Journal of Energy Storage 27 (2020) 101124

Contents lists available at ScienceDirect

Journal of Energy Storage

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

Hardware-in-the-loop Implementation of ANFIS based Adaptive SoC T

Estimation of Lithium-ion Battery for Hybrid Vehicle Applications
Krishna Veer Singha, , Hari Om Bansala, Dheerendra Singha
⁎

a
Power Electronics & Drives Lab, Department of Electrical and Electronics Engineering, Birla Institute of Technology and Science, Pilani, Rajasthan, India

ARTICLE INFO ABSTRACT

Keywords: This paper presents an effective method to estimate the state of charge (SoC) of a Lithium-ion battery. This
Open circuit voltage parameter is very crucial as it indicates the performance and health of the battery. The battery SoC estimation
state-of-charge equivalent circuit provided in MATLAB has been modified by adding the 3- RC pairs in series with its internal
state-of-health resistance. The values of the RC pairs have been calculated mathematically by solving the circuit model, based
lithium-ion battery
on charging and discharging dynamics of the battery. The values of these parameters have also been optimized
and hardware-in-the-loop and hybrid vehicle
using a “lsqnonlin” function. The SoC of the battery is estimated using the combination of coulomb counting and
open-circuit voltage methods to minimize the error in estimation. The obtained SoC is further corrected for
errors using ANFIS based algorithms. The effect of temperature has also been accounted for modelling the
battery and in SoC estimation. These obtained SoCs for 3 cases, i.e. without RC/with RC pairs and then tuned
with ANFIS based optimization are compared for the same load. The parameter calculation method adopted here
results in an efficient and accurate model that keeps track of correct battery SoC. The complete system is va-
lidated in real-time using hardware-in-the-loop laboratory setup.

1. Introduction density [2–5].

The batteries yield non-linear and complex characteristics making
The electric vehicles (XEVs) are gaining a lot of popularity, as they their behaviour unpredictable. The batteries undergo certain cycles
are the sustainable mode of transportation and are also the alternatives with consideration of environmental condition, therefore battery circuit
of the international combustion engine (ICE) based vehicles. The XEVs modelling is a prominent task in determining the accurate state of
are an effective way to overcome the fossil fuel crisis and environmental charge (SoC) and its performances. The model validation using para-
pollution, which are considered as the prominent hurdles for the au- meter optimization is an essential tool to accurately predict the beha-
tomobile sector. The battery management system is an integral part of viour.
any XEV. The more energy storage capacity means a higher driving The major objectives and outcomes of this paper are as below:
range of the vehicle and its better fuel economy. However, more ca-
pacity requires the bigger size battery which will increase the weight of 1 The modelling of the battery by adding the multiple RC circuits.
the vehicle and occupy more space, so a trade-off is needed between 2 Identification of time constant and then the calculation of the cor-
capacity and the size of the energy storage system (ESS). responding resistance and capacitances (RC).
There are various batteries available in the market, but the lithium- 3 Optimization of the obtained parameters from step 2 by “lsqnonlin”
ion batteries are gaining popularity[1]. They are equipped with several function provided in MATLAB.
advantages such as high specific energy, low self-discharge rate, high 4 Estimation of SoC using hybrid of coulomb counting and OCV
energy density, prolong life, high charge/discharge and high cell po- methods and then the obtained results were fine-tuned using ANFIS
tential, therefore, they have found a unique place in EV applications as based algorithm to minimize the error between proposed and rea-
ESS. Traction batteries need attributes like high energy density and listic data.
power density. However, it is also associated with some demerits such 5 Analysis of temperature variation on battery behaviour.
as high cost, safety issues, ageing and immature technology. The se- 6 Real-time validation of entire system on HIL testing platform.
lection of Li-ion batteries for traction application can be held true by
Fig. 1 where various batteries have been compared based on energy

⁎
Corresponding author.
E-mail address: [email protected] (K.V. Singh).

https://doi.org/10.1016/j.est.2019.101124
Received 25 July 2019; Received in revised form 14 November 2019; Accepted 30 November 2019
Available online 11 December 2019
2352-152X/ © 2019 Elsevier Ltd. All rights reserved.
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Nomenclature Vter Voltage across terminal of battery

Voc Open circuit volatge
ESS Energy storage system V0 Voltage across R0
OCV Open circuit voltage V1 Voltage across R1
SoC State of charge V2 Voltage across R2
XEVs (BEVs, HEVs & PHEVs) V3 Voltage across R3
SoH State of health R0 Internal resistance of battery
ANFIS An adaptive neuro-fuzzy inference system R1 Resistance across C1
ICE Internal combustion engine R1 Resistance across C2
ANN Artificial neural network R2 Resistance across C3
SoF State of function I Current flowing in the circuit
KF Kalman filter Tem Operating temperature
CC Coulomb counting C1 Capacitor of the branch R1C1
MHE Moving horizon estimation C2 Capacitor of the branch R2C2
GA Genetic algorithm C3 Capacitor of the branch R3C3
PSO Particle swarm optimization τ1 Time constant of the branch R1C1
BMS Battery management system τ2 Time constant of the branch R2C2
DEKF Dual extended Kalman filter τ3 Time constant of the branch R3C3
EKF Extended Kalman filter ToT0 and T0+ Initial condition
UKF Unscented Kalman filter V1zero, V2zero and V3zero Zero input response of voltages V1,V2 and
NN Neural network V3
EV Electric vehicle T1, T2 and T Various time instant of the waveform
SoH State of health a0 Battery terminal voltage when SoC = 0%
SoE State of energy a1 Battery terminal voltage when SoC = 100%.
PI Proportional integral Cp Battery capacity in Ah
FIS Fuzzy inference system. K0 Reaction constant
BP Back propagation R Gas constant
MF Membership function Ea Activation energy
DSO Digital storage oscilloscope pi Firing strength

1.1. Literature review on the SoC estimation methods estimation of SoC. Here, OCV was divided into three parts based on
charging, discharging, and neutral mode and were fed to NN for correct
There are various papers reported in the literature citing different estimation of SoC.
SoC estimation methods. Some recent papers have been presented in In [17], the reuse of EV batteries in the applications other than
this section. transportation has been suggested. The flow management of energy has
The SoC basically is the measure of the remaining capability of the also been explained. The author in [18] has discussed the role of the
battery. The various methods which have been proposed to estimate the dual EKF (DEKF) for battery SoC and state of health (SoH) estimation.
SoC can be categorized as: simple calculation methods, information- Based on the comparison of simulation results using KF and DEKF, they
driven strategies, and model-based methods. The simple calculation recommended DEKF. In [19], a new method of combining DEKF algo-
method can further be divide into impedance, open-circuit voltage rithm and charging voltage curve for SoC estimation has been proposed.
(OCV) and ampere-hour counting methods. The information-driven In [20], the adaptive SoC estimator has been proposed under various
strategies are alike simple calculation methods except they use in- assumption and condition. The error estimate in SoC is less than 3% for
telligent (neural networks, fuzzy logic, etc.) algorithms to train the both the constant current-voltage charging conditions and the dynamic
model. Model-based methods consider battery as a dynamic system and discharging conditions. In[21], the SoC and state-of-energy (SoE) are
involve state space to carry out battery modelling. This also uses ob- evaluated by the particles filter method and EKF by minimizing the
servers like extended Kalman filter (EKF) and unscented Kalman filter errors in voltage or current signal measurement. The [22] addresses the
(UKF) to estimate the state. Model-based techniques have some merits estimation of battery SoC from the joint perspectives of dynamic data-
like high accuracy, fast convergence, and a good computing power driven and model-based recursive analysis.
[6–15]. In [16], the author has used the neural network (NN) for In [23], the battery degradation capacity and the corresponding SoC
has been estimated by using three-dimensional response surface-based
battery OCV. In [24], a SoH estimation method has been proposed for
lithium-ion batteries. Here, OCV and Thevenin's equivalent circuit
model was used to improve the model accuracy and also to discuss the
relation between internal parameters and states of the battery. In [25],
a new kind of battery model which gets turned on/off based on the load
response has been proposed. The various switching algorithm for the
load has also been designed as well. The SoC and online model para-
meters co-estimation for liquid metal batteries has been proposed in
[26]. In [27], SoC estimation of battery has been carried by applying
two proportional-integral (PI) filters to minimize the sate error. In [28]
a comparative study of ANN/KF for on-board SOC estimation has been
carried out. They concluded that UKF presents the best overall perfor-
mance. In [29] the SoC estimation with accuracy has been proposed. An
Fig. 1. Comparison of various batteries based on energy density.
attempt to estimate the OCV and SoC of the battery during normal

2
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Table 1
Literature review of papers based on SoC, SoE, SoF, SoH, cell balancing and parameters estimation.
REF. PARAMETER ESTIMATED METHOD/APPROACH USED OUTCOME/ FINDINGS

