Modeling, State of Charge Estimation, and Charging of Lithium-Ion Battery in Electric Vehicle: A Review

1099114x, 2022, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/er.7339 by Nanyang Technological Uni, Wiley Online Library on [16/02/2023].
See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Received: 10 April 2021 Revised: 16 September 2021 Accepted: 17 September 2021
DOI: 10.1002/er.7339
REVIEW PAPER
Modeling, state of charge estimation, and charging of

lithium-ion battery in electric vehicle: A review
Maheshwari Adaikkappan | Nageswari Sathiyamoorthy
Department of Electrical and Electronics

Engineering, Alagappa Chettiar
Summary
Government College of Engineering and Extension of driving range and battery run time optimization are necessary
Technology, Karaikudi, India key points in the modeling of Electric Vehicle (EV). In this view, Battery Man-
Correspondence agement System (BMS) plays a major role to ensure a safe and trustworthy bat-
Maheshwari Adaikkappan, Department of tery operation, especially when using Lithium-ion (Li-ion) batteries in an
Electrical and Electronics Engineering,
electric vehicle. Key function of BMS is State of Charge (SoC) estimation. A
Alagappa Chettiar Government College of
Engineering and Technology, Karaikudi, well-parameterized battery model is required for accurate state estimation.
India. Consequently, the major factors to be considered in battery modeling are the
Email: [email protected]
SoC estimation and charging methodology of an effective BMS development.
Funding information By focusing on these features, in this paper, the well-known battery models
TEQIP such as the electrochemical model, equivalent circuit model, and data-driven
model are comprehensively reviewed along with their strengths and weak-
nesses. Further, the SoC estimation of a battery is also discussed by using stan-
dard methodologies such as direct estimation methods and model-based
estimation methods. The comparisons of the three most distinct battery models
and the classification of SoC estimation techniques to develop a proper BMS
for EV with the focus on accuracy, configuration effort, computational com-
plexity, ease of implementation, and real-time applications are systematically
reviewed. In addition to this, convenient battery charging approaches with the
consideration of some constraints such as charging time, charging efficiency,
state of charge, state of health, charging voltage threshold, capacity fade,
power fade, aging effect, capacity utilization, impedance rise, and temperature
rise of the battery in EV are presented. Finally, the perspectives of the existing
work and the recommended future research work of BMS are summarized.
KEYWORDS
battery management system, battery modeling, fault diagnosis, parameter identification,
state estimation
1 | INTRODUCTION battery and recharged by the plugging vehicle. Among

different types of batteries in the market, Li-ion batteries
All-electric vehicles produce zero direct emissions and as are the best suited for EV applications, due to their
a result, they contribute in improving the air quality of advantages, such as high energy density, low mainte-
the atmosphere. One stop solution for this is the Battery nance, and long service life.1 Despite the advantages, the
Electric Vehicle (BEV) which is completely powered by Li-ion battery is very sensitive to overcharge and deep
Int J Energy Res. 2022;46:2141–2165. wileyonlinelibrary.com/journal/er © 2021 John Wiley & Sons Ltd. 2141
1099114x, 2022, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/er.7339 by Nanyang Technological Uni, Wiley Online Library on [16/02/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2142 ADAIKKAPPAN AND SATHIYAMOORTHY
discharge, which can reduce its lifespan and even cause compares the stored data with the safe operating limits of
fire or explosion. Hence, the battery must be operated in the battery. If any abnormal condition occurs in the bat-
a Safe Operating Area (SOA).2 Specific care should be tery operating process, a safety module will protect the
taken to design a suitable BMS that detects faults and battery from damage. An effective communication net-
protects the battery from damage. BMS has some operat- work provides a user interface between the battery and
ing modules which are mainly divided into two groups, the BMS.
that is, hardware and software.3 The overview of the Fault detection in a high-capacity system is a very dif-
components of BMS is shown in Figure 1. The BMS soft- ficult fact to design.4 Major EV crash occurs due to bat-
ware is the heart of the system, since it controls all hard- tery cell inconsistency. When the current, temperature,
ware operations and analyses sensor data to make charge, and discharge exceed the limit value, they cause
decisions. The majority of faults will be detected through defects in the battery cells. Defects could cause fire and
online data processing. In order to deliver battery mal- battery damage due to the cell explosion. A proper charg-
function warnings, intelligent data analysis is required. ing strategy and a cell balancing circuit are necessary to
Data collection is more important to reveal the pre-alarm prevent cell inconsistency.
before the fault occurrence. Most cell balancing circuits are developed solely
Hardware components perform their work according based on voltage5 and they are considered to be very sim-
to the software instructions. The measurement of battery ple but end in poor performance. Based on an average
current and the voltage can be done with the help of sen- SoC, active cell balancing is established for better perfor-
sors. A good battery model is essential to monitor and mance.6 Although the cell balancing strategy based on
analyze the battery behavior, identify the faults, and pre- SoC and voltage is essential for fault diagnosis, single cell
vent a battery from thermal runaway. Furthermore, State SoC estimation is important before applying a cell
of Charge (SoC) plays a crucial role in regulating battery balancing method. As a result, the SoC is used as an
operation and it must consequently be monitored using input for power calculation, cell balancing, and fault
correct estimation methods. Therefore, a well-trained bat- diagnosis.
tery model together with a proper SoC estimation tech- In this paper, three key issues such as battery model-
nique is used to estimate the SoC of a battery precisely. ing, SoC estimation, and charging strategy of Li-ion bat-
After acquiring the behaviors of a battery, a suitable opti- tery cell are focused and the flow of the work is
mization algorithm has been applied to optimize the represented in Figure 2 as well as the relation of these
charging pattern of the battery. Battery charging is criti- three issues is illustrated in Figure 3. Among several bat-
cal in BMS, since it has a direct impact on battery protec- tery models, three well-known modeling techniques are
tion and accessibility. A well-designed charging approach
prevents the battery from damage, limits the temperature
changes, and increases the energy conversion efficiency.
Then, all the parameters of a battery can be stored using
Electrochemical Model
Battery
a data acquisition system. Finally, the system controller

models
Equivalent Circuit Model

Data driven Model
Battery Management System

estimation
Direct (Ampere hour, OCV method)

SOFTWARE HARDWARE
SoC
Model based (Adaptive filter methods,

computer intelligence algorithms)
Battery modeling Sensors and
Monitoring Circuit
State Estimation Charge and

Discharge Circuit Constant Trickle Charging
Charging algorithms
Constant Current Charging

Constant Voltage Charging
Fault Detection Safety Circuit Constant Current Constant Voltage
Charging
Data Acquisition Multi step Constant Current Charging
Data Storage Boost Charging
Pulse Charging
F I G U R E 1 Overview of the software and hardware

components of BMS FIGURE 2 Framework of key technologies in BMS
ADAIKKAPPAN AND SATHIYAMOORTHY 2143
Direct estimation
Li-ion battery cell
single dataset Coulomb
Counting/
Charging method Electrochemical Equivalent circuit Parameter
single dataset Open Circuit
model model identification
Voltage
Least method
Square
(or)
(or)
Genetic Model based estimation
Algorithm
Filter/
Observer/ State of Charge
Data driven model Learning
algorithm
multiple dataset
(or)
Artificial Deeplearning
neural network neural network
FIGURE 3 Relation between battery modelling, SoC estimation and charging method of a battery
discussed in Section 2. A battery model is essential for of BMS design is to build a proper battery model. Till
fault diagnosis, state estimation, cell balance, and ther- date, numerous battery models have been developed with
mal management. The standard models and the parame- various accuracy levels. The most distinct battery models
ters associated with a model are examined and described are the electrochemical model, equivalent circuit model,
to present the battery behavior. Each state is essential for and data-driven model.
driving range estimation, cell balancing, power demand,
energy management, thermal management, safety man-
agement, and the remaining useful life prediction of a 2.1 | Electrochemical model
battery for EV. In most of the studies carried out for SoC
estimation of a battery cell, SoC is the most important Electrochemical models are used to describe battery
parameter to find and control other states, since it is behaviors based on the electrolyte concentration, size of
mutually correlated with other states. Moreover, State of the electrodes (anode and cathode), and electrochemical
Health (SoH) is slowly changing in terms of capacity fad- process inside the battery using partial differential equa-
ing, temperature, internal resistance, and SoC. The State tions. Eventhough Electrochemical model (EM) provides
of Temperature (SoT) is varied on an intermediate scale accurate battery parameters, it needs more computation
due to the charging strategy, heat transfer, and electro- power and time to find numerous parameters such as
chemical properties of the battery. Although State of electrolyte potential, solid potential, open circuit poten-
Energy (SoE) and State of Power (SoP) are varied quickly, tial, overpotential, electrolyte concentration, solid con-
they highly depend on SoC. As a result, some standard centration, battery cell current, temperature and so
and popular SoC estimation algorithms are reviewed in on. Moreover, it is difficult to implement in real-time
terms of accuracy, complexity, and technical challenges applications. Based on its physical process, Doyle et al,7
in Section 3. Together with the state assessment, the have suggested a P2D electrochemical model. The
proper charging strategy can be well developed to avoid extended simulation time of the P2D model due to huge
failures. In Section 4, the popular battery charging number of nonlinear equations makes it computationally
methods are reviewed and the recommendations for inefficient for BMS applications.
future work are discussed in Section 5. Section 6 details Domenico et al,8 have proposed a reduced order EM
the review of three key issues of a battery cell for design- by taking an average solid electrolyte concentration
ing a BMS for EV applications. instead of the distribution along the electrodes.
Eventhough the model losses some information, it can be
implemented in real-time onboard buses. However, the
2 | BATTERY M ODELS identification of parameters is a difficult task. Ahmed
et al,9,10 have identified the battery parameters using a
In EV applications, accurate SoC estimation reduces the Genetic Algorithm and estimated SoC. For identification,
range anxiety of EV customers. A good battery model is the model is run 10 times and a population size of 1000 is
essential for battery state estimation. The starting point used as well as identification of 18 parameters is done in
6 hours. Even this model has the advantage of reducing Although this is simple to implement, it cannot accurately
computational power compared to the full order model; present the characteristics of the battery used in EV. A sin-
accuracy is reduced due to some assumptions that it is gle Resistance-Capacitance parallel network is added to
made to reduce the order of the model. Han et al,11 have Rint model to form a Thevenin model.18 The Thevenin
proposed an approximate model that monitors the diffu- model is widely used to capture the dynamic characteristics
sion process and the distribution of electrolyte concentra- of a battery. The modified version of Thevenin model is a
tion inside the battery. Zou et al,12 have proposed a Partnership for a New Generation of Vehicle (PNGV)
reduced-order model based on a singular perturbation model19 which includes a fictive capacitor to reflect the
approach and an averaging theory to estimate the SoC of effects of OCV variation.20 PNGV model also known as
Li-ion as well as to predict the discharging capacity of the FreedomCar model21 comprises open circuit voltage, ohmic
battery. This methodology of model simplification is suit- resistance, fictive capacitor, polarization resistance, and
able for all batteries. However, developing a high-fidelity capacitor. PNGV model is an ideal choice for low SoC area
model concerning temperature, aging, and capacity fad- whereas it is not suitable for high SoC region.22
ing not only improves the accuracy but also increases the Among the various ECM models available in the liter-
complexity. Comparison of various types of Electrochem- ature, Resistance-Capacitance (RC) network-based
ical models of battery is summarized in Table 1. models such as one RC network ECM,23,24 two RC net-
work ECM,25-27 and three RC network ECM28,29 are
widely accepted models for online applications.30,31
2.2 | Equivalent circuit model According to circuit theory, model equations, and param-
eters are specified in Table 2.
Equivalent circuit model (ECM) uses electrical components Among these three, two RC network model has high
such as voltage source, resistors, and capacitors that accuracy to predict the relation between the input cur-
describe the electrical behavior of the battery. In ECM, a rent and the output voltage (I-V) of the battery as well as
high valued capacitor13 or controlled voltage source14 is the charging and discharging time of the battery.14 Since
used to denote the Open Circuit Voltage (OCV) of battery the battery is a nonlinear system, the battery dynamics
and it is the most important parameter for many state esti- vary for different operating conditions such as SoC, tem-
mation techniques. Hu et al,15 have analyzed “a compara- perature, charging, and discharging rate of a battery.
tive study of twelve equivalent circuit models for Li-ion Therefore, the model parameterization has been consid-
batteries.” Among the 12 ECM models, Plett16 has projec- ered as an “identification problem” or “optimization
ted six battery models such as a combined model, a simple problem” to fit the model to the measured data. Hence,
model, zero-state hysteresis model, one state hysteresis the battery model is fixed, and the model parameters are
model, enhanced self-correcting model (with two state Low to be identified. The well-known Least Square
Pass Filter, four state Low Pass Filter). (LS) technique,25,26,32,33 and the Genetic Algorithm
The most simple ECM is the Rint model17 which con- (GA)23,31,34 are applied to identify the model parameters
siders the battery as a voltage source with series resistance. with experimental data. Sitterly et al,13,35,36 have
TABLE 1 Comparison of various types of Electrochemical models of battery
S.
No Model Merits Demerits
1 Pseudo-2-D (P2D) electrochemical model High accuracy High computational complexity
or Doyle-Fuller- Newman (DFN)
electrochemical battery model.
2 Reduced order model or Single particle Simple model that reduces the electrode Radial-domain PDEs in the SP model
(SP) model to a single particle and neglect the still need to be solved in order to
liquid phase. obtain the solid-phase Li-ion
concentration.
Lack accuracy at high C rates.
3 Extended Single Particle (ESP) Model. It simplifies the electrode into a single Model complexity is increased
active particle. Curve fitting or
approximate solution process is used to
solve PDEs. Improves the
computational efficiency.
TABLE 2 RC network based Equivalent Circuit models
Circuit diagram Model equations Model parameters

