and return to step 3 (2 in Fig. 2c), or modify the
i=0
Identification current profile and return to step 2 (3 in
The optimization method used for parameter identification Fig. 2c).
is known as Nonlinear Least Squares, described in [22]. This
method is used since the quadratic cost function is easier to III. EXPERIMENTAL SETUP
minimize than other cost functions, and it is sufficiently accurate
for this type of application. A. Battery Testing System
Fig. 2 shows the battery model identification procedure to The experimental setup is shown in Fig. 3a. It includes a bat-
estimate the optimum parameter values once the identification tery cell along with a data acquisition system and dynamic load.
current profile, model, and identification method have been Experiments have been conducted on pouch lithium-ion polymer
chosen. Fig. 2a shows a general equivalent circuit model that HRB 8048145, which specifications are listed in Table I. In
represents many of the battery electric models available in the order to verify the battery model’s accuracy when is used in
literature. Each model uses a different combination of elements, automotive applications, the experimental current profile and
as it is shown in Section IV. OCV represents the open-circuit the data acquisition system are conducted using the following
voltage; Vv (SOC) represents a variable voltage depending on set of equipment:

Authorized licensed use limited to: Harcourt Butler Technical University Kanpur. Downloaded on August 24,2023 at 06:33:46 UTC from IEEE Xplore. Restrictions apply.
MINIGUANO et al.: GENERAL PARAMETER IDENTIFICATION PROCEDURE AND COMPARATIVE STUDY OF LI-ION BATTERY MODELS 237

Fig. 2. Parameter estimation: (a) general equivalent circuit model, (b) equivalent circuit model in Simulink, (c) equivalent circuit model in Simscape, and
(d) general identification procedure.

TABLE I 4) Shunt resistor: Newtons 4th HF200 (0.5 mΩ) to measure

LITHIUM ION POLYMER HRB 8048145 SPECIFICATION
the cell current.
All these elements have been synchronized by LabVIEW
software in order to produce all current pulse and specific current
profile to verify the equivalent model of the battery.

B. Battery Test Schedule

The parameter identification procedure uses for each model,
either a constant current profile test or a pulsating current dis-
charge profile test, in order to obtain parameter estimation.
Fig. 3b shows the battery discharge curve obtained at constant
1) DC electronic load: Chroma 63206A-600-420 that repro- current (generally equal to 20% of the rated capacity until the
duces DC motor current consumption periods. cut-off voltage). From this curve, the real battery capacity is
2) DC power source: Sorensen SGI400/38 to inject energy obtained, Q = 5.58 Ah, by integrating the current as a function
from the regenerative braking. of time.
3) Datalogger: Agilent 34970A to measure cell voltage and On the other hand, the state of charge (SOC) estimation is a
current. challenging task due to the battery aging and temperature effect.

Authorized licensed use limited to: Harcourt Butler Technical University Kanpur. Downloaded on August 24,2023 at 06:33:46 UTC from IEEE Xplore. Restrictions apply.
 238 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 69, NO. 1, JANUARY 2020

Fig. 3. (a) Battery testing system. Current profiles for identification: (b) constant and (c) pulsating current discharge profile for verification: (d) HPPC and
(e) ECE15.

Some techniques are presented in [24], where the tendency of Fig. 3d shows the HPPC test, described in [25]. It is applied
estimation is a mixture of probabilistic and artificial intelligence repeatedly for almost one hour to validate how robust is the
techniques. battery model long-time response. Fig. 3e shows the ECE15
The Coulomb counting method is used in this paper for SOC test, described in [26], which is applied repeatedly for almost
calculation because it is simple and provides good accuracy. In one hour.
order to obtain the data for battery model identification, current
pulses of 10% SOC are applied, as shown in Fig. 3c.
In addition, to verify the robustness and accuracy of the battery IV. BATTERY MODELS
models, two verification test profiles are applied: HPPC profile This section analyses five battery models selected from the
and ECE15 profile. These profiles are widely used in the battery literature, on which the procedure shown in Section II has been
and supercapacitor modeling, in electric vehicles. applied.