[2] SoC Model-Based and Data-Driven Methods 1 . A review has been carried out on various model-based and data-driven
methods.
[6] SoC Ampere-hour and Kalman filter 1. The estimation error and the accumulated error of current is reduced to
obtain maximum accuracy.
[7] SoC and SoH OCV 1. SoC estimation accuracy has been improved compared to existing OCV
method.
[8] SoC Advanced binary search pattern for 1. This work deals with accuracy, efficiency, tolerance of disturbances and
impedance spectra temperature dependence of the methods which were proposed.
[10] SoC OCV and unscented Kalman filter 1 . A model has been developed which is based on the temperature effect to
estimate SoC.
[11] SoC Fuzzy based control 1 . A novel fuzzy model is proposed to characterize lead-acid battery
discharging.
2. An EKF based on a fuzzy model is proposed for SoC estimation.
3. The polarization resistance can be estimated using the algorithm proposed.
[12] SoC EKF 1. The average error of SoC estimation has been reduced by the proposed
algorithm.
[14] SoC EKF system based on a round-robin 1. The SoC has been accurately estimated based on voltage measurement.
approach
[16] SoC ANN 1. The proposed method results in a reduction of average estimation error.
2. The result depicts that the NN has boarder choice of training data and low
computational cost.
[18] SoC and SoH DEKF 1. The paper presents that the DEKF is responsible for increasing the accuracy
in SoC estimation as compared to EKF.
[19] SoC DEKF and Charging voltage 1. Based on the DEKF algorithm, the SoC is estimated. The results show that
the estimation error is within 3%.
[20] SoC Improved adaptive battery state estimator 1.An adaptive battery state estimator has been proposed for SoC estimation of
the lithium-ion battery.
2. The SoC estimation precision is evaluated under various operating
conditions.
3. Generality, robustness and dynamic characteristics are evaluated.
4. The error in SoC estimation is within the range of 3%.
[21] SoC and SoE The particle filter is used for simultaneous 1. Co-estimation of both SoC and SoE has been proposed.
SoC and SoE 2. The temperature effect was analysed to verify the robustness of the
proposed method.
[22] SoC A combination of symbolic time series 1. “Markov machine representation for dynamic data-driven analysis has
analysis and linear least-squares filtering been presented.”
[23] SoC and SoH GA (genetic algorithm) 1. “A three-dimensional response surface-based battery OCV model was
constructed.”
2. The proposed method has high accuracy and robustness.
3. The efficiency of online computation reported being increased.
[24] SoH PSO 1. The proposed method is compared with other methods and the advantage
of the PSO has been presented.
[25] SoC UKF and H-infinity observer for state 1. The proposed algorithm offers high accuracy under various load
estimation conditions.
2. An online optimization problem is formulated to find the best suitable
battery model.
3. Comparisons of the proposed algorithm with several existing ones are also
presented.
[26] SoC Adaptive unscented Kalman filter 1. Accuracy and robustness of SoC estimation have been proved to be higher.
[27] SoC Two proportional-integral filters 1. In order to compensate error two PI filters have been proposed.
2. In two RC equivalent circuit model of the battery, the presence of error
were analysed.
3. The fusion of the weights of the PI filter has been proposed as well.
[28] SoC ANN/KF 1. ANN and KF, SoC estimation methods are compared. ANN is proved
superior over KF.
[29] SoC ANFIS 1. SoC is compared to coulomb counting (CC) computation and ANFIS.
[31] SoC Hybrid algorithm with UKF 1. The SoC estimation has considered the effect of series batteries and the
differences between each cell i.e cell unbalancing
2. Error in the SoC estimation has been reduced.
[32] SoC and SoH KF 1. This paper describes the use of KF in order to estimate the parameters of
battery for photovoltaic applications
[33] SoC DEKF 1. KF overcomes the defects of the simplified battery modelling and separates
the sequence for estimation of the state and weight filter.
[34] SoC DEKF 1. Estimation of the remaining useful life with suitable margins of uncertainty
has been proposed.
[35] SoC/Capacity DEKF 1. DEKF algorithm combined with the pattern recognition based on the
hamming network for identification of suitable cell model parameters in
order to obtain improved SoC/battery capacity estimation has been
presented.
[36] SoC DEKF 1. Reliable SoC and battery parameter estimation algorithm with DEKF has
been proposed.
[37] SoC DEKF 1. This paper focuses on the behaviour of the modified DKF, based on the
state and parameter estimation and on the ideal tuning of the algorithm.
2. The algorithm enables an estimating error in SoC of less than 1.5%.
(continued on next page)

3
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Table 1 (continued)

REF. PARAMETER ESTIMATED METHOD/APPROACH USED OUTCOME/ FINDINGS

[38] SoC/ Capacity EKF 1. The proposed approach can estimate battery parameter, capacity and SoC
concurrently.
2. Less computation efficiency but higher estimation accuracy has been
suggested.
[39] SoC DKF 1. The behaviour of the DKF is analysed at various temperature.
[40] SoC KF 1. The benchmark proposed shows the same accuracy in the estimation of the
SoC with a one and two RC-term equivalent battery circuit.
2. The strong effect of temperature due to high-dynamic loads is observed.
3. By tuning individual filter, the temperature dependence is found to be
reduced.
[41] SoC, SoP and SoH DEKF 1. Online estimation of the battery's capacity and SoH
[42] SoC EKF 1. Parameterization and algorithm evaluation has been conducted.
2. Computational complexity is analysed.
3. Robustness and accuracy of the algorithms are compared.
[43] SoC GA 1. Multi-objective GA is utilized for extracting performance parameters.
2. Battery performance is accurately predicted once parameters are optimally
extracted.
[44] SoC ANFIS 1. Accurate estimation of the state of available capacity under variable
discharge currents.
[43] SoC and battery parameters GA 1. Battery charging/discharging characteristics are accurately predicted once
parameters are optimally extracted
[45] SoC EKF filter and least square identification 1. The results demonstrate the effectiveness of the fractional-order model and
method the estimation method.
[46] SoC NN 1. Increased SoC estimation accuracy and robustness by adding noise to
training data has been presented.
[47] SoC EKF 1. Accurate SoC for battery in automotive applications.
[48] SoC and Cell balancing AEKF 1. To obtain the OCV a data-driven filtering process is proposed
2. Parameter uncertainties are developed and analysed.
3. An adaptive SoC estimator which reduces the cell unbalancing.
[49] SoC ANFIS in combination with CC 1. SoC is estimated with a mean error of 4% and a maximum error of 7% in a
realistic driving scenario.
[50] SoC EKF 1. The proposed SoC estimation is accurate in terms of the error in estimation.
[51] SoC EKF and ANFIS 1. EKF and ANFIS have been compared.
2. It has been concluded that ANFIS-based SoC estimation has greater
precision and don't require iterations.
[52] SoC CC and SoC-OCV relationship 1. The CC algorithm has the drawback as it is associated with cumulative
errors and measurements uncertainties,to avoid them, the OCV has been
used, thus a linear SoC-OCV relationship has been established and
recalibration of the battery capacity periodically has been conducted.
[53] SoC EKF 1. Accurate estimation of SoC has been presented.
[54] SoC ANN and ANFIS 1. Comparisons of the two approaches have highlighted the potential of
ANFlS in modelling and prediction of the behaviour of complex nonlinear
dynamic systems.
[55] SoC AEKF 1. SoC estimation error is less than 2% compared to EKF.
[56] SoC ANN 1. The proposed technique showed that very accurate SoC estimation results
can be obtained provided enough training data are used to train the ANN
models.
[57] SoC OCV 1. The superiority of the adaptive SoC estimation which is based on the
parameter identification algorithms and SoC estimation schemes over the
existing methods has been demonstrated in terms of accuracy and robustness.
[58] SoC KF 1. The fractional-order model has higher accuracy in SoC estimation, and the
error can reach up to 0.5% SoC.
[59] SoC UKF and H-infinity filter 1. The joint SoC estimation method has high accuracy, fast convergence,
excellent robustness and adaptability.
[60] SoC Split battery model 1. The cross-interference between RC voltages and SoC is reduced, which
effectively reduces the oscillation in the estimation and improves the
estimation accuracy
[61] SoC and Cell balancing Quasi-sliding mode 1. Bidirectional Modified ´Cuk converters are utilized as the cell equalizers
due to their advantages of integrated infrastructure and modular design.
[62] SoC, SoH and SoF EKF 1. The continuous online upgrading of battery capacity and its related
parameters helps in accurate state estimation.
[63] SoC Review 1. This paper presents the reviews the SoC estimation methods that are
suitable for online usage and classify them into five categories.
[64] SoC CC, KF and OCV 1. Hybrid (which is a combination of CC and OCV SoC estimation technique
to improve accuracy under varying conditions.
[65] SoC A joint moving horizon estimation (joint- 1. Compared to joint EKF, the proposed approach based on MHE can offer a
MHE) approach more reliable, robust, and accurate SoC estimation.
[66] SoC EKF and AEKF 1 . A moving window technology and partial least squares (PLS) regression
can establish a series of piecewise linear battery models automatically.
2. Results have been compared with EKF and Adaptive-EKF (AEKF).
[67] SoC Gravitational search algorithm 1 . A detailed comparative study between the proposed model and existing
SoC strategies is conducted, which also demonstrates the superiority of the
proposed.

(continued on next page)

4
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Table 1 (continued)

REF. PARAMETER ESTIMATED METHOD/APPROACH USED OUTCOME/ FINDINGS

[68] SoC Long Short-Term Memory Network 1. The proposed method have the advantage of quick convergence to the true
SoC, with root mean square errors within 2% and mean average errors within
1%.
[69] SoC Gaussian process regression and 1. The proposed method reports great improvement in SoC estimation with
polynomial regression the bias characterization.
2. The polynomial regression model and the gaussian process regression
model are proposed to examine the effects of the two methods on bias
modelling and SoC estimation using a typical battery circuit model.
[70] SoC SoC estimation strategies 1. Review on SoC estimation strategies for HEVs and EVs
[71] SoC and cell balancing ANFIS 1. The proposed method has the potential to overcome the drawbacks of
traditional SoC estimators caused by cell-to-cell variations in a battery pack,
and the corrected SoC is more reasonable than the traditional “averaged
SOC.”
[72] SoC Estimation strategies 1. The paper presented the common methods currently adapted by
researchers in estimating battery SoC by analysing their pros and cons.
[73] SoC Estimation and management strategies 1 . A review of lithium-ion battery SoC estimation and management system in
EV applications has been proposed.
[74] SoC OCV 1. An algorithm to determine the battery system's impedance and SoC are
presented including parameter and state estimation techniques.
[75] SoC KF 1. An adaptive co-estimator is presented to update pack parameters in dual
time-scale using an optimized recursive least squares algorithm and to
estimate the battery pack SoC using an EKF.
[76] SoC, capacity, impedance parameters, Review on monitoring of lithium-ion 1. The methods for monitoring of the battery life are reviewed with the focus
available power, SoH, and remaining useful batteries in electric and hybrid vehicles on the elaboration of their strengths and weaknesses for the use in on-line
life. BMS applications.
[77] Characterization of lithium-ion batteries Electrochemical impedance spectroscopy 1. The cell impedance is presented and analyzed and strong nonlinear
temperature correlation has also been demonstrated for Li-ion battery
[78] On-line adaptive impedance and state Butlere volmer kinetics on the charge- 1 . A critical review of methods, modelling and requirement has been
estimation of battery. transfer process presented.
[79] Cell balancing chain structure of the switched capacitor 1. To improve cell balancing speed, two circuits with chain structure are
proposed to increase balancing speed, particularly among outer cells.
[80] SoC and battery parameter estimation DEKF 1. Compared to the other estimators the proposed estimator is found to be
between 16.4% and 28.2% more accurate for SoC estimation.
[81] SoH DEKF 1. The DKEF estimate the parameters of equivalent circuit network and
maximum available power of the battery at any specified instant irrespective
of the age of the battery.
[82] SoC ANFIS 1. It has been proposed that the identification error can change unknown
battery parameters.
2 . A novel analytical formula for identification has also been presented..
[83] SoC and battery parameters 1. The results demonstrate that even with a high noise level in measured data,
the proposed identification and estimation algorithms are able to work well
in real-time.
[84] Battery capacity, resistance and power Reference performance test Methodology 1. The reference performance test methodology is typically used to evaluate
capability. the performance of the batteries in terms of capacity, shuttle current,
resistance and power capability.
[85] Battery parameters and SoC OCV and Prediction-error minimization 1. An SoC observability analysis was also performed for both case studies,
algorithm NiMH and Li-S. The results show that the system is observable in the case of
NiMH whereas it is not fully observable for Li-S because of its flat OCV–SoC
curve.
[86] SoC Bayesian filters, EKF, UKF and a particle 1. The estimators (Bayesian filters, EKF, UKF and a particle filter) are tested
filter. through practical experimentation.
[87] Battery degradation Reference performance test Methodology 1 . A methodology for a reference performance test for the Li-S batteries is
proposed.
[88] Review on batteries Review paper 1. The transition from the use of lithium-ion to lithium-sulphur battery has
been proposed in the form of review.