One RC network: V Bat ¼ OCVðSoCÞ I Bat R0 V 1 OCV,R0 , R1 ,C 1

t
V1 ¼ C 1 þ I Bat R1
Q
exp R1 C 1 I Bat
R1
Two RC network: V Bat ¼ OCVðSoCÞ I Bat R0 V 1 V 2 OCV,R0 , R1 ,C 1 ,R2 ,C2

V 1 ¼ CQ1 þ I Bat R1 exp R1t
C 1 I Bat R1

C 2 I Bat R2
Three RC network: V Bat ¼ OCVðSoCÞ I Bat R0 V 1 V 2 V 3 OCV ,R0 , R1 , C1 , R2 ,

V 1 ¼ CQ1 þ I Bat R1 exp R1t C2 , R3 ,C 3
C 1 I Bat R1

C 2 I Bat R2

C 3 I Bat R3
established that the model must be identified. Without has the ability of computational intelligence to approxi-
laboratory experimentation, the model is identified with mate highly nonlinear battery characteristics. Various
real-time operating data from EV for BMS development. data-driven models including Artificial Neural Network
An automated real-time battery characterizer has been (ANN),37 Adaptive Neuro-Fuzzy Inference Systems
developed to capture and update cell characteristics in (ANFIS),38 Deep Neural Network (DNN),39 Support Vec-
real-time. For evaluation, the RC model that is developed tor Machine (SVM)40-42 describe the behavior of a battery
in the National Renewable Energy Laboratory (NREL) without prior knowledge of the battery's internal
has been taken. Real-time data are obtained from the structure.
ADVISOR tool. In addition to that, the identification of ANN is a powerful tool to develop a model for any
the system, the convergence rate of the estimation algo- nonlinear function when the training of data is done
rithm, the bias in the identification, and the correction of properly.43 The accuracy of this model demands large
the bias are thoroughly verified. quantities of high-quality data with different load condi-
However, ECM circuit elements do not describe the tions and also abundant training time. Feed Forward
proper physical representation of batteries. Besides, the Neural Network (FFNN) is the least complex among
model parameters are updated at all times with the SoC, other data driven models because of its simple learning
temperature, and charge-discharge rate of a battery. process and non-intensive computation. Like FFNN,
DNN has high computational efficiency compared to
other learning algorithms. But, FFNN and DNN are one-
2.3 | Data-driven model way where the history of data is not considered.
Both are well trained, that is, network biases and
Data-driven model (DDM) is significantly more efficient weights are learned through offline and there is no need
than ECM as well as EM but it highly depends on the for a backward pass during online SoC estimation. There-
amount of data along with the training method. DDM fore, the main challenge is that the models have to be
trained well in offline. In Reference 44, the authors have preprocessing and noise removal are also important in all
pointed out that the time required to form a DNN is from the abovementioned methods. Extreme calculations
a few hours to 40 to 50 hours at different ambient tem- could be another major obstacle for the implementation
peratures; although the training is performed on the of economic applications.
Graphics Processing Unit (GPU), and it has the advan- Chen et al,57 have introduced the Grey Model
tage of parallel computing. In Reference 37, the (GM) approach for describing a system as a “grey box”
Nonlinear Auto Regressive model with eXogenous input and it can predict unknown parameters using little-
(NARX) is used to obtain the required battery voltage. known battery information. Zhang et al,58 have stated
Unlike ANN, the Recurrent Neural Network (RNN) has that the Metabolic Grey Model (MGM), a new methodol-
the advantage of using the history of data with recurrent ogy is used to track the dynamic data calculations and
units.45 The only problem is memory loss; that means it can- the SoC estimation accuracy of a battery is enhanced
not capture the information for a long-time because of the through the metabolic mechanism. Each model has its
gradient vanishing phenomenon. Hence, Gated Recurrent strength and some drawbacks. Table 3 summarizes the
Unit (GRU)-RNN46-49 and Long Short Term Memory advantages and disadvantages of various Data-Driven
(LSTM)-RNN50-53 are established to capture long-term models.
requirements. Since GRU-RNN is structured over time; Data collection is critical in the development of a data-
another method of backpropagation has to be applied, driven model because these models require a large amount
known as Back-Through-Time-Propagation (BPTT) to build of data for training. With the increased deployment of bat-
a network. Proper data normalization is essential to improve teries in a variety of applications, batteries experience vary-
the GRU-RNN training process and it increases the conver- ing degradation rates under various operating conditions.
gence rate. In LSTM, the memory block is added to a hidden As a result, application-specific data are required to
unit to store the information for a longer time. develop an accurate battery model.70 Since data collection
ANFIS combines the advantages such as flexibility, is a time-consuming and costly process, researchers can
subjectivity of fuzzy systems and the learning capability use publicly available data instead of conducting experi-
of neural networks.39 The Takagi–Sugeno (T-S) fuzzy ments on batteries under various operating conditions.
model54 has an inherent multiple-model structure that Table 4 summarizes different combinations of battery
can cope with the nonlinear dynamics of battery. But, it models with parameter identification algorithms.
is more rule-based and the accuracy of the model varies Figure 4, shows the comparison of various battery models
with the number of rules, whereas the black-box models with indexing values in terms of accuracy, configuration
can provide accurate results and they require more com- effort, computational complexity, ease of implementa-
putational complexity as well as it is hard to analyze their tion, real-time applications, and interpretability. It can be
black box nature. concluded that EM provides high accuracy and inter-
In SVM, a small number of samples has been found preting ability yet still poor in computational complexity,
among huge volume of data to describe the system dynam- configuration effort, ease of implementation, and real-
ics with kernel trick.40 Though SVM has a simple structure time applications. The performance of ECM is most
than ANN, a computationally expensive optimization prob- opposite to that of the Electrochemical model. Data-
lem has been solved to determine kernel parameters.42 driven models show moderate performance.
Radial Basis Function (RBF) kernel is the more popular The performance comparisons of various types of bat-
choice for determining kernel parameters, due to its good tery models, including their strengths and weaknesses,
generalization capability.55 Since SVM fails to handle a are outlined in Table 5. As per the above discussion, both
large amount of data, it is difficult to implement in SoC accuracy and simplicity are the main concerns for choos-
and perform SoH estimation of battery packs. ing a battery model to design a BMS.23 According to
ANN, DNN, and SVM methods use machine learning Table 4, care must be taken to select the appropriate
algorithms to estimate the SoC of a battery based on sta- model, model parameter type and parameter identifica-
tistical data and they are capable to predict nonlinear tion algorithm to enhance the accuracy as well as reduce
parameters of a battery. Among these, DNN performs the complexity.
better than ANN and SVM.39 Eventhough Deep learning
algorithms need high computational time and more
resources than others, they are well suited for Remaining 3 | S O C ES T I M A T I O N
Useful Life (RUL) prediction of the battery.56
However, DDM requires an intensive calculation to According to the statement of Plett,16,79 the SoC of bat-
understand the properties of battery in real-time based tery is the ratio of remaining charge to the total charge
on the data acquisition and training process. Data capacity of the battery.
TABLE 3 Comparison of data-driven model of battery