Authorized licensed use limited to: Harcourt Butler Technical University Kanpur. Downloaded on August 24,2023 at 06:33:46 UTC from IEEE Xplore. Restrictions apply.
MINIGUANO et al.: GENERAL PARAMETER IDENTIFICATION PROCEDURE AND COMPARATIVE STUDY OF LI-ION BATTERY MODELS 239

TABLE II
SHEPHERD MODEL BATTERY PARAMETERS

Fig. 4. Shepherd mathematical model in Simulink.

In [27], different datasets of an HEV Li-polymer cell was

used to compare mathematical models, where enhanced-self-
correcting with 4 filters states gave the best performance in all
cases, at the cost of greater complexity. An overview of generic
battery models (mathematical and electrical) was presented in
[28], where the continuous and transients current test of the RC
model showed a higher performance.
Another comparative study between several models was per-
formed in [29], where the third-order RC model with hysteresis
shows a better performance. In addition, [14] shows a systematic
comparative study of equivalent circuit models, in which the
2RC model has better accuracy. However, if one-state hysteresis
is added to the model (2RCH), the accuracy and the reliability
do not improve for LiMNC cells.
In this paper, the selected models are the Sheperd Model [30],
the Nonlinear Model [31], PNGV Model [32], the 3rd Order
RC Model with One-State Hysteresis [29] and the Three RC
Network Model [33]. These models were selected to cover most
of the applications, and they are briefly reviewed below.

A. Shepherd Model
Fig. 5. Nonlinear RC Battery: (a) model structure and (b) OCV vs. SOC.
The Li-ion battery model proposed in [30] is used due to the
fact that this model requires only a few data from manufacturer
datasheet and the battery discharge curve. This model is based
B. The Nonlinear Model
on the modified Shepherd model, and it can accurately reflect
the macro-level characteristics of current and voltage that are The proposed model in [31] is demonstrated to be more
important for system-level simulations, as shown in Fig. 4. appropriate than the well-known Randles’s battery circuit model
This battery model is implemented in Simulink, and it is com- applied for lead-acid battery technologies. Despite the good
posed of a controlled voltage source and an internal resistance. results obtained with the modified Randles’s model, however,
The controlled voltage source value depends on two expressions, for lithium-polymer battery, this model is not appropriated due
one for battery discharge and other for battery charge. Where E0 to the battery nonlinearly behavior. For this reason, a nonlinear
is the battery constant voltage (V), K is the polarization constant RC battery model is proposed in [35]. The proposed model
(V Ah−1 ), Q is the maximum battery capacity (Ah), i∗ is the consists of a nonlinear voltage source, Open Circuit Voltage,
filtered battery current (A), it is the actual battery charge (Ah), A OCV(SOC) as a function of SOC in order to represent nonlinear
is the exponential zone amplitude (V), B is the exponential zone characteristics; a capacitance Cp to model polarization effect; a
time constant inverse (Ah−1 ), Tc is the battery time constant (s), propagation resistor Rb ; a diffusion resistor Rp ; and an ohmic
and R is the battery internal resistance (Ω) [34]. resistance Rt , as shown in Fig. 5a.
Applying a discharge curve of 0.2 C and using the Parameter The obtained parameter values are shown in Fig. 5b and
Estimation tool shown in Section II along with the battery model Table III, after using the identification procedure proposed in
in Simulink, the different estimated parameters for this model Section II. The model parameters depend on the state of charge
are obtained, as shown in Table II. at each current level during discharging.

Authorized licensed use limited to: Harcourt Butler Technical University Kanpur. Downloaded on August 24,2023 at 06:33:46 UTC from IEEE Xplore. Restrictions apply.
 240 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 69, NO. 1, JANUARY 2020

TABLE III
BATTERY DYNAMIC MODEL PARAMETERS

Fig. 7. The 3rd Order RC Model with One-State Hysteresis (3RCH): (a)
electric circuit and (b) OCV versus SOC.

the procedure described in Section II, are shown in the Appendix

and are depicted in Fig. 6b.