operation of EV was made by means of charge/discharge current and individual cell.

voltage in [30].
The various paper related to SoC, SoH (state of health), SoF (state of 1.2. Fixing the number of time constants (RC branches)
function), SoE (state of energy), cell balancing and parameters esti-
mation of battery has been listed in Table 1. It has also been observed that many of the researchers have used
From the above literature, it has been concluded that the major different combination of RC circuits in battery modelling (Table 2). To
challenges which make battery inefficient are as follows: cell un- accurately estimate the SoC, the fixing of RC circuits in battery model is
balancing, battery modelling, life of a battery, the effect of temperature, a must and the same is explained below.
self-discharge, rate of charge and discharge, challenges of monitoring In Fig. 2, five different time constant combinations have been con-
battery health, estimation of maximum capacity and communication sidered. The OCV is compared for these five combinations (1/2/3/4/5
method. RC pairs respectively) and for a realistic battery. The zoom-in a portion
Out of these challenges, the battery modelling and temperature ef- of the waveform has been shown on the right side so that the clear
fect has been taken into consideration for our work and for cell bal- waveform of all five times constants can be seen. In Fig. 2, it can be
ancing it has been assumed that the cell equalizer has been in- clearly seen that out of five combinations, the one with single RC pair is
corporated with BMS which will uniformly distribute the SoC of an not at all acceptable. Out of remaining combinations, the blue dotted

5
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Table 2
The various RC model reported in the literature review.
Reference Model used Estimate Initialization (Tuned/Measured) Year

[32] Ro model State/parameters Tuned 2007

[33] 1-RC model State/parameters N.S 2008
[34] 2-RC model State/parameters N.S 2009
[35] 1-RC model State/parameters N.S 2013
[36] Ro model State/parameters N.S 2013
[89] 3-RC model State/parameters Tuned 2013
[37] 1-RC model State/parameters Measured 2014
[38] 1-RC model State/parameters N.S 2014
[39] 1-RC model State/parameters N.S 2014
[40] 2-RC model State/parameters Tuned 2016
[41] 2-RC model State/parameters Tuned 2016
[90] 2-RC model State/parameters Measured 2017
[42] 2-RC model State/parameters Tuned 2018

N.S*: Not Specified.

Fig. 2. Time constant comparison by curve fitting method.

line (3-RC pair) is in close resemblance with the data of the OCV (green 3 The obtained parameters are optimized by means of “lsqnonlin”
line). This also provides a minimal error in SoC estimation with realistic function.
data. Therefore 3- RC combination is chosen here which is less explored 4 A combination of coulomb counting and open-circuit voltage
in the literature. methods is used to estimate the SoC.
5 The obtained SoC is further corrected (calibrated) using ANFIS
1.3. Paper organization method.
6 The effect of temperature variation on battery behaviour is ana-
This paper has been organized in various sections where section 1 is lyzed.
an introduction, literature review, paper organization, and Novelty. 7 Real-time validation of a complete system using dSpace
Section 2, establishes the Thevenin equivalent circuit model with 3-RC MicroLabBox.
for Li-ion battery. The detailed procedures for identifying the model
parameters and SoC estimation are also presented in this section.
Section 3 presents the simulations results and discussion. Section 4 2. Equivalent circuit model
gives the ANFIS optimized SoC. Section 5 presents the variation of SoC
and OCV with temperature. Section 6 reveals the real-time validation In this study, a Li-ion battery model is established. A reliable and
and hardware setup, Section 7 is the validation of the proposed work in
ADVISOR tool and Section 8 draws the conclusion.

1.4. Novelty or peculiarity of the paper

1 3-RC model of battery has been proposed, which has not been much
explored.
2 The various parameters of the battery, namely (R0, R1, C1, R2, C2,
R3, and C3) have been estimated. Where R0 is the internal resistance,
R1, R2, & R3 and C1, C2, & C3 are the resistance-capacitance pairs, as
shown in Fig. 3.
Fig. 3. The equivalent circuit battery model.

6
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

accurate battery model is required to estimate battery SoC while using response
the model-based method. The third-order RC model of battery is crafted
here in order to imitate the behaviour of the real battery. The model V1zero = V1 (T1)
considered in this work consists of DC source, Ro, i.e. internal resistance V2zero = V2 (T1)
and 3-parallel RC circuit in series as shown in Fig. 3. V3 zero = V3 (T1) (9)
In a Li-ion battery, Li-ions move from anode to cathode through an
organic electrolyte during discharging. During charging the reverse From Eq (3), (4) and (5)
process is followed. Anode and cathode have a layered structure and Li- T T1
ions move in between layers. The governing chemical reactions are as V1 (T T1) = V1zero e 1

follows:
T T1
V2 (T T1) = V2zero e 2
T T1
Atcathode: Li1 x CoO2 + xLi+ + xe LiCoO2 V3 (T T1) = V3zero e 3 (10)
Atanode:Lix C6 xLi++xe + 6C
Since I = 0, from (1) and (2)
The specification of the battery used has been shown in Table 3.
Vter = Voc V1 (T T1) V2 (T T1) V3 (T T1) (11)
2.1. Identification of battery model parameter
Considering H = V1 + V2 + V3
This yields
The battery performance depends on three main parameters, i.e.
SoC, operating temperature, and ageing. The battery model parameter T T1 T T1 T T1
H (T T1) = V1zero e 1 + V2zero e 2 + V3zero e 3 (12)
identification is obtained by the discharging/charging current pulse
experiments, and the specific experimental steps are described as fol- From Eq (10)
lows:
H (T T1) = Voc Vter (T T1) (13)
1 Rest the single battery for more than 24 hours, to ensure that the
To establish a regression equation that can be used for parameter
polarization voltage of the battery is reduced to zero.
identification, the following two variables are taken from [93].
2 Discharge the single battery with a constant current.
3 Rest the battery at the end of the experiment. T
X (T T1) = H( T1) d
From Fig. 4, it is clear that the response voltage has a downward T1 (14)
jump to discharge the battery once the battery is at rest for long en-
T
ough. Similarly, the response voltage has an upward jump for charging Y (T T1) = X( T1) d
the battery. The voltage drop due to Ro is sudden, whereas it is gradual T1 (15)
for the RC combination.
According to the principle of the circuit, the calculation of ohmic
X (T T1) = (V1zero 1 + V2zero 2 + V3zero 3)
resistance could be obtained by the formula T T1 T T1 T T1
(V1zero e 1 + V2zero e 2 + V3zero e 3 )
(16)
Vter = VOC V0 V1 V2 V3 (1)
Where Vter is the terminal voltage, Voc open-circuit voltage, Vo is the Y (T T1) = (V1zero 1 + V2zero + V3zero 3)(T T1)
2
voltage across the resistance Ro, V1 voltage across resistance R1, V2 2 2
voltage across resistance R2 and V3 voltage across resistance R3 (V1zero 1 + V2zero 2 + V3zero 32 )
2 T T1 2 T T1 2 T T1
V0 = IR 0 (2) + (V1zero 1e 1 + V2zero 2e 2 + V3zero 3e 3 ) (17)

dV1 V By using Eq. (13), (16) and (17)

C1 + 1 =I
dt R1 (3)
Y (T T1) = X (T T1)( 1 + 2 + 3) H (T T1) 1 2 3
dV2 V
C2 + 2 =I + (V1zero 1 + V2zero 2 + V3zero 3)(T T1)
dt R2 (4)
+ (V1zero + V2zero + V3zero) 1 2 3 (18)
dV V
C3 3 + 3 = I When these expressions are substituted into the third equation, a
dt R3 (5)
regression equation is obtained as follows, which has no exponential
The time constant of the circuit can be expressed as terms in it:
1 = R1 C1
Table 3
2 = R2 C2
Specification of the Li-ion battery [91,92].
3 = R3 C3 (6)
Commercialization year 1992
Determination of Voc: Since the battery has been at rest for suffi-
ciently long and the test starts from zero state, we have: Cell voltage 3.7
Energy by weight (Wh/kg) 90-180
VOC = Vter (T0 ) (7) Specific power (W/kg) 760
Energy by volume (Wh/L) 220 –350
The R0 can be obtained by Maximum discharge rate 40C
Recharge time < 3 Hours
Vter (T0 ) Vter (T0+) Cycle life cycle 1200
R0 =
I (8) Self-discharge per month 5% – 10%
Temperature range 0C -20°C to + 60°C-
Where T0 and T0+represent
the initial conditions Preferred charge method Constant voltage and constant current
The time duration from the T0+ to T2 has been used to calculate the Average energy cost $250
three time constants. During this period the circuit is in zero input