Training algorithm/ Activation
Ref. Model Parameters optimizer function Advantages Disadvantages
He et al,43 Ben Feed forward Hidden neurons Gradient descent Hyperbolic It is simple to execute Output of the FNN depends
sassi et al59 neural tangent and does not only on the current
network sigmoid necessitate intensive input, and it is inaccurate
(FFNN) activation computing or a to use the FNN process
function lengthy learning for time sequence
process. problems such as SOC
estimation because it fails
to model the temporal
information.
Hannan et al60 Back Learning rate, no. Levenberg– Sigmoid Simple execution and The robustness and
propagation of hidden layers, Marquardt (LM) activation flexible universality are
neural no. of neurons algorithm and function inadequate.
network backtracking Slow convergence rate, data
(BPNN) search algorithm overfitting and can easily
(BSA) be trapped into local
minima.
Zhang et al61 Radial basis Gaussing function Stochastic gradient Radiated Global approximation, Training at a slow pace
function centre, weights Gaussian rapid training/
neural kernel learning, and
network function proficiency in
(RBFNN) interpolation
Cui et al62 Wavelet Hidden neurons, Levenberg Marquardt Morlet wavelet Easy learning Need more layers and
neural wavelet -Weights function neurons to achieve better
network translation and accuracy
(WNN) dilation
parameter
Nagulapati Gaussian Amplitude of Kernel function Squared Nonparametric Kernel selection.
et al,63 process kernel function, exponential modeling and Accuracy of model depends
Deng et al,64 regression length scale of kernel probabilistic on the data diversity and
Babaeiyazdi (GPR) the distance function prediction, Able to size of the training
et al65 measure (RBF kernel) adapt hyperparameter dataset; When the
values training data are huge, it
will cause a large
computational burden.
Weng et al42 Support Kernel function Logistic regression Gaussian Simple structure, Less accurate than GPR.
vector and radial RBF computationally Training model depends on
regression regularization kernel efficient prognostic the correlation between
(SVR) parameter algorithm the training data features
and the target data.
Du et al,66 Extreme Hidden neurons Moore-Penrose Sigmoid Good generalisation Performance highly
Hossain learning generalized inverse function performance, better depends on training
Lipu et al67 machine operation scalability and fast accuracy.
(ELM) learning. No need to
tune the model
parameters during
training.
Wu et al68 Recurrent Grid search and the It can learn Limitation in capturing the
neural adaptive moment characteristics and length of the data;
network estimation method time dependencies Not suitable for long-term
(RNN) from sequential data sequences due to
using its internal state exploding gradient and
(memory). vanishing gradient.
Li et al46 GRU-RNN Time step, Back propagation Sigmoid Overcomes the short- The performance of GRU-
sampling through time function term dependence RNN increases with the
interval, batch (BPTT) problem of the simple hyper parameters such as
size, iteration, RNN, more robust to time step and iteration as
Weights and vanishing gradients, well as the training data
biases of self-learn network size; Takes more training
network parameters by time, Complex structure.
adaptive gradient
descent algorithms.
(Continues)
TABLE 3 (Continued)
Training algorithm/ Activation
Ref. Model Parameters optimizer function Advantages Disadvantages
Zhang et al45 LSTM-RNN Mini batch size, Back propagation Hyperbolic Capture longer More training time and
Yang et al51 Learning rate, through time tangent sequences of complex training process
gradient (BPTT) and function information without
threshold, no. of adaptive moment gradients vanishing.
nodes in the (ADAM) method.
layer.
Hasan et al69 NARX Hidden neurons Levenberg Marquardt Sigmoid Can model the temporal Accuracy depends on no. of
and delay for information (changes past inputs and feedback
input and in input/output over outputs
feedback. time) in a given time
series
Khumprom DNN Learning rate and Back-propagation ReLU Simple to perform, High computational time
et al39 momentum, stochastic gradient activation acceptable level of and more resources are
dropout rate descent function generalization ability, required
accurate results for
complicated
applications.
Remaining Charge
SoC ¼ :
Total Charge battery. If this relationship is linear, it gives accurate
results. But, it is not exactly linear for all types of batte-
More specifically, the reduction in SoC accuracy ries because it depends not only on size and type of elec-
reflects lower battery efficiency. Therefore, the accuracy trode materials but also on the electrolyte concentration
of SoC estimation is important in designing a BMS and it of the battery.80 Also, this method needs too much rest
is suitable for electric vehicles. time to estimate the accurate SoC of a battery. Of course,
The classification of SoC estimation methods is pri- this method is not practical to implement in a running
marily divided into two such as direct estimation electric vehicle, since this is valid only for the unloaded
methods and model-based estimation methods. condition of the battery.84 As a result, only direct mea-
surement methods are suitable only for low power appli-
cations not for real-time EV. Thus, the model-based
3.1 | Direct estimation methods estimation methods have been developed to save estima-
tion time by predicting the performance of battery with
The SoC can be calculated by directly measuring the volt- modeling.85
age and current of the battery. For example, the Ampere
hour (Ah) method is used to estimate the state of a battery
by integrating the charging and discharging currents. This 3.2 | Model based estimation methods
is a very simple and low computational method.80 Despite
the advantages of the Ah method, it is difficult to measure This method is further classified into two, such as Adap-
the initial SoC in real-time EV applications. Moreover, the tive filter and observer-based SoC Estimation and com-
accuracy of SoC is limited due to the unknown initial puter intelligence algorithm-based SoC Estimation.
capacity of the battery, the self-discharge rate, and the loss
of battery capacity. In general, the estimation using Ah is
done by calculating Peukert's effect and coulombic effi- 3.2.1 | Adaptive filter and observer-based
ciency. This can give accurate estimation when the battery SoC estimation
is discharged at a constant current rate and constant tem-
perature. Since EV batteries experience dynamic variations In general, well-known adaptive techniques of control
in load current and temperature during the charging and theory are combined with the battery model to achieve
discharging processes, they fail to estimate accurate SoC. an online SoC estimate. The familiar Adaptive filter and
Hence, some modifications have been done to calculate the observer algorithms are Kalman Filter (KF),85-87
Peukert's effect and coulombic efficiency subject to the cur- Extended Kalman Filter (EKF),88-91 Unscented Kalman
rent rate and temperature.81-83 Filter (UKF),92-96 Fading Kalman Filter (FKF),97,98
In OCV method, the SoC is directly measured by one- Cubature Kalman filter,72,99-101 Particle filter45 and H∞
to-one relation between the SoC and the OCV of the observer method.102,103 In all these methods, each
TABLE 4 Combination of models and identification algorithms used in literature

Parameter Parameters type
Battery identification (time invariant/
Reference model type algorithm time variant) Accuracy Complexity Issues
Chen 2RC-run time model Consider experimental Time variant Medium Medium Suitable only for the low-capacity
et al14 middle curve using battery, the temperature must be
Curve fitting constant
Yao et al25 2RC-ECM LS Time variant High Medium Suitable only for constant
temperature applications
Cao et al29 3RC-ECM LS Time invariant High. High Parameters must be corrected for
real-time EV
Zhang 2RC-ECM decoupled weighted Time invariant High High Parameters are not updated
et al30 recursive least squares
(DWRLS)
Meng 2RC-ECM Partial least square Time variant High High Selection of suitable width of the
et al71 (online) (PLS) moving window is difficult
regression
Xia et al72 2RC-ECM forgetting factor least- Time invariant High High Choosing a forgetting factor is
squares difficult.
Yang et al 2RC-ECM variable forgetting factor Time variant High High The effects of different C- rate and
73
recursive least square Temperature are not considered.
(VFFRLS) method
Xu et al74 Fractional order ECM Recursive LS, mixed- Time variant High High Initial Parameters are identified
swarm-based using MCPSO in offline then
cooperative particle updated by RLS.
swarm optimization
(MCPSO)
Liu et al75 1RC-ECM AUKF Time variant High High Choosing the value of alpha and
beta is difficult.
Yu et al76 1RC-ECM H ∞ filter Time variant High High Need noises must be a gaussian as
well as requires some knowledge
to choose noise covariances
based on statistics.
Fotouhi Simple R model and GD, GA, prediction error Time variant High High Temperature effect is not
et al23 1RC minimization (PEM). considered
Liu et al77 Electrochemistry model Internal resistance Time variant High High Parameters depend on SoC and
based on the Nernst model using Design of aging is not considered
equation Experiment, multiple
input parameter
fitting model
Ahmed Reduced EM and aging GA Time variant Medium Medium Erroneous SoC for the aged battery.
et al9,10 model
Du et al66 DDM-NN extreme learning Time invariant High High Computationally expensive model
machine (ELM)
Dang Dual NN fusion model Electrochemical Time variant Medium Low Improvement of the model
et al78 parameters are concerning aging and the
identified using linear temperature effects is complex.
NN, OCV determined
by Backpropagation
NN.
Samadi EM + state space = T-S Recursive LS Time variant Medium High More rule-based
et al54 fuzzy model
Chemali DDM Deep feedforward NN Time invariant Low Medium More training time, Selection of no.
et al44 of hidden layers, and no. of
neurons influence accuracy, as
well as the history of data.
Ignored
Boujoudar DDM- Nonlinear Auto Backpropagation Time variant Medium Medium Choosing the number of hidden
et al37 Regressive model with algorithm layer and number of neurons per
eXogenous input layer needs optimization
(NARX) technique, Over-learning of
ANN cause error.
Yang DDM LSTM-RNN Time variant High High More memory is required.
et al51
TABLE 5 Comparison of various battery models

4 Model Strength Weakness
3
Electrochemical • High accuracy • Complex model
model (EM) • Completely • Need high
Indexed values
2 describes the computational

electrochemical power
1 reaction of the • Not advisable for
DM
ECM
battery online control
0
EM
tasks
Equivalent • Simple model • Lack to describe
y
rt
ity
ns
y
ac
lit
tio
fo
io
ex
bi
ur
• Can implement
ef
ta
at
circuit model internal
pl
0-Very low
ta
cc
en
lic
n
m
tio
re
A
em
pp
o
1-Low
rp
lc
ra
(ECM) in real-time characteristics of

ea
pl
na
te
gu
2-Medium
im
im
In
tio
fi
a battery
on
of
lt
3-High
ta
ea
C
se
pu
4-Very high • Low accuracy