D. The 3rd Order RC Model With One-State Hysteresis

The Thevenin circuit with three RC parallel networks with
hysteresis is proposed in [29]. In this model, a hysteresis volt-
age Vh is added in series with an ohmic resistance and three
RC parallel networks, Fig. 7a. The battery hysteresis model,
proposed in [27], is defined by (2):

  
 i(k) · γ · Δt 
h(k) = exp −   · h(k)

Fig. 6. PNGV model: (a) electric circuit and (b) parameter values. Q
   
 i(k) · γ · Δt 
− 1 − exp −    · sgn (i (k)) (2)
Q 
C. Partnership for a New Generation of
Vehicles (PNGV) Model
The PNGV model is accurate and able to describe the battery The hysteresis voltage, Vh (in the order of few millivolts), is
behavior during the discharge process [32]. In [36], research Mh(k), where M is the maximum hysteresis value, γ adjusts the
about Li-ion battery enhances the PNGV model by adding changing rate of hysteresis voltage and Q is the total capacity. In
variable RC components, as shown in Fig. 6a. This model is [37], the authors provide a robust and consistent methodology
composed of the OCV0 represents the initial constant voltage to assess the OCV of different chemistry lithium-ion cells. This
source (V), and C0 (SOC) represents a variable capacitor, both method is applied to the under-test cell to get a variation of 10%
connected in series represent the nonlinear voltage characteris- of SOC using a 1C current until the cut-off voltage, as shown in
tics of the battery, which is dependent on SOC. R0 (SOC) rep- Fig. 7b. This information allows to obtain the parameter used in
resents the battery internal resistance (Ω), and also two parallel (2) needed to calculate Vh.
networks based on R1 , C1 , R2, and C2 as a function of SOC. The Table IV shows the average values of RC and the average
values of the different parameters estimated for this model, with hysteresis parameters from charge and discharge curves.

Authorized licensed use limited to: Harcourt Butler Technical University Kanpur. Downloaded on August 24,2023 at 06:33:46 UTC from IEEE Xplore. Restrictions apply.
MINIGUANO et al.: GENERAL PARAMETER IDENTIFICATION PROCEDURE AND COMPARATIVE STUDY OF LI-ION BATTERY MODELS 241

TABLE IV TABLE V
DYNAMIC MODEL BATTERY PARAMETERS OUTPUT VOLTAGE ERROR WITH HPPC TEST

represents the internal resistance of the battery, and three parallel

networks based on R1 , C1 , R2 , C2 , R3, and C3 . The values
of the different parameters estimated for this model, with the
procedure described in Section II, are shown in the Appendix
and are depicted in Fig. 8b.