7
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Y (T T1)
=[ X (T T1) H (T T1) (T T1) 1]
1+ 2 + 3
1 2 3
V1zero 1 + V2zero 2 + V3zero 3
(V1zero + V2zero + V3zero) 1 2 3 (19)

The time constants and initial states V1zero, V2zero and V3zero can be
calculated by Eq (19) by considering all the sampling point from T1+ to
T2
Then from Eq. (19) we have

Y1 X1 H1 T T1 1 1 + 2 + 3
1 2 3
Y2 = X2 H2 T2 T1 1
… … … … … … …… … V1zero 1 + V2zero 2 + V3zero 3
Yn X2 Hn Tn T1 1 (V1zero + V2zero + V3zero) 1 2 3 (20)

Y1 X1 H1 T T1 1
B= Y
…
2 A = X2 H2 T2 T1
…… …… ……
1
… P
Yn X2 Hn Tn T1 1
Fig. 4. Discharge test at constant current pulse.
1 + 2 + 3
1 2 3
= V
1zero 1 + V2zero 2 + V3zero 3
(V1zero + V2zero + V3zero) 1 2 3 (21)

The Eq. (20) can be written as

B = AP (22)

By least squared error solution

P = (AT A) AT B (23)

T1
V1zero = IR1 (1 e 1)
T1
V2zero = IR1 (1 e 2)
T1
V3zero = IR1 (1 e 3)
(24)

Since V1zero and V2zero are already obtained, R1 and R2 can be cal-
culated from (24) and subsequently, C1 and C2 can be calculated from
(6).

2.2. State of charge estimation:

The output voltage can be calculated as [94]

Vter = Voc (t )
R1 + 2R0 ( t1 ) R2 + 2R 0 ( t2 )
i (t ) R 0 e e
R1 C1 R2 C2

R3 + 2R 0
e ( t3 )
R3 C3 (25)

A) SOCV Calculation: Cell voltage when all the reactions are ba-
lanced, is called equilibrium voltage, which is occasionally referred to
as OCV. With this OCV, voltages based SoC (SoCv) can be estimated
using [64,95–97]
1
SoCv = (OCV a0)
a1 (26)
Fig. 5. Flow chart of estimation of parameters.
Where a0 is battery terminal voltage when SoC = 0% and a1 is
battery terminal voltage when SoC = 100%. But, due to change in
temperature, equilibrium voltage of the battery at any temperature
(Tem) gets changed as
dV
V (Tem) = V298 + ( )(Tem 298)
dTem (27)
dV
is temperature coefficient and is constant for the considered
dTem

8
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Fig. 6. The data of the battery and the located pairs due to R0 and RC.

Fig. 7. The real-time data and simulated data of battery.

temperature range. So the consideration of this OCV with temperature depends on temperature. As for every 10 °C temperature increase, the
effect will lead to modifying SOCV and will contribute in the final SOC current gets doubled so for ΔTem temperature change, final equation
calculation. can be written as.
B) SoC Calculation: The coulomb counting (CC) method involves the
current integration flowing through the battery to get SoC I(s d) =
1
K 0 (Tem) e (ln 2( 10R ) ) (Tem×(Tem+10))
cp (30)
1
SoCi = cp Idt
cp (28) Temperature effect and self-discharge current are included for
analyzing run-time characteristics. A number of Li-ion cells are con-
Cp is battery capacity in Ah. Due to change in temperature, the cell nected in series and in parallel to simulate a powerful battery to be used
reaction rate gets changed. The Arrhenius equation, the reaction rate is in hybrid vehicles. The output voltage is calculated as Eq.(25) where V
given as (t) is time-dependent OCV, i(t) is the time-dependent current, and
Ea multiplier term associated with i(t) represents the impedance offered.
K = K0 × e R × Tem (29) From Eq. (26), temperature-dependent voltage-based SoC, that is,
SoCV(Tem), can be intended as Eq. (31) and from Eq. (28) temperature-
K0 is reaction constant, R is gas constant, Ea is activation energy, dependent current based SoC, that is, SoCi(Tem), can be calculated as
and Tem is operating temperature. During the electron transfer reac- Eq. (32)
tion, electrons require the additional amount of energy to surmount the
energy barrier called the activation energy (Ea = J⋅mol−1) which

9
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Fig. 8. The parameter obtained for the 3-RC circuit.

Fig. 9. The optimized parameter obtained for the 3-RC circuit.

SoCv (Tem)

Vter (t ) + i (t ) × 2( ) + I(s
1 Tem
= 10 d) (t ) × R 0
a1

R1 + 2R0 ( t1 ) R2 + 2R 0 ( t2 )
e e
R1 C1 R2 C2

R3 + 2R0
e ( t3 ) +
dV
(Tem 298) a0
R3 C3 dTem (31)

Fig. 10. The new and original SoC comparison and their variation with respect
to time.
SoCi (Tem ) =
1
cp
cp { i (t ) × 2(
Tem
10 ) + I(s d) }
(t ) dt
(32)

3. Simulations Results and Discussion

The SoC indicates the power available in a battery and its correct
information is of utmost importance. The parameters namely resistance
and capacitance, play a vital role in system dynamic performance so to
accurately estimate the SoC, these parameters are first optimized.
The procedure of parameter estimation starts with the data

10
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Fig. 11. The MATLAB/Simulink circuit used to simulate and analyze SoC.

Fig. 12. Structure of ANFIS.

collection of the battery, which involves OCV and charge/discharge with the help of a lookup table. After that the OCV and current dis-
current pulse data collection. The collected data is processed to identify charge pulse have been tracked with the SoC at the same time. The pairs
the effects of Ro and the RC pairs on OCV while tracking the charge/ due to Ro and RC also need to be identified for a sampling of the data, as
discharge current pulse. The OCV is sampled so that its magnitude can shown in Fig. 6:
be located at various time intervals concerning the current pulse. Then The difference in the behaviour of the real data of a battery and the
parameters are mathematically estimated by the Eq.(1) to Eq.(24). The simulated battery, which has been demonstrated in Fig. 7:
obtained parameters are again optimized with the help of a function By locating the pairs and by sampling the OCV and corresponding
called “lsqnonlin” provided in the MATLAB. The obtained parameters current pulse for the time interval T0 to T2, the parameters can be
are put up in the battery block, and an accurate SoC can be obtained. calculated using the equations described in section 2. The obtained
The obtained SoC with the 3-RC circuit is compared with the original parameters have been shown in Fig. 8. The time constant and resistance
circuit (circuit having no change). The procedure of the parameter es- are calculated by applying the equations mentioned in section 2 and are
timation of the 3-RC circuit has been shown in Fig. 5 in the form of a given below.
flowchart: The obtained parameters have been optimized utilizing “lsqnonlin”
The simulation has been carried out in MATLAB/Simulink, and the function provided in MATLAB. The values have been provided using the
temperature has been kept 303° K to 315° K for the whole experiment. graphs shown below in Fig. 9:
The data of a real battery has been collected and is fed in the MATLAB The obtained parameters have been fed to the modified block of

11
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

consideration, as the authors can generate the data set. As expected, the
ANFIS optimized SoC provides lesser variation in parameters during
dynamically load change and lesser settling time. The real-time ver-
ification ensures that this can be implemented practically. ANFIS is a
hybrid of ANN and FLC, which enjoys the benefits of both. A dynamic
and parallel processing system estimates input-output functions. A
fuzzy method adaptively deduces and alters its fuzzy relationship from
representative mathematical samples. The NN, on the other hand, can
erratically create and refine fuzzy rules from the training data. The
fuzzy sets are thought to be constructive in the intelligent field and in
dealing with higher processing. The performance of FLC depends on the
rule basis, number, and forms of membership functions (MFs). In this
work, various MFs like triangular, trapezoidal, and Gaussian are eval-
uated. Out of these, triangular is chosen here as it offers minimum error
and less computational burden. The authors have tested the system with
fuzzy MF rules like 3*3, 5*5, and 7*7, but they all yield the same result
and settling time. Therefore, 3*3 rules are chosen here because of the
lesser computational burden. The fuzzy input, along with triangular
MF, is fed into the NN block. The NN block consists of a rule base,
which is connected to the fuzzy inference system (FIS). Here, back
propagation (BP) algorithm issued to train the FIS, and the BP learning
algorithm has been used for training feedforward NNs. The most
commonly used FIS are Mamdani and Sugeno. In this work, the Sugeno-
type FIS is used because it is computationally efficient than the
Mamdani type, and can be trained by real data as opposed to the latter,
which is more susceptible on an expert system. Also, the output MF of
the Sugeno system is either linear or constant. Takagi Sugeno FIS is
executed here using ANFIS and has a five-layered structure, as shown in
Fig. 12.
Fig. 13. Flowchart showing the operation of an adaptive neuro-fuzzy inference
Fuzzification of the input variable has been done in the first hidden
system (ANFIS) algorithm. layer and T-standard operators to compute the rule are conveyed to the
second hidden layer. Third and fourth hidden layers standardize the
rule strengths and determine consequent parameters of the rule, re-
3-RC pair circuit and are compared with the SoC of the original circuit,
spectively. The fifth layer computes the overall input by summing all
as shown in Fig. 10 below. In the Fig.10 the new SoC is the SoC ob-
the incoming signals. ANFIS architecture is an adaptive, feedforward
tained by modifying the battery model by incorporating 3-RC branches
network structure. Every node plays out a specific function on input
and original SoC is the SoC of the battery without any modification or
signals as well as on the set of parameters of that node. Some of the
for default battery model. The curves show the real value curve of SoC
nodes are adaptive, which means that the node parameters are de-
estimation. The coulomb-counting and OCV methods have been used to
pendent on the other nodes; whereas others are fixed nodes, where the
estimate the SoC.
node parameters are independent.
The MF chosen should be able to minimize the error between the
4. The ANFIS based Optimized SoC actual output data and ANFIS input sketched output data. The para-
meters connected with the MF will change within the learning process.
The obtained parameters have been installed in the circuit shown in Layer 1: This layer contains two inputs, one as error and other as a
Fig. 11 and are used to obtain the new SoC. The obtained SoC has again change in error represented by x and y respectively. It is commonly
been optimized by ANFIS, as shown below: known as the fuzzification layer [98].
ANFIS algorithm: The ANFIS has many advantages like it does not Layer 2: Every node of the 2n (the number of fuzzy if-then rules) is
necessitate the use whole mathematical model only input-output data fixed and labelled as Π. It multiplies the input signal and forwards and
are required generated from the simulation. The rousing features of the outputs the product as given by Eq. 33
NN are its ability to learn and adapt, whereas the ability of a fuzzy
system is to consider imprecision and prevailing uncertainties. In order pk = ai (x ) × bi (y ) (33)
to exploit the advantages of both methods, an adaptive neuro-fuzzy
Where k = 1, 2, 3… 9. The output of each node represents the firing
inference system (ANFIS) algorithm comprising of both the NN and the
strength pi (i = 1 to 2n) of the i the fuzzy if-then rule given by Eq. 34 as
fuzzy logic is adapted for the proposed work. The ANFIS do have a high
computational cost but due to the redundancy, complementarity, het- pi
pi =
erogeneity and its performance make it a fascinating tool for such case. 2n
p (34)
i=1 i
As the learning rate of the NN is relatively low, which makes it less
suitable as a stand-alone technique for time-critical applications, in- Layer 3: Each node in this layered circle calculates the ratio of the
tegrating it with fuzzy logic brings a better real-time response. individual rule's firing strength to the sum of all rules given by Eq. 35,
The EV applications deals with high uncertainty of the power de- as shown in Fig. 13.
mand, temperature, and load variations. Concurrently, the system The ANFIS model for the controller design uses the architecture of
should also be designed in such away so that it could be adaptive and the adaptive network. Fig. 12 shows the flow chart for the working
responsive to these changes. Therefore, application of ANFIS is appro- operation of the ANFIS algorithm. The ANFIS model is designed,
priate for the proposed system. Further, the literature survey establishes trained, and analyzed using MATLAB 2017b Neuro-Fuzzy Designer
that ANFIS results in accurate modelling of the system provided a good tool-box and graphical user interface (GUI).
data bank is available. This is the icing on the cake for the system under A data set is collected and stored in the MATLAB workspace to train