R
Ea
om
C
than other
models
FIGURE 4 Comparison of Battery Models with indexed values
Data-driven • The high • High
model (DDM) nonlinear computational
prediction complexity
algorithm has some advantages and disadvantages. For capability • Accuracy fully
example, KF is an optimal estimator and is often used to • Easy to dependent on
estimate the states of linear systems. In the case of implement Quality of
nonlinear systems, KF requires complex calculations. • Suitable for training data
Plett88 has proposed an EKF method for SoC estimation online • Need high
applications storage
of nonlinear battery model. However, if the system is
extremely nonlinear, linearization error occurs due to the
lack of accuracy, and also it necessitates more computa-
tional effort. appropriate one for the same. But, it also exhibits similar
Compared to EKF, the UKF may be implemented to problems as in KF based methods. Also, Accuracy and
provide better results for highly nonlinear model. Though convergence rate of this observer depend on gain. Deter-
it eliminates the linearization error, Cholesky factoriza- mining an observer gain is based on the adjustment coef-
tions on every step and sigma point selection affect the ficient which can be chosen by an optimization
performance. FKF is capable to correct the modeling technique. KF algorithms can handle initial SoC error
error using a fading concept. But, it needs more comput- due to their self-correcting nature. Therefore, KF algo-
ing power to implement the hardware. However, filter rithms are suitable to estimate the state of fast time-
parameters like noise covariance matrices affect the esti- varying systems when a well-parameterized model is
mation accuracy and convergence rate and those noises available.
must be a Gaussian. KF algorithms fail to handle the
non-gaussian noises. In general, noise covariance matri-
ces are chosen by trial and error method based on statisti- 3.2.2 | Computer intelligence algorithm-
cal knowledge.71,77,86,104 Besides, Adaptive Extended based SoC estimation
Kaman filter (AEKF),83,89,105 and Adaptive Unscented
Kaman filter (AUKF)75,106-108 are developed to automati- According to the rapid development of computer intelli-
cally update noise covariance matrices. Despite accurate gence algorithms, some machine learning algorithms
estimation, computational time and complexity have have been used to estimate the SoC of the battery.109,110
been increased. These algorithms can use training data to estimate SoC
However, the calculation complexities of all KF algo- without prior knowledge of battery chemistry. SoC has
rithms are identical. Besides, it requires a well- been estimated using fuzzy logic.111 In this fuzzy model,
parameterized battery model but often the parameters a maximum as well as a minimum voltage value is col-
are varied drastically for temperature changes and battery lected with the cycle number used as an input to generate
aging. For this purpose, the recursive least square algo- a fuzzy interference system.
rithm is recommended by many authors. If the system Due to self-learning, the ANN37,112 has been familiar
has parameter uncertainties and often errors occur in sys- with the validation of complex nonlinear model.
tem modeling, the H∞ observer method is the most Although it is extremely dependent on collective
information by training, it requires a smaller computa- terms of computational complexity, real-time applica-
tional cost. But, the challenge is that the ANN has the tions, and ease of implementation but lack accuracy,
risk of overlearning. As a result, the optimization algo- require more training, and high storage size than the
rithm is necessary to find the number of hidden layers others. Table 6 summarizes the information about the
and the number of neurons per layer.37 When large real- combinational technique that may comprise the model,
world data are available, DNN can be used to estimate identification technique, and an algorithm used for SoC
SoC without the need of filtering algorithms.44,113 DNN estimation.
can estimate SoC through online with low computational Based on the review, it is concluded that an adaptive
time, but it needs more time for an offline training pro- filter based SoC estimation can give accurate results with
cess, since it is a multi-layer architecture that increases the proper selection of battery model (EM/ECM/DDM),
the number of neurons. SVM uses a regression algorithm parameter type (Time variant/Time Invariant), and
to predict the parameters of the battery and estimates the parameter identification algorithm (Least square/
SoC of a battery. According to the structural risk minimi- optimization algorithm). On the other hand, computer
zation principle, SVM works for state estimation.114 intelligence algorithm-based SoC estimation requires an
Eventhough the standard SVM has a simple structure, appropriate model selection, an algorithm for tuning
it needs more computational effort to find the kernel hyperparameters, a proper training algorithm and a large
parameters. SVM provides a modular framework that can amount of data collection and normalization to extend a
adapt to any kind of problem with kernel functions. trade-off between accuracy and complexity.
While the ANN has a fixed structure and provides more
outputs, the size of SVM has increased much more and
offers only one output. 4 | CHAR GING AND
Hu et al,115 and Shen et al116 have estimated SoC DISCHARGING OF BATTERY
using ANFIS model. ANFIS utilizes the best features of
ANNs and Rule-based Fuzzy systems to learn the behav- To deploy the growth of EV, quick charging of Li-ion bat-
ior of a battery from the dataset and adjusts the parame- tery is a challenging task among EV manufacturers.136
ters of a battery according to error generation.38,117 Due to fast charging, the temperature increases inside the
Taimoor Zahid et al,118 have proposed the SoC estimation cell rapidly and hence, the battery life deteriorates.137
method based on Subtractive Clustering Neuro-Fuzzy Therefore, the optimal charging of Li-ion batteries has
architecture under 10 different unseen driving cycles. become an essential method to solve the problems in EV
Some of the researchers have proposed a combined applications.138-142 There are more influential factors in
model that uses both the types of algorithms. For example, the Li-ion battery charge curve and they include the ini-
Charkhgard et al,119 have created a battery model using tial state of charge, charge current rate, discharge current
neural network concepts and estimated the capacity using rate, temperature rise, depth of discharge, used cycle
EKF. Yang et al,51 have developed an estimation technique times, charging strategy, overcharge, over discharge, etc,
using LSTM and UKF. Likewise, Du et al,66 have devel- Hence, there are more constraints like charging time,
oped a neural network model using an extreme learning temperature rise, current rate, energy loss, charging effi-
machine and AUKF is used for SoC estimation. But, these ciency, state of charge, state of health, charging voltage
are not more attractive and they also make the calculations threshold, capacity fade, power fade, aging effect, capac-
more complex as well as increase the computational cost. ity utilization, and impedance rise to develop an optimal
Some joint estimation algorithms are developed like SoC charging strategy. Some constraints for developing an
and SoH estimation,33,120-122 SoC and SoE estimation,123-125 optimal charging strategy are shown in Figure 5.
and SoC and SoP estimation126-128 using aforementioned Many battery charging methods have been used so
estimation techniques. far, such as Constant Trickle Charging (CTC), Constant
Both the algorithms have been implemented in real- Current Charging (CC), Constant Voltage Charging (CV),
time applications and those methods include some Constant Current Constant Voltage Charging (CCCV),
advantages and certain drawbacks. It can be concluded Multi-Step Constant Current Charging (MCC), Boost
that direct estimation methods are simple and low-cost Charging (BC), and Pulse charging (PC). The constant
but not suitable for real-time applications. Adaptive filter trickle charging method is a simple, safe, and low-cost
and observer methods are the best in terms of accuracy, charging method but it takes more than 10 hours for full
storage size, and suitable for real-time applications yet charging by making it an “Overnight Charger.”57
still poor in terms of computational complexity, configu- To reduce the charging time, the CC method has been
ration effort, and ease of implementation. Computer used. In the CC method, the value of charging current
intelligence methods furnish superior performance in has been increased to achieve faster charging times. But,
TABLE 6 Summary of the SoC estimation at different combinations
Authors Algorithm execution Remarks Accuracy Complexity Technical challenges

Leksono Modified coulomb counting Peukert's effect is modified Low Low The temperature effect is
et al82 with the consideration of not considered.
current rate parameters.
Therefore, SoC can be
calculated at different
current rates.
Xie et al 81 Enhanced coulomb Peukert's law and Low Medium Accuracy is reduced under
counting coulombic efficiency are dynamic load variations
considered.
Nugroho Coulomb counting with Overcome the shortcomings Medium Medium The temperature effect is
et al32 dynamic model (integral of traditional Coulomb not considered
controller) counting method
Weng et al OCV + EKF This model gives accuracy High Medium It requires life cycle data
129
like EM. ICA gives whereas practically it is
charging capacity with difficult.
respect to a terminal
voltage that gives an
accurate V-Q curve. A
small error is corrected by
EKF.
Barai OCV New OCV hysteresis test is High Low Need more time
et al130 performed with respect to
parameters which affect
the hysteresis voltage.,
Hysteresis voltage is a
function of SoC that
improves accuracy
Zheng Low current OCV and Incremental OCV gives High Low The incremental OCV test
et al131 Incremental OCV better accuracy than the consists of many SoC
low current OCV method. intervals and rest periods
after which the OCV with
the corresponding SoC is
observed.
Dong Invariant-imbedding- Parameters are updated Medium Medium Possibility of Measurement
et al132 method (IIM) + OCV using IIM error reduces accuracy.
Xiong H∞ filter + OCV Reduces time taken for a Low High The open-loop method, low
et al102 full test; OCV can be accuracy
obtained using current–
voltage measurement by
the filter.
Awadallah ANFIS + OCV Accurate than CC Medium High The OCV-SoC relationship
et al38 estimated SoC of a may vary due to aging.
battery pack with the
consideration of
temperature effects.
Feng et al Enhanced coulomb SoC is estimated using Medium Medium Measuring OCV is very
133
counting + multi-state Enhanced coulomb difficult while EV is
OCV counting with the running.
consideration of
temperature. OCV
method is used to reduce
initial SoC error.
TABLE 6 (Continued)

Meng et al 71
2RC-ECM + Partial least Parameters are Very high High Selection of suitable width
square (online) (PLS) automatically updated, of the moving window is
regression +KF and SoC is considered difficult
only as a state, compared
to 2RC ECM-EKF and
AUKF with LSSVM
Wang et al87 ECM + Fixed parameter by considering the modeling Medium Low Model parameters are
experimental test + KF error, SoC-OCV curve constant, not suitable for
and Bias Corrected KF error, system noise error running EV.
and measurement noise
error are analyzed, KF-
based SoC and model
error estimators have
been established
Liu et al77 Electrochemistry model Temperature effect during High High Computationally complex
based on Nernst eqn. SoC estimation is because two different
+ Internal resistance considered and it reduces parameter identification
model using Design of experimental workload techniques are used and
Experiment + multiple because it needs current SoC is further corrected
input parameter fitting measurement data by EKF. Parameters
model + EKF depend on SoC and aging
is not considered.
Xu et al74 Fractional order ECM Reduces battery model High Very high Three methods such as
+ Recursive LS, Mixed- error using DEKF and EKF, KF and Ah are used
swarm-based Cooperative current measurement for single-cell SoC
Particle Swarm error using DKF estimation. Initial
Optimization (MCPSO) Parameters are identified
+ Dual KF, Dual EKF, using MCPSO in offline
Adaptive dual KF and then updated by RLS.
A very complex method
as well as too difficult to
implement in pack level.
Yang et al105 2RC-ECM-split battery By splitting model, cross- High High Model parameters are only
model + AEKF interference between RC considered as a
voltages and SoC is univariant function of
eliminated; Therefore, SoC; Not considering the
convergence rate is effect of temperature and
increased aging.
Xia et al72 2RC-ECM + forgetting More accuracy compared to High Very high Choosing a forgetting factor
factor least-squares + EKF and Cubature KF is difficult. Accurate
Adaptive Cubature KF measurement data are
needed for parameter
identification. Parameters
are not updated as well as
computational cost is
more.
Peng 2RC-ECM + LS-curve High accuracy, more Very high Very high High computational cost,
et al107,108 fitting + AUKF convergence rate, Need predetermined
parameters are updated; sampling points for
process, and AUKF.
measurement noise
matrix are automatically
updated.
(Continues)
TABLE 6 (Continued)