V. COMPARISON RESULT AND DISCUSSION

After obtaining parameters for each model, using in all cases,
the same procedure shown in Section II, the output voltage
accuracy and robustness analysis for the five models of the
polymer Li-ion battery is performed with the HPPC test and the
ECE15 test. The results are illustrated in Fig. 9a–9d for HPPC
test and Fig. 9e–9h for the ECE15 test. In these figures, the
current profiles are shown in Fig. 9a and Fig. 9e, the simulated
and the experimental voltage are shown in Fig. 9b and Fig. 9f,
the voltage relative error is shown in Fig. 9c and Fig. 9g, and
the maximum and average voltage relative error are shown in
Fig. 9d and Fig. 9h.
In Fig. 9c the 3rd Order RC Model with One-State Hysteresis
(3RCH) and the Three RC Network Model (3RC(SOC)) have
a lower mean and maximum voltage relative error, as well as
with convergent tendency compared to other models, while the
Nonlinear RC Model (Nonlinear RC) has the highest average
voltage relative error without a convergent tendency. The Part-
nership for a New Generation of Vehicles (PNGV) Model has
a good response for a short simulation time but without enough
convergent tendency for a long simulation time. A summary
of the results is shown in Table V and Fig. 9d, for the HPPC
profile.
Fig. 8. Three RC Network Model: (a) electric circuit and (b) parameter values. The obtained results under the ECE15 test are shown in
Fig. 9g, where again the 3rd Order RC Model with One-
State Hysteresis (3RCH) and the Three RC Network Model
(3RC(SOC)) have a lower average voltage relative error with a
E. The Three RC Network Model convergent tendency. However, the Nonlinear RC Model (Non-
Each element of the equivalent circuit of Fig. 8a is a func- linear RC) has a higher average and maximum voltage relative
tion of SOC, with three RC networks that represent the slow, error without a convergent tendency. The Partnership for a New
intermediate, and fast dynamic response of the battery. In [33], Generation of Vehicles (PNGV) Model has a good response for
a comparison is made when one, two, and three branches are a short simulation time but without enough convergent tendency
used, obtaining better accuracy as more branches in parallel are for a long simulation time, as in the HPPC test. The Shepherd
used. The disadvantage of using a higher number of RC branches Model (Shepherd) has the maximum voltage peak error and sim-
is that the mathematical model becomes more complex since it ilar voltage relative error to the PNGV model, with a convergent
includes more lookup tables [22]. tendency. A summary of the results is shown in Table VI and
In this case, three RC branches are used in order to obtain Fig. 9h.
better precision, without increasing the complexity of the model In general, models based on RC networks have a greater ca-
response, where OCV represents the open-circuit voltage, R0 pacity to reproduce the different effects that appear in the battery,

Authorized licensed use limited to: Harcourt Butler Technical University Kanpur. Downloaded on August 24,2023 at 06:33:46 UTC from IEEE Xplore. Restrictions apply.
 242 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 69, NO. 1, JANUARY 2020

Fig. 9. Time response of lithium-ion battery: (a)–(d) current profile HPPC and (e)–(h) current profile ECE15. (a) current profile HPPC, (b) simulated and
experimental voltage, (c) voltage relative error, (d) maximum and average voltage relative error; (e) current profile ECE15, (f) simulated and experimental voltage
and (g) voltage relative error and (h) maximum and average voltage relative error.

whether short, medium or long term, since with a sufficient The results obtained indicate that the Three RC Network
number of RC networks all phases of polarization of the battery Model, followed by the 3rd Order RC Model with One-State
can be represented. Hysteresis are the two models with the best results, although
However, the PNGV model produces long-term accumulation most of the parameters of the 3rd Order RC Model with
of voltage on the capacitor, which increases the model voltage One-State Hysteresis are constant, so this model is easier to
error. implement.

Authorized licensed use limited to: Harcourt Butler Technical University Kanpur. Downloaded on August 24,2023 at 06:33:46 UTC from IEEE Xplore. Restrictions apply.
MINIGUANO et al.: GENERAL PARAMETER IDENTIFICATION PROCEDURE AND COMPARATIVE STUDY OF LI-ION BATTERY MODELS 243