12
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Fig. 14. A) Training data, B) Epochs and test data error, and C) Training data testing. The predicted output is indicated by red, and the test data is represented by blue
circles.

Fig. 15. A) SoC comparison with 3-RC and ANFIS optimized & without 3-RC circuit B) OCV versus SOC curve.

the ANFIS. The ANFIS is tested with different MFs such as triangular, individual output values “f” from the inferring of the rule base. Each
trapezoidal, and Gauss MFs, and it is found that the training error is node in this layer is a square node, with the function as given by Eq 36
minimum with triangular MF. The epochs (iterations) and error are
4
shown in Fig. 14 i = pi fi = pi (gi x + h1 y + ri), (36)

3
pi Where i = 1, 2, 3 and pi is the output of layer 3 and {g1, h1, ri} is a
= pi = , i = 1, 2, 3.
p1 + p2 (35) parameter set. Parameters in this layer are also referred to as con-
sequent parameters.
Layer 4: This layer is called a defuzzification layer. It calculates the Layer 5: This layer is called as the output layer. There is a single

13
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Fig. 16. Effect of temperature on SoC and voltage of battery A and battery B.

circle of the output neuron labelled as Σ, and the output signal is ob- Step 7: Implement the “. fis” in the fuzzy toolbox and obtain the results.
Step 8: Implement the same work in hardware to check its validity in real-time.
tained by adding up input signals as it can be seen in the following
Step 9: The algorithm uses an ANFIS based system for fusing the data from multiple
Eq 37 sources.
The two inputs are given to the ANFIS based algorithm by means of taking the Input/
5 ipi fi
= i pi fi = ipi, Output data in the workspace of the MATAB tool. The generated “. mat” file is
i
(37) loaded in the ANFIS. The training has been carried by setting the epoch value and
MF is generated using “trimf.” The output which is “. fis” file is loaded in a fuzzy
Comparison of simulated SoCs: Fig. 15 (A) provides the comparison of
toolbox in the simulation model then the output is taken, and the same is checked
estimated SoCs, where red line indicates estimation without RC, and the in the HIL.
blue line indicates with the 3-RC circuit. The circuit with 3-RC involves
more losses due to the impedance whereas in another case losses due to
internal resistance are only considered. The results obtained using RC 5. Variation of SoC and OCV with temperature
pairs are closer to that of real data available in [38]. Therefore, the
proposed approach provides a faster and accurate estimation of SoC in Two batteries, namely, A and B, have been taken with the 3-RC
real-time. Fig. 15 (B) provides the OCV Vs SoC curve of the battery. circuits. The behaviour of battery A has been captured for temperature
This paper presents the design of a data fusion algorithm based on variations from 40 °C to – 40 °C as shown in Fig. 16 below. The effect of
ANFIS and provides minimal root mean square error and mean absolute temperatures is neglected in Battery. The current pulse of 2A has been
percentage error. The redundancy, complementarity, timeliness, het- considered for this purpose.
erogeneity etc. are the advantages of the systems based on data fusion
algorithms. As the learning rate of the NN is relatively low, which 5.1. Simulation and analysis
makes it less suitable as a stand-alone technique for time-critical ap-
plications, integrating it with fuzzy logic brings a better real-time re- At t = 0s, the Battery A and B are discharged with 2A at an ambient
sponse. The fuzzy logic creates a fuzzy engine based on the rules de- temperature of 40°C.
rived from the input/output data set. The neural network trains the At t = 150s, the internal temperature has increased to its steady-
engine for more precision. The steps of the algorithm have been pre- state value of 42.2°C due to heat losses from the discharge process. This
sented below: causes the output voltage of Battery A to slightly increase, while battery
B output voltage continues to decrease.
At t = 1000s, the ambient temperature is decreased to -40°C. This
Algorithm
Step 1:Collect the data from individual sensors. causes the output voltage of Battery A to greatly decrease as the internal
Step 2: Create a fuzzy engine with pre-determined input/output variables using A- temperature decreases rapidly. Also, the SoC of Battery A decreases due
NFIS. to the reduction of its capacity. The battery B output voltage continues
Step 3: Train/re-train the fuzzy engine until there is no improvement in the accuracy. to decrease slowly to its steady-state voltage.
Step 4: Train the engine to obtain the minimum root mean square error.
Step 5: If No, go to step 4.
At t = 2000s, the ambient temperature is increased from -40°C to
Step 6: If Yes, stop training further and save the “. fis” file. 0°C. As the internal temperature increases, the output voltage of Battery
A increases. Also, as the capacity increases, the SoC of Battery A

14
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Fig. 17. A) Hardware setup and B) SoC comparison.

increases. The Battery B output voltage remains constant to its steady- calculations and to write all outputs. The interval size is known as the
state value. step size, Ts. Fixed-step solvers solve the model at regular time intervals
At t = 2500s, the Battery A and B are charged with 3A at an am- from the beginning to the end of the simulation.
bient temperature of 0°C. This causes the internal temperature to in- Physical test system setups are large, costly, and required highly
crease due to heat losses during the charging process, which increases experienced people to handle the repetitive occupations of setting up
the charging voltage of Battery A. Afterwards, Battery A and B continue networks and maintaining extensive inventories of difficult hardware.
to charge up until fully charged. With the improvement of microprocessor and floating-point digital
The variation of SoC has also been taken care of in the developed signal processing technologies, physical test systems have been re-
design. placed with completely real-time simulator. The HEVs are highly non-
linear and complex in nature, and testing on the actual network is very
6. Hardware setup and Real-time results expensive, tedious, and risky. Hence, a speedier approach to test and
validate a new algorithm or research in a real-world environment using
Real-time is often used to describe the time-critical technology in real-time simulator is an essential part of the design and improvement
various industries. MicroLabBox uses real-time in reference to an em- of the system.
bedded system. These systems are devices which interface with the real There are various types of real-time simulators; the popular real-time
world and provide control. The embedded device is given a pre-set platforms are Opal-RT, Typhoon HIL, and MicroLabBox (d-SPACE). In this
time, like 1,5,20ms to read input signals, to perform all necessary work, authors have used MicroLabBox hardware controller.

15
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

Fig. 18. SoC comparison.