Ben Sassi 1RC-ECM + Non-linear Comparison has been done. High High A complex method,
et al59 Recursive LS + UKF UKF gives reasonable Parameters are not
accuracy and suitable for updated, predetermined
online estimation sampling points are
required and statistical
knowledge is required to
choose noise covariance
matrices.
Du et al66 DDM-NN + extreme Fast and good learning High Very high Computationally expensive
learning machine (ELM) performance with a model and complex
+ AUKF simple process, compared estimation algorithm
to EKF, UKF, and AEKF. gives high accuracy. Need
predetermined sampling
points for AUKF.
Xie et al134 1RC-ECM + Multiple Estimated SoC of battery High Very high Parameters are identified in
linear regression (MLR) packs. offline only, not updated
+ EKF, Multi Model EKF In MMEKF, different initial concerning with SoC,
(MMEKF), Adaptive values of noises are used aging, temperature.
fading EKF (AFEKF) for different models. In Statistical knowledge
AFEKF, the forgetting which are required to
factor is introduced to choose noise covariance
restrict the memory usage matrices. Three
of EKF. algorithms are used with
switching strategy.
Roscher 1RC-ECM + Recursive LS Reasonable accuracy Medium Medium Feedforward structure,
et al135 + Luenberger observer Parameters are not
updated
Xue et al103 1RC-ECM + linear Handle the uncertainty of Medium Medium Accuracy and convergence
Interpolation + H∞ the parameters. The effect speed depend on the
observer of parameter correction is OCV-SoC curve and an
analyzed. optimization algorithm is
needed to find an
adjustment coefficient
that determines the
observer gain matrix.
Samadi EM + state-space = T-S The stability and robustness High High More knowledge is required
et al54 fuzzy model + Recursive of the observer are to create rules.
LS + T-S Luenberger and addressed. T-S
H∞ observer Luenberger is better than
H∞ observer
Charkhgard DDM- NN + EKF Without the prior Low High Leads to Computationally
et al119 knowledge of battery complex. Learning rate
chemistry, battery and no. of neurons are
modeling has been done chosen by trial and error
and the process noise method; These are very
covariance matrix value is important for the
also estimated to improve accuracy of SoC.
the convergence rate of Linearization error occurs
EKF. on EKF.
Fotouhi 1RC-ECM + Prediction- Parameters are identified High Medium The open-loop method,
et al23 Error Minimization and updated at every time Need more time
(PEM) algorithm + step. ANFIS is used to
ANFIS find the relationship
between OCV and SoC.
TABLE 6 (Continued)

Chemali Deep feedforward NN Better computational Medium Medium Bulk data is required for
et al44 efficiency as well as it different C-rates, SoC,
needs low memory. and temperature range.
Computational time is More training time is
less than EKF. needed. Selection of no.
of hidden layers and no.
of neurons influences
accuracy. History of data
ignored.
Boujoudar DDM (NARX) + BPNN + Model created without the High Medium Accuracy depends on data
et al37 FFNN knowledge of battery collection and training.
chemistry Choosing a number of
hidden layer and number
of neurons per layer
needs optimization
technique. Over-learning
of ANN causes error.
Yang et al51 DDM (LSTM-RNN) + UKF Mainly focuses on the effect High Very high High Complexity and more
of temperature and OCV- memory are required.
SoC relationship to Need predetermined
estimate SoC of battery. sampling points.
F I G U R E 5 Constraints to
develop optimal charging strategy
Charging
time
Capacity Temperatur
utilization e rise
Aging Charging
effect efficiency
Charging
Strategy
Current
Energy loss
rate
Charging
State of
voltage
charge
threshold
Impedance State of
rise health
the large charging current requires additional control cir- swarm optimization,145-147 genetic algorithm,137,148,149
cuitry to determine the charging status and to terminate dynamic programming algorithm,150 and multi- objective
the charging process, when the battery attains 100% full biogeography-based optimization141,151 have been used to
charge. Also, increasing the charging current leads to find the optimal values for each step of MCC. However,
increase loss of capacity and reduces the battery life cycle capacity loss due to electrolyte decomposition may occur
due to significant negative effects on the ion concentra- when switching at different current rates. As a result, a
tion between two electrodes. multi-step constant current constant voltage (MCCCV)
On CV charging, the charger gets current to make the method147,149,152,153 is employed by combining short time
battery reach its nominal voltage. After that, it provides the constant voltage with MCC to avoid the capacity loss and
required amount of current to keep the voltage constant at ensure full charging.
the same point. The accuracy of setting this voltage is To reduce the charging time, the BC method has been
essential, since the battery life cycle is reduced, when the recommended. In this method, high voltage is applied to
voltage is high. At the same time, the battery cannot be charge the battery for a short period (ie, 5 or 10 minutes)
fully charged, when the voltage is set to a low point. Fur- and is known as the boost period. In this boost period,
thermore, the temperature increases, due to rapid changes the battery attains a significant amount of charge. After
in current by forcing the voltage to reach its nominal level. that, the standard CCCV approach has been used with
Subsequently, the CCCV charging method has been pro- the low constant current value. Eventhough it is faster
posed by combining two conventional charging methods. than the CCCV approach, it is necessary to discharge the
For Li-ion battery, the CCCV method is the most accepted battery fully before starting the charging. Hence, this
and widely used charging method in the case of fast char- method is not advisable for real-time EV.
gers.139 Under the CCCV method, the first CC mode is ON, PC is a verified one for quick and efficient charging.
that is, the constant current is provided to charge the bat- The drawback of PC is its difficulty in picking the correct
tery until the terminal voltage reaches the nominal value charge pulse.136,154,155 Recently, the modified form of PC
and then, the CV mode is ON, that is, the battery is uses a sinusoidal wave in the place of square wave. How-
charged with constant voltage to the extent that the battery ever, both the types of PC methods require optimal
current reaches its lower limit. Nonetheless, the CV mode charging frequency that can be achieved at low imped-
takes the charging time as many (approximately 3 times) ance which maximizes the charging current. Table 7
as the CC mode. Consequently, it prolongs the overall bat- summarizes the advantages and disadvantages of those
tery charging time. charging methods.
The MCC reduces the charging time and controls the Optimization-based charging strategy is too difficult
temperature rise, but it becomes problematic, when to fix to implement in real time applications because optimiza-
the constant current value for each charging step. To tion process involves iterations that consume more com-
overcome the issue, some soft computing algorithms like putation time. Therefore, a database-based dynamic
Taguchi method,143 ant-colony algorithm,144 particle programming method has been proposed to find the
TABLE 7 Keynotes of different charging methodologies
Core idea Advantages Disadvantages

CTC Simple method Too long charging time
Low-cost charging
CC Fast charging Capacity loss increases, shorten life cycle of battery
CV Simple charging Produces high temperature inside the cell
Deteriorates the battery
CCCV Computationally efficient More charging time due to slow charging in CV mode
Easy to implement
MCC Reduces the charging time Difficult to find charging points at each step
Prolongs the life span of a battery
BC-CCCV Fast charging Discharge first before charging
Not suitable for EV
PC Fast and efficient charging Difficult to pick correct charging pulses
S-PC Medium level charging Optimal frequency achieved at low impedance only which increases current
resulting high temp and shorten the life cycle of battery
suboptimal charging strategy with a trade-off between that give the best charging speed and time are worst in
the charging loss and the charging time. battery deterioration, ease of implementation, and vice
In the CCCV method, high energy loss in CC mode versa. No such strategy can fulfill all the requirements of
causes a temperature rise and more charging time in CV the users.
mode reduces the charging efficiency. To accomplish fast Both charging and discharging have impacts on bat-
charging, high charge current is applied during the initial tery life. In the constant current discharging test, a higher
stage and then, the current rate is reduced in later stages. discharge rate results in a lower discharge capacity,
Although high current rate reduces the charging time, it which means the battery can discharge less energy at a
increases the temperature rise, energy loss, and perfor- higher discharge rate,83 since a small amount of energy is
mance degradation such as Solid Electrolyte Interphase released and the battery requires less charging time.
(SEI) growth and Lithium plating deposition. To control Because of the effects of high discharge rate and aging,
the temperature rise, the Fuzzy Temperature Rise (FTR) the time required for the CC mode of CCCV charging
controller,156 Recurrent fuzzy neural network,157 Gener- method has also decreased and it has an impact on the
alized Regression Neural Network (GRNN) controller158 entire charging process.
and Deep neural network159 have been developed to fine EV operates in a variety of driving conditions includ-
tune the current rate with the consideration of charging ing sharp curves, rough roads, straight lines, and so
time and temperature rise. on. Due to these harsh conditions, electric vehicles con-
Fast intelligent charging algorithm has been devel- sume various amounts of power and heat dissipation and
oped using a Neural-Fuzzy approach160 In this approach, the current rate is required from the battery. The Federal
the charging current rate is controlled by the SoC, volt- Urban Driving Schedule (FUDS),166 the dynamic stress
age, and temperature. Based on these parameters, the test (DST),104 the New European Driving Cycle
current rate is frequently altered to prevent overcharging (NEDC),167 the Urban Dynamometer Driving Schedule
and temperature rise. Moreover, Liu et al,161 have pro- (UDDS)24 and the Highway Fuel Economy Test
posed a Fuzzy Logic Control-based (FLC-based) five- (HWFET)10 are the most common driving profiles used
stage charging strategy to reduce the charging time, in the literature to verify the effectiveness of the SoC esti-
increase the charging efficiency, reduce the average tem- mation algorithm in recent years.
perature rise and maximize the cycle life of Li-ion
battery.
To overcome the drawback of CV mode in the stan- 5 | RECOMMENDATIONS TO
dard CCCV method of charging, CV mode is replaced by DEVELOP PROPER BMS FOR EV
Grey Predicted (GP) mode.57 Grey controlled current
source gives near optimal charging current to decrease To enhance the precision, scalability, extension of driving
the charging time. A model-free reinforcement learning range and other challenges of the BMS in electric vehi-
framework162 is used to develop charging strategy with cles, some issues, such as joint estimation of states, scal-
the trade-off between fast charging and aging. Likewise, able state estimation, smart learning and optimization,
Suresh et al163 have developed a model predictive and safety management are worth studying in-depth to
control-based optimal current profile with the consider- fulfill the requirements of EV. Table 8 shows the recom-
ation of charging speed and capacity fade during charg- mendations of the research work to develop a suitable
ing/discharging of the battery. BMS for EV.
Closed-loop charging systems, that use instantaneous
cell voltage and temperature to adjust charging current
magnitude, are required. Patnaik et al,164 have presented 5.1 | Joint estimation of states
a constant-temperature constant-voltage (CT-CV) charg-
ing strategy that takes the cell temperature into account Generally, battery states are joined together. Most stud-
as a critical deterioration measure. However, fast charg- ies concentrate on single state estimation. But, a few
ing deteriorates the battery's electrical characteristics and studies focus on joint estimation. Even though it gives
affects its life time and the authors165 have developed a satisfactory results under certain constraints, it has
closed-loop optimization (CLO) system with an early out- some limitations, because of a powerful bond between
come prediction for an efficient optimization over large three or more states in real-time applications. Conse-
parameter spaces. quently, a proper BMS has to be developed to estimate
So far, many charging strategies have been discussed all the states (the State of Charge, State of Health, State
here in terms of charging speed, time, battery deteriora- of Energy, State of Power and State of Temperature) of
tion, and ease of implementation. However, the methods the battery reliably.
TABLE 8 Challenges for recommended work
Recommended work Remarks Challenges