TABLE VI The Nonlinear RC model does not accurately reflect the

OUTPUT VOLTAGE ERROR WITH ECE15 TEST
dynamic characteristics of the battery because this model cannot
reflect the polarization phases of the Li-ion battery. Therefore,
its accuracy is limited.
The 3RCH and 3RC(SOC) models do not show a cumulative
error in long simulation time, and the simulated voltage is better
than the previous ones. The 3RCH includes the hysteresis phe-
nomenon that depends on the battery chemistry, and it is simpler
to estimate than the 3RC(SOC) model, which is more accurate,
although more complex. Therefore, it can be concluded that
both models have higher accuracy and robustness than the other
models in order to show the static and dynamic battery voltage.
A tradeoff between complexity, robustness, and accuracy is
obtained with the 3RCH model.
VI. CONCLUSION
Considering these results, Shepherd Model is recommended
This paper describes a general parameter identification proce- for short simulation time and 3RCH model for medium and long
dure with a reduced number of steps and an easy user interface, simulation time.
which requires circuits or equations to implement any type
of battery model. The Parameter Estimation Tool is available APPENDIX
in Simscape or Simulink of Matlab, and it can estimate and
Models parameters of Nonlinear, PNGV and Three RC Net-
validate multiple parameters of the model at the same time, using
work Models are shown in Tables VII, VIII, and IX.
multi-experiment data with specific bounds for each parameter.
This parameter identification proposal can be used in offline TABLE VII
applications. NONLINEAL MODEL PARAMETERS
A comparative study of five different battery models available
in the literature has been made. In all cases, the model param-
eters have been obtained from the proposed general parameter
identification procedure.
The comparison result obtained from the five models indicates
that the Shepherd Model is straightforward, and it can be used
as a first approximation, needing only the discharge curve of the
battery to be parameterized. However, when it is used for a long
simulation time, this model produces a cumulative voltage error,
and it cannot represent the voltage peaks of the battery response.
The PNGV model is excellent for a short simulation time;
however, once again, this model produces long-term accumu-
lation of voltage on the capacitor, which increases the model
voltage error.

TABLE VIII
PNGV MODEL PARAMETERS

Authorized licensed use limited to: Harcourt Butler Technical University Kanpur. Downloaded on August 24,2023 at 06:33:46 UTC from IEEE Xplore. Restrictions apply.
 244 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 69, NO. 1, JANUARY 2020