The laboratory setup of HIL has been presented in Fig. 17 (A). The was minimized using ANFIS based. The system has been tested on HIL
whole system has been simulated in the MATLAB, and then it is burnt in platform by applying MicoLabBox hardware controller. The HIL results
the MicroLabBOX to verify the results obtained in real-time. The wa- are very close to the simulated results. Proposed SoC estimation shows
veform in Fig. 17 (B) shows two different lines corresponding to two very little deviation from the real battery data, thus verifies the utility
SoCs. The blue line is the SoC due to the original battery (without RC) of the proposed design in real-time systems.
and the yellow line shows the ANFIS optimized SoC of the 3-RC circuit.
The results obtained on the HIL platform are in line with the simulated Declaration of Competing Interest
results. Further, the results obtained using ANFIS based algorithm is
very close to the real data, thus verifies the proposed design. In the All authors have participated in (a) conception and design, or
MicroLabBox it is difficult to show the exact waveform as the signal analysis and interpretation of the data; (b) drafting the article or re-
division is quite large and digital storage oscilloscope (DSO) can only vising it critically for important intellectual content; and (c) approval of
display till M 50s the signal magnitude is further divided by 100so that the final version. This manuscript has not been submitted to, nor is
it can be step down and can be displayed in DSO in real-time. The under review at, another journal or other publishing venue. The authors
MATLAB file needs to undergo some setting like fixed solver is selected, have no affiliation with any organization with a direct or indirect fi-
step time needs to be fixed and will have to be kept the same for the nancial interest in the subject matter discussed in the manuscript
whole system.
In figure 17(B) the original SoC indicates the default SoC of the Reference
battery i.e battery without any modification which is indicated by the
blue line and the SoC of the battery with ANFIS with the 3-RC and has [1] H. Budde-Meiwes, J. Drillkens, B. Lunz, J. Muennix, S. Rothgang, J. Kowal,
been indicted by the yellow line which runs parallel to blue line but D.U. Sauer, A review of current automotive battery technology and future pro-
spects, Proceedings of the Institution of Mechanical Engineers, Part D: Journal of
lesser in magnitude. The result shown in 17(B) is only for 500s as the Automobile Engineering 227 (2013) 761–776, https://doi.org/10.1177/
sampling time of the DSO is 50s so in display at a time in one screen it 0954407013485567.
can show result for 500s. [2] D.N.T. How, M.A. Hannan, M.S. Hossain Lipu, P.J. Ker, State of Charge Estimation
for Lithium-Ion Batteries Using Model-Based and Data-Driven Methods: A Review,
IEEE Access 7 (2019) 136116–136136, https://doi.org/10.1109/access.2019.
2942213.
7. Validation of the proposed work by ADVISOR tool
[3] Battery Performance Characteristics – How to specify and test a battery, (n.d.).
https://www.mpoweruk.com/performance.htm (accessed October24, 2019).
To reconfirm the accuracy of the proposed work for SoC estimation, [4] G. Zubi, R. Dufo-López, M. Carvalho, G. Pasaoglu, The lithium-ion battery: State of
the art and future perspectives, Renewable and Sustainable Energy Reviews 89
it has been implemented in the ADVISOR tool (for Toyota Prius) that is
(2018) 292–308, https://doi.org/10.1016/j.rser.2018.03.002.
developed by NREL and considered as a very significant tool for car- [5] Battery Comparison of Energy Density – Cylindrical and Prismatic Cells, (n.d.).
rying out the research on XEVs. The Fig. 18 demonstrates the difference https://www.epectec.com/batteries/cell-comparison.html (accessed October24,
between the estimated SoC (after implementation of the proposed 2019).
[6] L. Guo, C. Hu, G. Li, The SOC estimation of battery based on the method of im-
work) and the default SoC (without any modification). The proposed proved Ampere-hour and Kalman filter, Proceedings of the 2015 10th IEEE
results are in line with the default battery and are close to realistic Conference on Industrial Electronics and Applications, ICIEA 2015 (2015)
situation. 1458–1460, https://doi.org/10.1109/ICIEA.2015.7334337.
[7] C. Weng, J. Sun, H. Peng, A unified open-circuit-voltage model of lithium-ion
batteries for state-of-charge estimation and state-of-health monitoring, Journal of
Power Sources 258 (2014) 228–237, https://doi.org/10.1016/j.jpowsour.2014.02.
8. Conclusion 026.
[8] P. Jansen, M. Vollnhals, D. Renner, D. Vergossen, W. John, J. Götze, Advanced
This paper presents an accurate SoC estimation involving parameter binary search pattern for impedance spectra classification for determining the state
of charge of a lithium iron phosphate cell using a support vector machine, Advances
optimization for a hybrid electric vehicle. In this work whole battery
in Radio Science 14 (2016) 55–62, https://doi.org/10.5194/ars-14-55-2016.
system of 420V has been considered instead of a single cell. 3-RC pairs [9] D.I. Stroe, V. Knap, M. Swierczynski, T. Stanciu, E. Schaltz, R. Teodorescu, An
are included in the circuit in series with the internal resistance which Electrochemical Impedance Spectroscopy Study on a Lithium Sulfur Pouch Cell, ECS
has produced realistic results for EV applications. The values of the Transactions 72 (2016) 13–22, https://doi.org/10.1149/07212.0013ecst.
[10] Y. Xing, W. He, M. Pecht, K.L. Tsui, State of charge estimation of lithium-ion bat-
parameters have been calculated mathematically and then installed in teries using the open-circuit voltage at various ambient temperatures, Applied
the circuit to find the new SoC from the modified battery. The obtained Energy 113 (2014) 106–115, https://doi.org/10.1016/J.APENERGY.2013.07.008.
parameters are then optimized to get accurate results. To get a precise [11] C. Burgos, D. Sáez, M.E. Orchard, R. Cárdenas, Fuzzy modelling for the state-of-
charge estimation of lead-acid batteries, Journal of Power Sources 274 (2015)
estimation, this estimated SoC was compared with real data and error