Joint estimation of states Each state is important for vehicle Developing a single algorithm for all states is too difficult.
performance and all states are States are varying in different time scales.
correlated.
Scalable state estimation The behavior of a battery pack is The algorithm developed for a single cell is not enough for
needed to determine the driving the pack level. Because, the characteristics of each cell are
range and manage the available varied to operating conditions. An essential cell balancing
energy for different tasks. technique is needed.
Smart learning and The learning algorithm can handle a Proper data collection, data pre-processing and training
optimization large amount of data that will be algorithm are required. An algorithm is required to handle
collected from a battery pack. different time scale data. Training time could be reduced
and it must be adopted for fast time-varying systems.
Safety Management High energy density batteries Developing an Electro-thermal model is difficult to estimate a
necessitate a safe operation. state of temperature. The safety management circuit must
Because it is very sensitive to handle the electrical, thermal, electrochemical, and
temperature. mechanical issues.
5.2 | Scalable state estimation explosions. As a result, calculating the state of safety
linked with thermal processes is very important. Even
Most of the literatures concentrate on cell level estima- though thermal management plays a vital role in battery
tion. But for EV applications, there is a need for a safety, the state of safety monitoring involves various
research effort to address better scalability of state esti- fields such as electrical, electrochemical, mechanical, and
mation from battery cell level to battery module level and thermal as well as they will become noteworthy for the
battery pack level. In the case of a lumped arrangement, lifecycle prediction of battery.
the entire battery system can be viewed as a single large
battery cell, and as a result, only a single state estimator
is used to seek minimal computing cost, with no capture 6 | CONCLUSION
of cell interactions and inconsistencies. Extensive study is
still required to effectively address the scalability issue. The battery management system plays a vital role in
monitoring and controlling battery energy storage sys-
tems in electric vehicles. Since the development of bat-
5.3 | Smart learning and optimization tery technology has been increasing, the performance
enhancement of the battery management system in elec-
Artificial intelligence algorithms are often used for SoH tric vehicles is an important factor to be considered
and Remaining Useful Life (RUL) prediction of batteries. because it controls the life of the costly battery. The most
Whereas these algorithms are unable to describe the important facts of the battery management system have
internal characteristics of the battery. In addition to that, been reviewed in this paper, especially in the fields of
a large amount of training data is needed and the train- battery modeling, state of charge estimation, and battery
ing process is also time-consuming. Much more effort is charging. To improve the accuracy while reducing the
needed to develop an intelligent BMS to describe the bat- complexity of battery modeling, selection of a suitable
tery behaviors as well as RUL prediction of batteries. model type as well as a parameter identification algo-
rithm is required. Based on the review of SoC estimation,
it is stated that an adaptive filter-based state of charge
5.4 | Safety management estimation may provide reliable results when the battery
model, parameter type, and parameter identification
Most researchers have focused on thermal management algorithm are properly selected. On the other hand, com-
for battery safety. For example, if a battery's actual oper- puter intelligence-based state of charge estimation neces-
ating temperature surpasses the upper permitted temper- sitates an appropriate model selection, an algorithm for
ature, the SEI layer and the active components may tuning hyperparameters, a proper training approach, and
decompose and they may potentially lead to extreme exo- a considerable amount of data collection as well as nor-
thermic reactions, thermal runaway, and even malization to provide a trade-off between accuracy and
complexity. Careful attention is needed to develop an J Energy Res. 2020;44(4):2535-2548. https://doi.org/10.1002/er.