TABLE IX
THEVENIN MODEL PARAMETERS

REFERENCES [18] L. Turos, I. Szekely, G. Csernáth, and K. Gyorgy, “Offline battery pack
model optimization,” in Proc. IEEE Int. Conf. Expo. Elect. Power Eng.,
[1] X. Luo, J. Wang, M. Dooner, and J. Clarke, “Overview of current develop- 2014, pp. 973–977.
ment in electrical energy storage technologies and the application potential [19] H. Yuan and L. Dung, “Offline state-of-health estimation for high-power
in power system operation,” Appl. Energy, vol. 137, pp. 511–536, 2015. lithium-ion using three-point impedance extraction method,” IEEE Trans.
[2] X. Zhang, H. Peng, H. Wang, and M. Ouyang, “Hybrid lithium iron Veh. Technol., vol. 66, no. 3, pp. 2019–2032, Mar. 2017.
phosphate battery and lithium titanate battery systems for electric buses,” [20] D. Dvorak, T. Bauml, A. Holzinger, and H. Popp, “A comprehensive algo-
IEEE Trans. Veh. Technol., vol. 67, no. 2, pp. 956–965, Feb. 2018. rithm for estimating lithium-ion battery parameters from measurements,”
[3] M. U. Cuma and T. Koroglu, “A comprehensive review on estimation IEEE Trans. Sustain. Energy, vol. 9, no. 2, pp. 771–779, Apr. 2018.
strategies used in hybrid and battery electric vehicles,” Renewable Sustain. [21] N. Kim, A. Rousseau, and E. Rask, “Parameter estimation for a lithium-
Energy Rev., vol. 42, pp. 517–531, 2015. ion battery from chassis dynamometer tests,” IEEE Trans. Veh. Technol.,
[4] N. Nitta, F. Wu, J. T. Lee, and G. Yushin, “Li-ion battery materials: Present vol. 65, no. 6, pp. 4393–4400, Jun. 2016.
and future,” Mater. Today, vol. 18, no. 5, pp. 252–264, 2015. [22] Simulink Design Optimization User’s Guide R2018a, Mathworks, Natick,
[5] Z. Yang, D. Patil, and B. Fahimi, “Electrothermal modeling of lithium-ion MA, USA, 2018.
batteries for electric vehicles,” IEEE Trans. Veh. Technol., vol. 68, no. 1, [23] R. Ahmed et al., “Model-based parameter identification of healthy and
pp. 170–179, Jan. 2019. aged Li-ion batteries for electric vehicle applications,” SAE Int. J. Alter-
[6] M. Farag, M. Fleckenstein, and S. Habibi, “Continuous piecewise-linear, native Powertrains, vol. 4, no. 2, pp. 233–247, 2015.
reduced-order electrochemical model for lithium-ion batteries in real-time [24] J. Rivera-Barrera, N. Muñoz-Galeano, and H. Sarmiento-Maldonado,
applications,” J. Power Sources, vol. 342, pp. 315–362, 2017. “SoC estimation for lithium-ion batteries: Review and future challenges,”
[7] G. Fan, X. Li, and M. Canova, “A reduced-order electrochemical model Electronics, vol. 6, no. 4, pp. 1–33, 2017.
of Li-ion batteries for control and estimation applications,” IEEE Trans. [25] J. P. Christopherson, “Battery test manual for electric vehicles,” U.S. Dept.
Veh. Technol., vol. 67, no. 1, pp. 76–91, Jan. 2018. Energy Veh. Technol. Program, Idaho Nat. Lab., Idaho Falls, ID, USA,
[8] C. Zhang, W. Allafi, Q. Dinh, P. Ascencio, and J. Marco, “Online estimation INL/EXT-15-34184, 2015, pp. 6–9.
of battery equivalent circuit model parameters and state of charge using [26] C. Raga, A. Barrado, A. Lazaro, I. Quesada, M. Sanz, and P. Zumel,
decoupled least squares technique,” Energy, vol. 142, pp. 678–688, 2018. “Driving profile and fuel cell minimum power analysis impact over the
[9] S. Li and B. Ke, “Study of battery modeling using mathematical and circuit size and cost of fuel cell based propulsion systems,” in Proc. 9th Int. Conf.
oriented approaches,” in Proc. IEEE Power Energy Soc. Gen. Meeting, Compat. Power Electron., 2015, pp. 390–395.
2011, pp. 1–8. [27] G. L. Plett, “Extended Kalman filtering for battery management systems
[10] E. Raszmann, K. Baker, Y. Shi, and D. Christensen, “Modeling stationary of LiPB-based HEV battery packs—Part 2. Modeling and identification,”
lithium-ion batteries for optimization and predictive control,” in Proc. J. Power Sources, vol. 134, no. 2, pp. 262–276, 2004.
IEEE Power Energy Conf. Illinois, 2017, pp. 1–7. [28] A. A. H. Hussein and I. Batarseh, “An overview of generic battery models,”
[11] C. Zhang, K. Li, J. Deng, and S. Song, “Improved realtime state-of-charge in Proc. IEEE Power Energy Soc. Gen. Meeting, 2011, pp. 4–9.
estimation of LiFePO4 battery based on a novel thermoelectric model,” [29] M. Farag, M. Fleckenstein, and S. R. Habibi, “Li-ion battery SOC esti-
IEEE Trans. Ind. Electron., vol. 64, no. 1, pp. 654–663, Jan. 2017. mation using non-linear estimation strategies based on equivalent circuit
[12] H. E. Perez, X. Hu, S. Dey, and S. J. Moura, “Optimal charging of Li-ion models,” SAE Tech. Paper, vol. 1, pp. 1–11, 2014.
batteries with coupled electro-thermal-aging dynamics,” IEEE Trans. Veh. [30] O. Tremblay and L. A. Dessaint, “Experimental validation of a battery
Technol., vol. 66, no. 9, pp. 7761–7770, Sep. 2017. dynamic model for EV applications,” World Elect. Veh. J., vol. 3, no. 2,
[13] S. Nejad, D. T. Gladwin, and D. A. Stone, “A systematic review of lumped- pp. 289–298, 2009.
parameter equivalent circuit models for real-time estimation of lithium-ion [31] C. R. Gould, C. M. Bingham, D. A. Stone, and P. Bentley, “New battery
battery states,” J. Power Sources, vol. 316, pp. 183–196, 2016. model and state-of-health determination through subspace parameter esti-
[14] X. Lai, Y. Zheng, and T. Sun, “A comparative study of different equivalent mation and state-observer techniques,” IEEE Trans. Veh. Technol., vol. 58,
circuit models for estimating state-of-charge of lithium-ion batteries,” no. 8, pp. 3905–3916, Oct. 2009.
Electrochim. Acta, vol. 259, pp. 566–577, 2018. [32] X. Liu, W. Li, and A. Zhou, “PNGV equivalent circuit model and SOC
[15] M. Cacciato, G. Nobile, G. Scarcella, and G. Scelba, “Real-time model- estimation algorithm for lithium battery pack adopted in AGV vehicle,”
based estimation of SOC and SOH for energy storage systems,” IEEE IEEE Access, vol. 6, pp. 23639–23647, 2018.
Trans. Power Electron., vol. 32, no. 1, pp. 794–803, Jan. 2017. [33] R. Jackey, M. Saginaw, P. Sanghvi, J. Gazzarri, T. Huria, and M. Ceraolo,
[16] L. B. Cells, H. Rahimi-eichi, S. Member, and F. Baronti, “Online adap- “Battery model parameter estimation using a layered technique: An exam-
tive parameter identification and state-of-charge coestimation for lithium- ple using a lithium iron phosphate cell,” SAE Tech. Paper, vol. 2, pp. 1–14,
polymer battery cells,” IEEE Trans. Ind. Electron. vol. 61, no. 4, pp. 2053– 2013.
2061, Apr. 2014. [34] S. N. Motapon, A. Lupien-Bedard, L.-A. Dessaint, H. Fortin-Blanchette,
[17] J. Meng et al., “An overview and comparison of online implementable and K. Al-Haddad, “A generic electro-thermal Li-ion battery model for
SOC estimation methods for lithium-ion battery,” IEEE Trans. Ind. Appl., rapid evaluation of cell temperature temporal evolution,” IEEE Trans. Ind.
vol. 54, no. 2, pp. 1583–1591, Mar./Apr. 2018. Electron., vol. 46, no. 2, pp. 998–1008, Feb. 2017.