16
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

355–366, https://doi.org/10.1016/J.JPOWSOUR.2014.10.036. Estimation of Lithium-Ion Batteries Based on a Dual Linear Kalman Filter -,
[12] Z. Zhu, D. Liu, Y. Rong, J. Sun, Estimation Based on a Second-order LiFePO 4 Technological Advances in Electrical, Electronics and Computer Engineering (2014)
Battery Model, (2013). https://www.semanticscholar.org/paper/Estimation-Based- 1–10.
on-a-Second-order-LiFePO-4-Battery-Zhu-Liu/ [38] R. Xiong, F. Sun, Z. Chen, H. He, A data-driven multi-scale extended Kalman fil-
bb7880e6c1d09d9882f2bb62c127e56cdfd2a69e (accessed May 29, 2019). tering based parameter and state estimation approach of lithium-ion olymer battery
[13] Z. He, C. Dong, C. Pan, C. Long, S. Wang, State of charge estimation of power Li-ion in electric vehicles, Applied Energy 113 (2014) 463–476, https://doi.org/10.1016/
batteries using a hybrid estimation algorithm based on UKF, Electrochimica Acta j.apenergy.2013.07.061.
211 (2016) 101–109, https://doi.org/10.1016/j.electacta.2016.06.042. [39] C. Campestrini, G. Walder, A. Jossen, M. Lienkamp, Temperature Influences on
[14] M. Paschero, G.L. Storti, A. Rizzi, F.M.F. Mascioli, G. Rizzoni, A Novel Mechanical State and Parameter Estimation Based on a Dual Kalman Filter, n.d. https://
Analogy-Based Battery Model for SoC Estimation Using a Multicell EKF, IEEE mediatum.ub.tum.de/doc/1226678/1226678.pdf (accessed June 5, 2019).
Transactions on Sustainable Energy 7 (2016) 1695–1702, https://doi.org/10.1109/ [40] C. Campestrini, T. Heil, S. Kosch, A. Jossen, A comparative study and review of
TSTE.2016.2574755. different Kalman filters by applying an enhanced validation method, Journal of
[15] A. Fotouhi, K. Propp, D.J. Auger, S. Longo, State of Charge and State of Health Energy Storage 8 (2016) 142–159, https://doi.org/10.1016/J.EST.2016.10.004.
Estimation Over the Battery Lifespan, in: Springer, Cham, 2018: pp. 267-288. 10. [41] S. Nejad, D.T. Gladwin, D.A. Stone, On-chip implementation of Extended Kalman
1007/978-3-319-69950-9_11. Filter for adaptive battery states monitoring, IECON Proceedings (Industrial
[16] S. Tong, J.H. Lacap, J.W. Park, Battery state of charge estimation using a load- Electronics Conference) (2016) 5513–5518, https://doi.org/10.1109/IECON.2016.
classifying neural network, Journal of Energy Storage 7 (2016) 236–243, https:// 7793527.
doi.org/10.1016/j.est.2016.07.002. [42] S. Li, S. Pischinger, C. He, L. Liang, M. Stapelbroek, A comparative study of model-
[17] S. Bobba, F. Mathieux, F. Ardente, G. Andrea, M. Anna, A. Podias, A. Pfrang, Life based capacity estimation algorithms in dual estimation frameworks for lithium-ion
Cycle Assessment of repurposed electric vehicle batteries : an adapted method batteries under an accelerated aging test, Applied Energy 212 (2018) 1522–1536,
based on modelling energy fl ows, Journal of Energy Storage 19 (2018) 213–225, https://doi.org/10.1016/j.apenergy.2018.01.008.
https://doi.org/10.1016/j.est.2018.07.008. [43] A. Malik, Z. Zhang, R.K. Agarwal, Extraction of battery parameters using a multi-
[18] N. Wassiliadis, J. Adermann, A. Frericks, M. Pak, C. Reiter, B. Lohmann, objective genetic algorithm with a non-linear circuit model, Journal of Power
M. Lienkamp, Revisiting the dual extended Kalman fi lter for battery state-of-charge Sources 259 (2014) 76–86, https://doi.org/10.1016/j.jpowsour.2014.02.062.
and state-of-health estimation : A use-case life cycle analysis, Journal of Energy [44] K.T. Chau, K.C. Wu, C.C. Chan, W.X. Shen, A new battery capacity indicator for
Storage 19 (2018) 73–87, https://doi.org/10.1016/j.est.2018.07.006. nickel-metal hydride battery powered electric vehicles using adaptive neuro-fuzzy
[19] L. Wang, D. Lu, Q. Liu, L. Liu, X. Zhao, Electrochimica Acta State of charge esti- inference system, Energy Conversion and Management 44 (2003) 2059–2071,
mation for LiFePO 4 battery via dual extended kalman fi lter and charging voltage https://doi.org/10.1016/S0196-8904(02)00249-2.
curve, Electrochimica Acta 296 (2019) 1009–1017, https://doi.org/10.1016/j. [45] C. Liu, W. Liu, L. Wang, G. Hu, L. Ma, B. Ren, A new method of modeling and state
electacta.2018.11.156. of charge estimation of the battery, Journal of Power Sources 320 (2016) 1–12,
[20] W. Zhang, L. Wang, L. Wang, C. Liao, An improved adaptive estimator for state-of- https://doi.org/10.1016/j.jpowsour.2016.03.112.
charge estimation of lithium- ion batteries, Journal of Power Sources 402 (2018) [46] E. Chemali, P.J. Kollmeyer, M. Preindl, A. Emadi, State-of-charge estimation of Li-
422–433, https://doi.org/10.1016/j.jpowsour.2018.09.016. ion batteries using deep neural networks: A machine learning approach, Journal of
[21] Y. Wang, C. Zhang, Z. Chen, A method for joint estimation of state-of-charge and Power Sources 400 (2018) 242–255, https://doi.org/10.1016/j.jpowsour.2018.06.
available energy of LiFePO 4 batteries, Applied Energy 135 (2014) 81–87, https:// 104.
doi.org/10.1016/j.apenergy.2014.08.081. [47] K.S.R. Mawonou, A. Eddahech, D. Dumur, D. Beauvois, E. Godoy, Improved state of
[22] Y. Li, P. Chattopadhyay, S. Xiong, A. Ray, C.D. Rahn, Dynamic data-driven and charge estimation for Li-ion batteries using fractional order extended Kalman filter,
model-based recursive analysis for estimation of battery state-of-charge q, Applied Journal of Power Sources 435 (2019) 226710, , https://doi.org/10.1016/j.
Energy 184 (2016) 266–275, https://doi.org/10.1016/j.apenergy.2016.10.025. jpowsour.2019.226710.
[23] R. Yang, R. Xiong, H. He, H. Mu, C. Wang, A novel method on estimating the de- [48] R. Xiong, H. He, Cell state-of-charge estimation for the multi-cell seriesconnected
gradation and state of charge of lithium-ion batteries used for electrical vehicles, battery pack with model bias correction approach, Energy Procedia 61 (2014)
Applied Energy 207 (2017) 336–345, https://doi.org/10.1016/j.apenergy.2017.05. 172–175, https://doi.org/10.1016/j.egypro.2014.11.932.
183. [49] A. Fotouhi, D.J. Auger, K. Propp, S. Longo, Lithium-Sulfur Battery State-of-Charge
[24] D. Yang, Y. Wang, R. Pan, R. Chen, Z. Chen, State-of-health estimation for the li- Observability Analysis and Estimation, IEEE Transactions on Power Electronics 33
thium-ion battery based on support vector regression, Applied Energy 227 (2018) (2018) 5847–5859, https://doi.org/10.1109/TPEL.2017.2740223.
273–283, https://doi.org/10.1016/j.apenergy.2017.08.096. [50] L. Guo, J. Li, Z. Fu, Lithium-ion battery SOC estimation and hardware-in-the-loop
[25] X. Tang, F. Gao, C. Zou, K. Yao, W. Hu, T. Wik, Load-responsive model switching simulation based on EKF, Energy Procedia 158 (2019) 2599–2604, https://doi.org/
estimation for state of charge of lithium- ion batteries, Applied Energy 238 (2019) 10.1016/j.egypro.2019.02.009.
423–434, https://doi.org/10.1016/j.apenergy.2019.01.057. [51] B. Sun, J. Jiang, Z. Wang, The Comparative Study of SOC Estimation Based on EKF
[26] G. Liu, C. Xu, H. Li, K. Jiang, K. Wang, State of charge and online model parameters and ANFIS Algorithm, International Conference on Information and Network
co-estimation for liquid metal batteries, Applied Energy 250 (2019) 677–684, Technology 4 (2011) 207–210.
https://doi.org/10.1016/j.apenergy.2019.05.032. [52] I. Baccouche, S. Jemmali, A. Mlayah, B. Manai, N.E. Ben Amara, Implementation of
[27] J. Meng, M. Ricco, A. Budnar, G. Luo, M. Swierczynski, D. Stroe, R. Teodorescu, an improved Coulomb-counting algorithm based on a piecewise SOC-OCV re-
Low-complexity online estimation for LiFePO 4 battery state of charge in electric lationship for SOC estimation of Li-ion battery, International Journal of Renewable
vehicles, Journal of Power Sources 395 (2018) 280–288, https://doi.org/10.1016/j. Energy Research 8 (2018) 178–187.
jpowsour.2018.05.082. [53] D.W. Chung, S.H. Yang, SOC estimation of lithium-ion battery based on kalman
[28] H. Ben Sassi, F. Errahimi, N. Es-Sbai, C. Alaoui, Comparative study of ANN/KF for filter algorithm for energy storage system in microgrids, E3S Web of Conferences 57
on-board SOC estimation for vehicular applications, Journal of Energy Storage 25 (2018) 2316–2319, https://doi.org/10.1051/e3sconf/20185702006.
(2019) 100822, , https://doi.org/10.1016/j.est.2019.100822. [54] C.H. Cai, D. Du, Z.Y. Liu, Battery state-of-charge (SOC) estimation using adaptive
[29] M.A. Awadallah, B. Venkatesh, Accuracy improvement of SOC estimation in li- neuro-fuzzy inference system (ANFIS), IEEE International Conference on Fuzzy
thium-ion batteries, Journal of Energy Storage 6 (2016) 95–104, https://doi.org/ Systems 2 (2003) 1068–1073, https://doi.org/10.1109/fuzz.2003.1206580.
10.1016/j.est.2016.03.003. [55] R. Xiong, H. He, F. Sun, K. Zhao, Evaluation on State of Charge estimation of bat-
[30] D. Carnevale, C. Possieri, M. Sassano, Control Engineering Practice State-of-charge teries with adaptive extended kalman filter by experiment approach, IEEE
estimation for lead – acid batteries via embeddings and observers, Control Transactions on Vehicular Technology 62 (2013) 108–117, https://doi.org/10.
Engineering Practice 85 (2019) 132–137, https://doi.org/10.1016/j.conengprac. 1109/TVT.2012.2222684.
2019.01.011. [56] A.A. Hussein, Capacity Fade Estimation in Electric Vehicle Li-Ion Batteries Using
[31] H. Zhigang, C. Dong, P. Chaofeng, C. Long, W. Shaohua, State of charge estimation Artificial Neural Networks, IEEE Transactions on Industry Applications 51 (2015)
of power Li-ion batteries using a hybrid estimation algorithm based on UKF, 2321–2330, https://doi.org/10.1109/TIA.2014.2365152.
Electrochimica Acta 211 (2016) 101–109, https://doi.org/10.1016/J.ELECTACTA. [57] C. Zhang, L.Y. Wang, X. Li, W. Chen, G.G. Yin, J. Jiang, Robust and Adaptive
2016.06.042. Estimation of State of Charge for Lithium-Ion Batteries, IEEE Transactions on
[32] M. Urbain, S. Rael, B. Davat, P. Desprez, State Estimation of a Lithium-Ion Battery Industrial Electronics 62 (2015) 4948–4957, https://doi.org/10.1109/TIE.2015.
Through Kalman Filter, in: 2007 IEEE Power Electronics Specialists Conference, 2403796.
IEEE, 2007: pp. 2804-2810. 10.1109/PESC.2007.4342463. [58] Y. Ma, X. Zhou, B. Li, H. Chen, Fractional modeling and SOC estimation of lithium-
[33] S. Lee, J. Kim, J. Lee, B.H. Cho, State-of-charge and capacity estimation of lithium- ion battery, IEEE/CAA Journal of Automatica Sinica 3 (2016) 281–287, https://doi.
ion battery using a new open-circuit voltage versus state-of-charge, Journal of org/10.1109/JAS.2016.7508803.
Power Sources 185 (2008) 1367–1373, https://doi.org/10.1016/J.JPOWSOUR. [59] Q. Yu, R. Xiong, C. Lin, W. Shen, J. Deng, Lithium-Ion Battery Parameters and State-
2008.08.103. of-Charge Joint Estimation Based on H-Infinity and Unscented Kalman Filters, IEEE
[34] M. Rubagotti, S. Onori, G. Rizzoni, Automotive Battery Prognostics Using Dual Transactions on Vehicular Technology 66 (2017) 8693–8701, https://doi.org/10.
Extended Kalman Filter, (2010) 257-263. 10.1115/dscc2009-2725. 1109/TVT.2017.2709326.
[35] J. Kim, B.H. Cho, Pattern recognition for temperature-dependent state-of-charge/ [60] J. Yang, B. Xia, Y. Shang, W. Huang, C.C. Mi, Adaptive state-of-charge estimation
capacity estimation of a li-ion cell, IEEE Transactions on Energy Conversion 28 based on a split battery model for electric vehicle applications, IEEE Transactions
(2013) 1–11, https://doi.org/10.1109/TEC.2012.2222884. on Vehicular Technology 66 (2017) 10889–10898, https://doi.org/10.1109/TVT.
[36] T. Dragicevic, S. Sucic, J.M. Guerrero, Battery state-of-charge and parameter esti- 2017.2728806.
mation algorithm based on Kalman filter, in: Eurocon 2013, IEEE, 2013: pp. 1519- [61] Q. Ouyang, J. Chen, J. Zheng, Y. Hong, SOC Estimation-Based Quasi-Sliding Mode
1525. 10.1109/EUROCON.2013.6625179. Control for Cell Balancing in Lithium-Ion Battery Packs, IEEE Transactions on
[37] G. Walder, C. Campestrini, M. Lienkamp, A. Jossen, Adaptive State and Parameter Industrial Electronics 65 (2018) 3427–3436, https://doi.org/10.1109/TIE.2017.