optimal charging strategy with the consideration of cer- 4876
tain constraints such as charging time, charging effi- 7. Doyle M, Fuller TF, Newman J. Modeling of Galvanostatic
charge and discharge. J Electrochem Soc. 1993;140(6):1526-
ciency, capacity fade, state of charge, Impedance rise,
1533.
temperature rise and so on. However, most of the key 8. Di Domenico D, Fiengo G, Stefanopoulou A. Lithium-ion bat-
technologies in the battery management system are tery state of charge estimation with a Kalman filter based on a
achieved and validated in specific laboratory test condi- electrochemical model. Paper presented at: 2008 IEEE Inter-
tions. But, the real-world performance is different from national Conference on Control Applications. 2008. https://
laboratory results and hence, it is difficult to extend guar- doi.org/10.1109/CCA.2008.4629639
antee. In order to reduce the range anxiety of electric 9. Ahmed R, El Sayed M, Arasaratnam I, Tjong J, Habibi S.
Reduced-order electrochemical model parameters identifica-
vehicle customers and prevent cell inconsistency in the
tion and state of charge estimation for healthy and aged Li-
battery pack, an accurate and simple algorithm has to be
ion batteries—part II: aged battery model and state of charge
developed for state of charge estimation. The algorithm estimation. IEEE J Emerg Sel Top Power Electron. 2014;2(3):
must be validated for the dynamic conditions of an elec- 678-690. https://doi.org/10.1109/jestpe.2014.2331062
tric vehicle. Much more consideration is required to 10. Ahmed R, El Sayed M, Arasaratnam I, Tjong J, Habibi S.
design and develop fast charging algorithms without Reduced-order electrochemical model parameters identifica-
affecting the life span of costly battery and to increase the tion and SOC estimation for healthy and aged Li-ion batteries
popularity of electric vehicles worldwide. Recommenda- part I: parameterization model development for healthy batte-
ries. IEEE J Emerg Sel Top Power Electron. 2014;2(3):659-677.
tions for research work on different categories of battery
https://doi.org/10.1109/jestpe.2014.2331059
management system are also listed in this paper. 11. Han X, Ouyang M, Lu L, Li J. Simplification of physics-based
electrochemical model for lithium ion battery on electric vehi-
A C K N O WL E D G E M E N T cle. Part I: diffusion simplification and single particle model.
The authors gratefully acknowledge The Alagappa J Power Sources. 2015;278:802-813. https://doi.org/10.1016/j.
Chettiar Government College of Engineering and Tech- jpowsour.2014.12.101
nology, Karaikudi for the financial support under TEQIP. 12. Zou C, Manzie C, Nesic D. A framework for simplification of
PDE-based lithium-ion battery models. IEEE Trans Control
Syst Technol. 2016;24(5):1594-1609. https://doi.org/10.1109/
ORCID TCST.2015.2502899
Maheshwari Adaikkappan https://orcid.org/0000-0002- 13. Sitterly M, Wang LY, Yin GG, Wang C. Enhanced identifica-
7721-8679 tion of battery models for real-time battery management.
Nageswari Sathiyamoorthy https://orcid.org/0000-0003- IEEE Trans Sustain Energy. 2011;2(3):300-308. https://doi.org/
10.1109/TSTE.2011.2116813
1563-4711
14. Chen M, Rincon-Mora GA. Accurate electrical battery model
capable of predicting runtime and I-V performance. IEEE
R EF E RE N C E S Trans Energy Convers. 2006;21(2):504-511. https://doi.org/10.
1. How DNT, Hannan MA, Hossain Lipu MS, Ker PJ. State of 1109/TEC.2006.874229
charge estimation for lithium-ion batteries using model-based 15. Hu X, Li S, Peng H. A comparative study of equivalent circuit
and data-driven methods: a review. IEEE Access. 2019;7: models for Li-ion batteries. J Power Sources. 2012;198:359-367.
136116–136136. https://doi.org/10.1109/ACCESS.2019.2942213 https://doi.org/10.1016/j.jpowsour.2011.10.013
2. Shrivastava P, Soon TK, Bin Idris MYI, Mekhilef S. Overview of 16. Plett GL. Extended Kalman filtering for battery management
model-based online state-of-charge estimation using Kalman fil- systems of LiPB-based HEV battery packs - part 2. Modeling
ter family for lithium-ion batteries. Renew Sust Energ Rev. 2019; and identification. J Power Sources. 2004;134(2):262-276.
113:109233. https://doi.org/10.1016/j.rser.2019.06.040 https://doi.org/10.1016/j.jpowsour.2004.02.032
3. Xing Y, Ma EWM, Tsui KL, Pecht M. Battery management 17. Nejad S, Gladwin DT, Stone DA. A systematic review of
systems in electric and hybrid vehicles. Energies. 2011;4(11): lumped-parameter equivalent circuit models for real-time esti-
1840-1857. https://doi.org/10.3390/en4111840 mation of lithium-ion battery states. J Power Sources. 2016;
4. Wang Z, Hong J, Zhang L, Liu P. Voltage fault detection and 316:183-196. https://doi.org/10.1016/j.jpowsour.2016.03.042
precaution of batteries based on entropy and standard devia- 18. Xu Z, Gao S, Yang S. LiFePO4 battery state of charge estima-
tion for electric vehicles. Energy Procedia. 2017;105:2163-2168. tion based on the improved Thevenin equivalent circuit model
https://doi.org/10.1016/j.egypro.2017.03.611 and Kalman filtering. J Renew Sustain Energy. 2016;8(2):
5. Gallardo-Lozano J, Romero-Cadaval E, Milanes-Montero MI, 024103. https://doi.org/10.1063/1.4944335
Guerrero-Martinez MA. Battery equalization active methods. 19. Siguang L, Chengning Z. Study on battery management sys-
J Power Sources. 2014;246:934-949. https://doi.org/10.1016/j. tem and lithium-ion battery. Paper presented at: Proceedings -
jpowsour.2013.08.026 2009 International Conference on Computer and Automation
6. Zhang Z, Zhang L, Hu L, Huang C. Active cell balancing of Engineering, ICCAE 2009, 2009, pp. 218–222. https://doi.org/
lithium-ion battery pack based on average state of charge. Int 10.1109/ICCAE.2009.11
20. Id JM, Luo G, Ricco M, Swierczynski M. Overview of lithium- 33. Qiu X, Wu W, Wang S. Remaining useful life prediction of
ion battery modeling methods for state-of-charge estimation lithium-ion battery based on improved cuckoo search particle
in electrical vehicles. Appl Sci. 2018;8:659. https://doi.org/10. filter and a novel state of charge estimation method. J Power
3390/app8050659 Sources. 2020;450:1-13. https://doi.org/10.1016/j.jpowsour.
21. Omar N, Widanage D, Abdel Monem M, et al. Optimization 2020.227700
of an advanced battery model parameter minimization tool 34. Malik A, Zhang Z, Agarwal RK. Extraction of battery parame-
and development of a novel electrical model for lithium-ion ters using a multi-objective genetic algorithm with a non-
batteries. Int Trans Electr Energy Syst. 2014;24:1747-1767. linear circuit model. J Power Sources. 2014;259:76-86. https://
22. Lai X, Gao W, Zheng Y, Ouyang M, Li J, Han X. A compara- doi.org/10.1016/j.jpowsour.2014.02.062
tive study of global optimization methods for parameter iden- 35. Sitterly M, Wang LY, Yin G. Enhanced Identification Algo-
tification of different equivalent circuit models for Li-ion rithms for Battery Models under Noisy Measurements. SAE
batteries. Electrochim Acta. 2019;295:1057-1066. https://doi. Int. 2010. https://doi.org/10.4271/2010-01-1768
org/10.1016/j.electacta.2018.11.134 36. Sitterly MW, Wang LY. Identification of battery models for
23. Fotouhi A, Auger DJ, Member S, Propp K, Longo S. Accuracy enhanced battery management. Paper presented at: NSTI-
versus simplicity in online battery model identification. IEEE Nanotech 2010, 2010, 3, pp. 809–812.
Trans Syst Man Cybern Syst. 2018;48(2):195-206. https://doi. 37. Boujoudar Y, Elmoussaoui H, Lamhamdi T. Lithium-ion bat-
org/10.17862/cranfield.rd.3545847 teries modeling and state of charge estimation using artificial
24. Fotouhi A, Propp K, Auger DJ. Electric vehicle battery model neural network. Int J Electr Comput Eng. 2019;9(5):3415-3422.
identification and state of charge estimation in real world https://doi.org/10.11591/ijece.v9i5.pp3415-3422
driving cycles. Paper presented at: 2015 7th Computer Science 38. Awadallah MA, Venkatesh B. Accuracy improvement of SOC
and Electronic Engineering Conference (CEEC): Conference estimation in lithium-ion batteries. J Energy Storage. 2016;6:
Proceedings: September 24-25, 2015, University of Essex, UK, 95-104. https://doi.org/10.1016/j.est.2016.03.003
2015, pp. 243–248. 39. Khumprom P, Yodo N. A data-driven predictive prognostic
25. Yao LW, Aziz JA, Kong PY, Idris NRN. Modeling of lithium- model for lithium-ion batteries based on a deep learning algo-
ion battery using MATLAB/Simulink. Paper presented at: rithm. Energies. 2019;12(4):1-21. https://doi.org/10.3390/
IECON 2013-39th Annual Conference of the IEEE Industrial en12040660
Electronics Society, 2013, pp. 1729–1734. 40. Klass V, Behm M, Lindbergh G. Capturing lithium-ion battery
26. Hu Y, Wang YY. Two time-scaled battery model identification dynamics with support vector machine-based battery model.
with application to battery state estimation. IEEE Trans Con- J Power Sources. 2015;298:92-101. https://doi.org/10.1016/j.
trol Syst Technol. 2015;23(3):1180-1188. https://doi.org/10. jpowsour.2015.08.036
1109/TCST.2014.2358846 41. Junping W, Quanshi C, Binggang C. Support vector machine
27. Yang K, Tang Y, Zhang Z. Parameter identification and state- based battery model for electric vehicles. Energy Convers
of-charge estimation for lithium-ion batteries using separated Manag. 2006;47(7–8):858-864. https://doi.org/10.1016/j.encon
time scales and extended Kalman filter. Energies. 2021; man.2005.06.013
14(1054):1-15. 42. Weng C, Sun J, Peng H. Model parametrization and adapta-
28. Wang A, Jin X, Li Y, Li N. LiFePO4 battery modeling and tion based on the invariance of support vectors with applica-
SOC estimation algorithm. Paper presented at: Proceedings of tions to battery state-of-health monitoring. IEEE Trans Veh
the the 29th Chinese Control and Decision Conference Technol. 2015;64(9):3908-3917. https://doi.org/10.1109/TVT.
(2017CCDC): May 28-30, 2017, Chongqing, China, 2017, 2014.2364554
pp. 7574–7578. 43. He W, Williard N, Chen C, Pecht M. State of charge estima-
29. Cao Y, Kroeze RC, Krein PT. Multi-timescale parametric elec- tion for Li-ion batteries using neural network modeling and
trical battery model for use in dynamic electric vehicle simu- unscented Kalman filter-based error cancellation. Int J Electr
lations. IEEE Trans Transp Electrif. 2016;2(4):432-442. https:// Power Energy Syst. 2014;62:783-791. https://doi.org/10.1016/j.
doi.org/10.1109/TTE.2016.2569069 ijepes.2014.04.059
30. Zhang C, Allafi W, Dinh Q, Ascencio P, Marco J. Online estima- 44. Chemali E, Kollmeyer PJ, Preindl M, Emadi A. State-of-
tion of battery equivalent circuit model parameters and state of charge estimation of Li-ion batteries using deep neural
charge using decoupled least squares technique. Energy. 2018; networks: a machine learning approach. J Power Sources.
142:678-688. https://doi.org/10.1016/j.energy.2017.10.043 2018;400:242-255. https://doi.org/10.1016/j.jpowsour.2018.
31. Thirugnanam K, Joy TPER, Singh M, Kumar P. Mathematical 06.104
modeling of li-ion battery using genetic algorithm approach 45. Zhang C, Zhu Y, Dong G, Wei J. Data-driven lithium-ion bat-
for V2G applications. IEEE Trans Energy Convers. 2014;29(2): tery states estimation using neural networks and particle fil-
332-343. https://doi.org/10.1109/TEC.2014.2298460 tering. Int J Energy Res. 2019;43(14):8230-8241. https://doi.
32. Nugroho A, Rijanto E, Wijaya FD, Nugroho P. Battery state of org/10.1002/er.4820
charge estimation by using a combination of coulomb cou- 46. Li C, Xiao F, Fan Y. An approach to state of charge estimation
nting and dynamic model with adjusted gain. Paper presented of lithium-ion batteries based on recurrent neural networks
at: Proceeding - 2015 International Conference on Sustainable with gated recurrent unit. Energies. 2019;12(9):1592. https://
Energy Engineering and Application: Sustainable Energy for doi.org/10.3390/en12091592
Greater Development, ICSEEA 2015, 2016. https://doi.org/10. 47. Ardeshiri RR, Ma C. Multivariate gated recurrent unit for bat-
1109/ICSEEA.2015.7380745 tery remaining useful life prediction: a deep learning
approach. Int J Energy Res. 2021;45:16633–16648. https://doi. discrete wavelet transform-based wavelet neural network.
org/10.1002/ER.6910 Energies. 2018;11(4):995-1012. https://doi.org/10.3390/
48. Xiao F, Li C, Fan Y, Yang G, Tang X. State of charge estima- en11040995
tion for lithium-ion battery based on Gaussian process regres- 63. Nagulapati VM, Lee H, Jung DW, et al. A novel combined
sion with deep recurrent kernel. Int J Electr Power Energy multi-battery dataset based approach for enhanced prediction
Syst. 2021;124:106369. https://doi.org/10.1016/j.ijepes.2020. accuracy of data driven prognostic models in capacity estima-
106369 tion of lithium ion batteries. Energy AI. 2021;5:100089.
49. Ungurean L, Micea MV, Cârstoiu G. Online state of health https://doi.org/10.1016/j.egyai.2021.100089
prediction method for lithium-ion batteries, based on gated 64. Deng Z, Hu X, Lin X, Che Y, Xu L, Guo W. Data-driven state
recurrent unit neural networks. Int J Energy Res. 2020;44(8): of charge estimation for lithium-ion battery packs based on
6767-6777. https://doi.org/10.1002/ER.5413 Gaussian process regression. Energy. 2020;205:118000. https://