Authorized licensed use limited to: Harcourt Butler Technical University Kanpur. Downloaded on August 24,2023 at 06:33:46 UTC from IEEE Xplore. Restrictions apply.
MINIGUANO et al.: GENERAL PARAMETER IDENTIFICATION PROCEDURE AND COMPARATIVE STUDY OF LI-ION BATTERY MODELS 245

[35] I.-S. Kim, “Nonlinear state of charge estimator for hybrid electric vehicle Antonio Lázaro was born in Madrid, Spain, in 1968.
battery,” IEEE Trans. Power Electron., vol. 23, no. 4, pp. 2027–2034, He received the M.Sc. degree in electrical engineering
Jul. 2008. from the Universidad Politécnica de Madrid, Madrid,
[36] S. Castano, L. Gauchia, E. Voncila, and J. Sanz, “Dynamical modeling Spain, in 1995, and the Ph.D. degree in electrical en-
procedure of a Li-ion battery pack suitable for real-time applications,” gineering from the Universidad Carlos III de Madrid,
Energy Convers. Manage., vol. 92, pp. 396–405, 2015. Leganés, Spain, in 2003. Since 1995, he has been
[37] A. Barai, W. D. Widanage, J. Marco, A. Mcgordon, and P. Jennings, “A an Assistant Professor with the Universidad Carlos
study of the open circuit voltage characterization technique and hysteresis III de Madrid. Since 1994, he has been involved
assessment of lithium-ion cells,” J. Power Sources, vol. 295, pp. 99–107, in power electronics, participating in more than 50
2015. research and development projects for industry. He
holds seven patents and software registrations and has
authored/coauthored nearly 140 papers in IEEE journals and conferences. His
research interests include switched-mode power supplies, power factor correc-
tion circuits, inverters (railway and grid-connected applications), modeling and
control of switching converters, and digital control techniques.