17
K.V. Singh, et al. Journal of Energy Storage 27 (2020) 101124

2750629. Industrial Electronics 61 (2014) 3989–3999, https://doi.org/10.1109/TIE.2013.

[62] P. Shen, M. Ouyang, L. Lu, J. Li, X. Feng, The co-estimation of state of charge, state 2288195.
of health, and state of function for lithium-ion batteries in electric vehicles, IEEE [80] K. Propp, D.J. Auger, A. Fotouhi, M. Marinescu, V. Knap, S. Longo, Improved state
Transactions on Vehicular Technology 67 (2018) 92–103, https://doi.org/10.1109/ of charge estimation for lithium-sulfur batteries, Journal of Energy Storage 26
TVT.2017.2751613. (2019) 100943, , https://doi.org/10.1016/j.est.2019.100943.
[63] J. Meng, M. Ricco, G. Luo, M. Swierczynski, D.I. Stroe, A.I. Stroe, R. Teodorescu, An [81] V. Knap, D.J. Auger, K. Propp, A. Fotouhi, D.I. Stroe, Concurrent real-time esti-
Overview and Comparison of Online Implementable SOC Estimation Methods for mation of state of health and maximum available power in lithium-sulfur batteries,
Lithium-Ion Battery, IEEE Transactions on Industry Applications 54 (2018) Energies (2018) 11, https://doi.org/10.3390/en11082133.
1583–1591, https://doi.org/10.1109/TIA.2017.2775179. [82] A. Fotouhi, D.J. Auger, K. Propp, S. Longo, Accuracy Versus Simplicity in Online
[64] G.S. Misyris, D.I. Doukas, T.A. Papadopoulos, D.P. Labridis, V.G. Agelidis, State-of- Battery Model Identification, IEEE Transactions on Systems, Man, and Cybernetics:
Charge Estimation for Li-Ion Batteries: A More Accurate Hybrid Approach, IEEE Systems 48 (2018) 195–206, https://doi.org/10.1109/TSMC.2016.2599281.
Transactions on Energy Conversion 34 (2019) 109–119, https://doi.org/10.1109/ [83] A. Fotouhi, K. Propp, L. Samaranayake, D. Auger, S. Longo, A hardware-in-the-loop
TEC.2018.2861994. test rig for development of electric vehicle battery identification and state estima-
[65] J.N. Shen, J.J. Shen, Y.J. He, Z.F. Ma, Accurate State of Charge Estimation with tion algorithms, International Journal of Powertrains 7 (2018) 227–248, https://
Model Mismatch for Li-Ion Batteries: A Joint Moving Horizon Estimation Approach, doi.org/10.1504/IJPT.2018.090391.
IEEE Transactions on Power Electronics 34 (2019) 4329–4342, https://doi.org/10. [84] V. Knap, D.I. Stroe, R. Purkayastha, S. Walus, D.J. Auger, A. Fotouhi, K. Propp,
1109/TPEL.2018.2861730. Reference performance test methodology for degradation assessment of lithium-
[66] J. Meng, D.I. Stroe, M. Ricco, G. Luo, R. Teodorescu, A simplified model-based state- sulfur batteries, Journal of the Electrochemical Society 165 (2018) A1601–A1609,
of-charge estimation approach for lithium-ion battery with dynamic linear model, https://doi.org/10.1149/2.0121809jes.
IEEE Transactions on Industrial Electronics 66 (2019) 7717–7727, https://doi.org/ [85] A. Fotouhi, D.J. Auger, K. Propp, S. Longo, Electric vehicle battery parameter
10.1109/TIE.2018.2880668. identification and SOC observability analysis: NiMH and Li-S case studies, IET
[67] M.S. Hossain Lipu, M.A. Hannan, A. Hussain, M.H. Saad, A. Ayob, M.N. Uddin, Power Electronics 10 (2017) 1289–1297, https://doi.org/10.1049/iet-pel.2016.
Extreme learning machine model for state-of-charge estimation of lithium-ion 0777.
battery using gravitational search algorithm, IEEE Transactions on Industry [86] K. Propp, D.J. Auger, A. Fotouhi, S. Longo, V. Knap, Kalman-variant estimators for
Applications 55 (2019) 4225–4234, https://doi.org/10.1109/TIA.2019.2902532. state of charge in lithium-sulfur batteries, Journal of Power Sources 343 (2017)
[68] F. Yang, X. Song, F. Xu, K.L. Tsui, State-of-Charge Estimation of Lithium-Ion 254–267, https://doi.org/10.1016/j.jpowsour.2016.12.087.
Batteries via Long Short-Term Memory Network, IEEE Access 7 (2019) [87] V. Knap, D.I. Stroe, R. Purkayastha, S. Walus, D.J. Auger, A. Fotouhi, K. Propp,
53792–53799, https://doi.org/10.1109/ACCESS.2019.2912803. Methodology for assessing the lithium-sulfur battery degradation for practical ap-
[69] Z. Xi, M. Dahmardeh, B. Xia, Y. Fu, C. Mi, Learning of Battery Model Bias for plications, ECS Transactions 77 (2017) 479–490, https://doi.org/10.1149/07711.
Effective State of Charge Estimation of Lithium-Ion Batteries, IEEE Transactions on 0479ecst.
Vehicular Technology 68 (2019) 8613–8628, https://doi.org/10.1109/tvt.2019. [88] A. Fotouhi, D.J. Auger, K. Propp, S. Longo, M. Wild, A review on electric vehicle
2929197. battery modelling: From Lithium-ion toward Lithium-Sulphur, Renewable and
[70] M.U. Cuma, T. Koroglu, A comprehensive review on estimation strategies used in Sustainable Energy Reviews 56 (2016) 1008–1021, https://doi.org/10.1016/j.rser.
hybrid and battery electric vehicles, Renewable and Sustainable Energy Reviews 42 2015.12.009.
(2015) 517–531, https://doi.org/10.1016/j.rser.2014.10.047. [89] R. Jackey, M. Saginaw, P. Sanghvi, J.G. Mathworks, Battery Model Parameter
[71] H. Dai, P. Guo, X. Wei, Z. Sun, J. Wang, ANFIS (adaptive neuro-fuzzy inference Estimation Using a Layered Technique: An Example Using a Lithium Iron Phosphate
system) based online SOC (State of Charge) correction considering cell divergence Cell, 2013. https://in.mathworks.com/content/dam/mathworks/tag-team/
for the EV (electric vehicle) traction batteries, Energy (2014) 1–11, https://doi.org/ Objects/s/76117-sae-2013-battery-estimation-layered-technique.pdf (accessed
10.1016/j.energy.2014.11.077. June 5, 2019).
[72] P. Venugopal, T. Vigneswaran, State-of-Charge Estimation Methods for Li-ion [90] A. Fotouhi, D.J. Auger, K. Propp, S. Longo, R. Purkayastha, L. O'Neill, S. Walus,
Batteries in Electric Vehicles, (2019) 37-46. Lithium-Sulfur Cell Equivalent Circuit Network Model Parameterization and
[73] M.A. Hannan, M.S.H. Lipu, A. Hussain, A. Mohamed, A review of lithium-ion bat- Sensitivity Analysis, IEEE Transactions on Vehicular Technology 66 (2017)
tery state of charge estimation and management system in electric vehicle appli- 7711–7721, https://doi.org/10.1109/TVT.2017.2678278.
cations : Challenges and recommendations, Renewable and Sustainable Energy [91] L. Limn2o4, L.-P. Linicoo2, L. Li, Lead Acid NiCad NiMH Zinc Air Aluminium Air
Reviews 78 (2017) 834–854, https://doi.org/10.1016/j.rser.2017.05.001. Nickel Iron Nickel Zinc Sodium Sulphur Zebra NaNiCl, (1881). http://www.
[74] M.A. Roscher, S. Member, O.S. Bohlen, D.U. Sauer, Reliable State Estimation of mpoweruk.com/specifications/comparisons.pdf.
Multicell Lithium-Ion Battery Systems, IEEE Transactions on Energy Conversion 26 [92] Lithium-based Batteries Information – Battery University, (n.d.). https://
(2011) 737–743, https://doi.org/10.1109/TEC.2011.2155657. batteryuniversity.com/learn/article/lithium_based_batteries (accessed Feburary 13,
[75] D. Huang, Z. Chen, C. Zheng, H. Li, A model-based state-of-charge estimation 2019).
method for series- connected lithium-ion battery pack considering fast-varying cell [93] S. Jiang, A Parameter Identification Method for a Battery Equivalent Circuit Model,
temperature, Energy 185 (2019) 847–861, https://doi.org/10.1016/j.energy.2019. SAE Technical Paper Series 1 (2011), https://doi.org/10.4271/2011-01-1367.
07.063. [94] A. Panday, H.O. Bansal, P. Srinivasan, Thermoelectric Modeling and Online SOC
[76] W. Waag, C. Fleischer, D.U. Sauer, Critical review of the methods for monitoring of Estimation of Li-Ion Battery for Plug-In Hybrid Electric Vehicles, 2016 (2016).
lithium-ion batteries in electric and hybrid vehicles, Journal of Power Sources 258 [95] H. Rahimi-Eichi, F. Baronti, M.Y. Chow, Online adaptive parameter identification
(2014) 321–339, https://doi.org/10.1016/j.jpowsour.2014.02.064. and state-of-charge coestimation for lithium-polymer battery cells, IEEE
[77] D. Andre, M. Meiler, K. Steiner, H. Walz, T. Soczka-Guth, D.U. Sauer, Transactions on Industrial Electronics 61 (2014) 2053–2061, https://doi.org/10.
Characterization of high-power lithium-ion batteries by electrochemical impedance 1109/TIE.2013.2263774.
spectroscopy. II: Modelling, Journal of Power Sources 196 (2011) 5349–5356, [96] K.S. Ng, C.S. Moo, Y.P. Chen, Y.C. Hsieh, State-of-charge estimation for lead-acid
https://doi.org/10.1016/j.jpowsour.2010.07.071. batteries based on dynamic open-circuit voltage, PECon 2008 – 2008 IEEE 2nd
[78] C. Fleischer, W. Waag, H.M. Heyn, D.U. Sauer, On-line adaptive battery impedance International Power and Energy Conference (2008) 972–976, https://doi.org/10.
parameter and state estimation considering physical principles in reduced order 1109/PECON.2008.4762614.
equivalent circuit battery models part 2. Parameter and state estimation, Journal of [97] W.-Y. Chang, The State of Charge Estimating Methods for Battery: A Review, ISRN
Power Sources 262 (2014) 457–482, https://doi.org/10.1016/j.jpowsour.2014.03. Applied Mathematics 2013 (2013) 1–7, https://doi.org/10.1155/2013/953792.
046. [98] R. Kumar, H.O. Bansal, Real-time implementation of adaptive PV-integrated SAPF
[79] M.Y. Kim, C.H. Kim, J.H. Kim, G.W. Moon, A chain structure of switched capacitor to enhance power quality, International Transactions on Electrical Energy Systems
for improved cell balancing speed of lithium-ion batteries, IEEE Transactions on 29 (2019) 1–22, https://doi.org/10.1002/2050-7038.12004.

18