Henry Miniguano was born in Ambato, Ecuador, in Pablo Zumel (M’06) received the B.S. degree in
1984. He received the M.Sc. and Ph.D. (cum laude) electrical engineering from the University of Burgos,
degrees in electrical engineering from the Carlos Burgos, Spain, in 1995, the first M.S. degree in electri-
III University of Madrid, Leganés, Spain, in 2013 cal engineering from the Universidad Politécnica de
and 2019, respectively. From 2014 to 2019, was in- Madrid (UPM), Madrid, Spain, in 1999, the second
volved in power electronics research and develop- M.S. degree in electrical engineering from the Ecole
ment projects with the Power Electronics Systems Centrale Paris, Paris, France, in 2000, and the Ph.D.
Group, Carlos III University of Madrid. His research degree in electrical engineering from UPM in 2005.
interests include modeling, identification, and con- From 1999 to 2003, he was a Researcher with the
trol of power electronics converters, batteries, and UPM. Since 2003, he has been with the Department
supercapacitors. Dr. Miniguano is the recipient of the of Electronic Technology, Carlos III University of
Young Engineer Award (IEEE International Conference ESAR-ITEC 2016) and Madrid, Leganés, Spain, where he is currently an Associate Professor. He
Best Presenter Award (18th International Symposium Doctoral School of Energy has authored/coauthored more than 100 scientific papers in main international
and Geotechnology III 2019). conferences and journals, has filed four patents, and is the Co-Founder of a
spin-off company targeting CAD for power electronics. He has participated in
more than 50 research projects in the field of power electronics. His research
interests include digital control in power electronics, modeling and control
techniques, modular power converters, design optimization, and educational
topics in power electronics.

Cristina Fernández received the M.Sc. and Ph.D. de-

Andrés Barrado received the M.Sc. degree in elec- grees in electrical engineering from the Universidad
trical engineering from the Polytechnic University of Politecnica de Madrid (UPM), Madrid, Spain. From
Madrid, Madrid, Spain, in 1994, and the Ph.D. degree 1997 to 2003, she researched in the field of power
from the Carlos III University of Madrid, Leganés, electronics with UPM. In Summer 2000, she was a
Spain, in 2000. He is currently a Full Professor with summer intern with the Center of Research and De-
the Carlos III University of Madrid, where he has been velopment, General Electric, Schenectady, NY, USA.
the Head of the Power Electronics Systems Group, Since 2003, she has been with the Carlos III Univer-
since 2004. His research interests are switching-mode sity of Madrid, Leganés, Spain, where she is currently
power supply, solar and fuel cell conditioning, be- an Associate Professor and works with the Group of
havioral modeling of converters and systems, fuel Power Electronics Systems. In 2016, she stayed as a
cell electric vehicle, power distribution systems for Visitant Researcher for three months at Tyndall National Institute, Cork, Ireland.
aircraft, and electromagnetic compatibility. He has authored/coauthored more She has authored/coauthored more than 100 scientific papers in international
than 200 scientific papers in international journals and conference proceedings conferences and journals, has filed four patents, and is the Co-Founder of a
and holds 12 patents. He has been actively involved in more than 85 R&D spin-off company targeting CAD for power electronics. Her research interests are
projects for companies in Europe and the USA, and more than 25 projects with contactless transference of energy, identification techniques applied to switching
public funding, in the field of power electronics. He is the founder of a spin-off power converters, design and modeling of high-frequency magnetic components,
company targeting to applications and CAD for power electronics. and design tools for dc/dc converters.

Authorized licensed use limited to: Harcourt Butler Technical University Kanpur. Downloaded on August 24,2023 at 06:33:46 UTC from IEEE Xplore. Restrictions apply.