50. Zhang Y, Xiong R, He H, Pecht MG. Long short-term memory doi.org/10.1016/j.energy.2020.118000
recurrent neural network for remaining useful life prediction 65. Babaeiyazdi I, Rezaei-Zare A, Shokrzadeh S. State of charge
of lithium-ion batteries. IEEE Trans Veh Technol. 2018;67(7): prediction of EV Li-ion batteries using EIS: a machine learn-
5695-5705. https://doi.org/10.1109/TVT.2018.2805189 ing approach. Energy. 2021;223:120116. https://doi.org/10.
51. Yang F, Zhang S, Li W, Miao Q. State-of-charge estimation of 1016/j.energy.2021.120116
lithium-ion batteries using LSTM and UKF. Energy. 2020;201: 66. Du J, Liu Z, Wang Y. State of charge estimation for Li-ion bat-
117664. https://doi.org/10.1016/j.energy.2020.117664 tery based on model from extreme learning machine. Control
52. Kaur K, Garg A, Cui X, Singh S, Panigrahi BK. Deep learning Eng Pract. 2014;26(1):11-19. https://doi.org/10.1016/j.
networks for capacity estimation for monitoring SOH of Li- conengprac.2013.12.014
ion batteries for electric vehicles. Int J Energy Res. 2021;45(2): 67. Hossain Lipu MS, Hannan MA, Hussain A, Saad MH,
3113-3128. https://doi.org/10.1002/ER.6005 Ayob A, Uddin MN. Extreme learning machine model for
53. Heinrich F, Noering FKD, Pruckner M, Jonas K. state-of-charge estimation of lithium-ion battery using gravita-
Unsupervised data-preprocessing for long short-term memory tional search algorithm. IEEE Trans Ind Appl. 2019;55(4):
based battery model under electric vehicle operation. J Energy 4225-4234. https://doi.org/10.1109/TIA.2019.2902532
Storage. 2021;38:102598. https://doi.org/10.1016/J.EST.2021. 68. Wu Z, Christofides PD. Economic machine-learning-based
102598 predictive control of nonlinear systems. Mathematics. 2019;
54. Samadi MF, Saif M. State-space modeling and observer design 7(6):1-20. https://doi.org/10.3390/math7060494
of Li-ion batteries using Takagi-Sugeno fuzzy system. IEEE 69. Hasan MM, Ali Pourmousavi S, Jahanbani Ardakani A,
Trans Control Syst Technol. 2017;25(1):301-308. https://doi. Saha TK. A data-driven approach to estimate battery cell tem-
org/10.1109/TCST.2016.2549270 perature using a nonlinear autoregressive exogenous neural
55. Hu JN, Hu JJ, Lin HB, et al. State-of-charge estimation for network model. J Energy Storage. 2020;32:101879. https://doi.
battery management system using optimized support vector org/10.1016/j.est.2020.101879
machine for regression. J Power Sources. 2014;269:682-693. 70. dos Reis G, Strange C, Yadav M, Li S. Lithium-ion battery
https://doi.org/10.1016/j.jpowsour.2014.07.016 data and where to find it. Energy AI. 2021;5:100081. https://
56. Kara A. A data-driven approach based on deep neural net- doi.org/10.1016/j.egyai.2021.100081
works for lithium-ion battery prognostics. Neural Comput 71. Meng J, Stroe DI, Ricco M, Luo G, Teodorescu R. A simplified
Applic. 2021;1–14. https://doi.org/10.1007/s00521-021-05976-x model-based state-of-charge estimation approach for lithium-
57. Chen LR, Hsu RC, Liu CS. A design of a grey-predicted Li-ion ion battery with dynamic linear model. IEEE Trans Ind Elec-
battery charge system. IEEE Trans Ind Electron. 2008;55(10): tron. 2019;66(10):7717-7727. https://doi.org/10.1109/TIE.2018.
3692-3701. https://doi.org/10.1109/TIE.2008.928106 2880668
58. Zheng D, Wang H, An J, Chen J, Pan H, Chen L. Real-time 72. Xia B, Wang H, Tian Y, Wang M, Sun W, Xu Z. State of
estimation of battery state of charge with metabolic grey charge estimation of lithium-ion batteries using an adaptive
model and LabVIEW platform. IEEE Access. 2018;6:13170– cubature Kalman filter. Energies. 2015;8(6):5916-5936. https://
13180. https://doi.org/10.1109/ACCESS.2018.2807805 doi.org/10.3390/en8065916
59. Ben Sassi H, Errahimi F, Es-Sbai N, Alaoui C. Comparative 73. Yang G, Li J, Fu Z, Guo L. Adaptive state of charge estimation
study of ANN/KF for on-board SOC estimation for vehicular of Lithium-ion battery based on battery capacity degradation
applications. J Energy Storage. 2019;25:100822. https://doi. model. Energy Procedia. 2018;152:514-519. https://doi.org/10.
org/10.1016/j.est.2019.100822 1016/j.egypro.2018.09.203
60. Hannan MA, Lipu MSH, Hussain A, Saad MH, Ayob A. Neu- 74. Xu Y, Hu M, Zhou A, et al. State of charge estimation for
ral network approach for estimating state of charge of lithium-ion batteries based on adaptive dual Kalman filter.
lithium-ion battery using backtracking search algorithm. Appl Math Model. 2020;77:1255-1272. https://doi.org/10.1016/
IEEE Access. 2018;6:10069–10079. https://doi.org/10.1109/ j.apm.2019.09.011
ACCESS.2018.2797976 75. Liu G, Xu C, Jiang K, Wang K. State of charge and model
61. Zhang L, Zheng M, Du D, et al. State-of-charge estimation of parameters estimation of liquid metal batteries based on adap-
lithium-ion battery pack based on improved RBF neural net- tive unscented Kalman filter. Energy Procedia. 2019;158:4477-
works. Hindawi. 2020;2020:1-10. 4482. https://doi.org/10.1016/j.egypro.2019.01.765
62. Cui D, Xia B, Zhang R, et al. A novel intelligent method for 76. Yu Q, Xiong R, Lin C, Shen W, Deng J. Lithium-ion battery
the state of charge estimation of lithium-ion batteries using a parameters and state-of-charge joint estimation based on H-
infinity and unscented Kalman filters. IEEE Trans Veh parameter estimation. J Power Sources. 2004;134:277-292.
Technol. 2017;66:8693-8701. https://doi.org/10.1109/TVT. https://doi.org/10.1016/j.jpowsour.2004.02.033
2017.2709326 89. Taborelli C, Onori S. State of charge estimation using
77. Liu X, He Y, Zhang J, Zeng G. A new state-of-charge estima- extended Kalman filters for battery management system.
tion method for electric vehicle lithium-ion batteries based on Paper presented at: 2014 IEEE Int. Electr Veh Conf IEVC
multiple input parameter fitting model. Int J Energy Res. 2017; 2014, 2014. https://doi.org/10.1109/IEVC.2014.7056126
41:1265-1276. https://doi.org/10.1002/er.3705 90. Li Z, Zhang P, Wang Z, Song Q, Rong Y. State of charge esti-
78. Dang X, Yan L, Xu K, Wu X, Jiang H, Sun H. Open-circuit mation for Li-ion battery based on extended Kalman filter.
voltage-based state of charge estimation of Lithium-ion bat- Energy Procedia. 2017;105(4):3515-3520. https://doi.org/10.
tery using dual neural network fusion battery model. Elec- 1016/j.egypro.2017.03.806
trochim Acta. 2016;188:356-366. https://doi.org/10.1016/j. 91. Al-Gabalawy M, Hosny NS, Dawson JA, Omar AI. State of
electacta.2015.12.001 charge estimation of a Li-ion battery based on extended
79. Plett GL. Sigma-point Kalman filtering for battery manage- Kalman filtering and sensor bias. Int J Energy Res. 2020;45:1-
ment systems of LiPB-based HEV battery packs. Part 2: simul- 19. https://doi.org/10.1002/er.6265
taneous state and parameter estimation. J Power Sources. 92. He W, Williard N, Chen C, Pecht M. State of charge estima-
2006;161(2):1369-1384. https://doi.org/10.1016/j.jpowsour. tion for electric vehicle batteries using unscented kalman fil-
2006.06.004 tering. Microelectron Reliab. 2013;53(6):840-847. https://doi.
80. Hannan MA, Lipu MSH, Hussain A, Mohamed A. A review of org/10.1016/j.microrel.2012.11.010
lithium-ion battery state of charge estimation and manage- 93. Santhanagopalan S, White RE. State of charge estimation
ment system in electric vehicle applications: challenges and using an unscented filter for high power lithium ion cells. Int
recommendations. Renew Sust Energ Rev. 2017;78:834-854. J Energy Res. 2016;34:152-163.
https://doi.org/10.1016/j.rser.2017.05.001 94. Zhang F, Liu G, Fang L. Battery state estimation using
81. Xie J, Ma J, Bai K. Enhanced coulomb counting method for unscented kalman filter. Proc - IEEE Int Conf Robot Autom.
state-of-charge estimation of lithium-ion batteries based on 2009;3:1863-1868. https://doi.org/10.1109/ROBOT.2009.5152745
peukert's law and coulombic efficiency. J Power Electron. 95. Peng N, Zhang S, Guo X, Zhang X. Online parameters identi-
2018;18(3):910-922. https://doi.org/10.6113/JPE.2018.18.3.910 fication and state of charge estimation for lithium-ion batte-
82. Leksono E, Haq IN, Iqbal M, Soelami FXN, Merthayasa IGN. ries using improved adaptive dual unscented Kalman filter.
State of charge (SoC) estimation on LiFePO4 battery module Int J Energy Res. 2021;45(1):975-990. https://doi.org/10.1002/
using coulomb counting methods with modified Peukert. Paper ER.6088
presented at: Proceedings of the 2013 Joint International Con- 96. Chen X, Chen X, Chen X. A novel framework for lithium-ion
ference on Rural Information and Communication Technology battery state of charge estimation based on Kalman filter
and Electric-Vehicle Technology, rICT and ICEV-T 2013, 2013. Gaussian process regression. Int J Energy Res. 2021;45(9):
1–5. https://doi.org/10.1109/rICT-ICeVT.2013.6741545 13238–13249. https://doi.org/10.1002/ER.6649
83. Yang A, Wang Y, Yang F, et al. A comprehensive investiga- 97. Lim KC, Bastawrous HA, Duong VH, See KW, Zhang P,
tion of lithium-ion battery degradation performance at differ- Dou SX. Fading Kalman filter-based real-time state of charge
ent discharge rates. J Power Sources. 2019;443:227108. https:// estimation in LiFePO4 battery-powered electric vehicles. Appl
doi.org/10.1016/j.jpowsour.2019.227108 Energy. 2016;169:40-48. https://doi.org/10.1016/j.apenergy.
84. Fathoni G, Widayat SA, Topan PA, Jalil A, Cahyadi AI, 2016.01.096
Wahyunggoro O. Comparison of state-of-charge (SOC) esti- 98. Yang Y, Cui N, Wang C, Liu M, Gao R. SOC estimation of
mation performance based on three popular methods: cou- lithium-ion battery based on new adaptive fading extended
lomb counting, open circuit voltage, and Kalman filter. Paper Kalman filter. Paper presented at: Proc. - 2017 Chinese
presented at: Proc. 2nd Int. Conf. Autom. Cogn. Sci. Opt. Autom. Congr. CAC 2017. 2017, ICCSEE, pp. 5630–5634,
Micro Electro-Mechanical Syst. Inf. Technol. ICACOMIT 2017. https://doi.org/10.1109/CAC.2017.8243786
2017, 2018. 70–74, 2017. https://doi.org/10.1109/ICACOMIT. 99. Peng J, Luo J, He H, Lu B. An improved state of charge esti-
2017.8253389 mation method based on cubature Kalman filter for lithium-
85. Premkumar M, Mohan Kumar R, Karthick K, Sowmya R. SoC ion batteries. Appl Energy. 2019;253:113520. https://doi.org/
estimation and monitoring of li-ion cell using kalman-filter 10.1016/j.apenergy.2019.113520
algorithm. Indones J Electr Eng Inform. 2018;6(4):418-427. 100. Cui X, He Z, Li E, Cheng A, Luo M, Guo Y. State-of-charge
https://doi.org/10.11591/ijeei.v6i1.548 estimation of power lithium-ion batteries based on an embed-
86. Asghar F, Talha M, Kim SH, Ra IH. Simulation study on bat- ded micro control unit using a square root cubature Kalman
tery state of charge estimation using Kalman filter. J Adv filter at various ambient. Int J Energy Res. 2019;43(8):1-51.
Comput Intell Intell Inform. 2016;20(6):861-866. https://doi. https://doi.org/10.1002/er.4503
org/10.20965/jaciii.2016.p0861 101. Shen J, Xiong J, Shu X, et al. State of charge estimation frame-
87. Wang W, Mu J. State of charge estimation for lithium-ion bat- work for lithium-ion batteries based on square root cubature
tery in electric vehicle based on Kalman filter considering Kalman filter under wide operation temperature range. Int J
model error. IEEE Access. 2019;7:29223–29235. https://doi. Energy Res. 2021;45(4):5586-5601. https://doi.org/10.1002/ER.
org/10.1109/ACCESS.2019.2895377 6186
88. Plett GL. Extended Kalman filtering for battery management 102. Xiong R, Yu Q, Wang LY, Lin C. A novel method to obtain
systems of LiPB-based HEV battery packs - part 3. State and the open circuit voltage for the state of charge of lithium ion
batteries in electric vehicles by using H infinity filter. Appl 115. Hu X, Li SE, Yang Y. Advanced machine learning approach
Energy. 2017;207:346-353. https://doi.org/10.1016/j.apenergy. for lithium-ion battery state estimation in electric vehicles.
2017.05.136 IEEE Trans Transp Electrif. 2016;2(2):140-149. https://doi.org/
103. Xue L, Jiuchun J, Caiping Z, Weige Z, Bingxiang S. Effects 10.1109/TTE.2015.2512237
analysis of model parameters uncertainties on battery SOC 116. Shen WX, Chan CC, Lo EWC, Chau KT. Adaptive neuro-
estimation using H-infinity observer. Paper presented at: IEEE fuzzy modeling of battery residual capacity for electric vehi-
International Symposium on Industrial Electronics, 2014. cles. IEEE Trans Ind Electron. 2002;49(3):677-684. https://doi.
https://doi.org/10.1109/ISIE.2014.6864862 org/10.1109/TIE.2002.1005395
104. Pang H, Guo L, Wu L, Jin X. An enhanced temperature- 117. Singh KV, Bansal HO, Singh D. Hardware-in-the-loop imple-
dependent model and state-of-charge estimation for a Li-ion mentation of ANFIS based adaptive SoC estimation of
battery using extended Kalman filter. Int J Energy Res. 2020; Lithium-ion battery for hybrid vehicle applications. J Energy
44:1-14. https://doi.org/10.1002/er.5435 Storage. 2020;27:101124. https://doi.org/10.1016/j.est.2019.
105. Yang J, Xia B, Shang Y, Huang W, Mi CC. Adaptive state-of- 101124
charge estimation based on a split battery model for electric 118. Zahid T, Xu K, Li W, Li C, Li H. State of charge estimation for
vehicle applications. IEEE Trans Veh Technol. 2017;66(12): electric vehicle power battery using advanced machine learn-
10889–10898. https://doi.org/10.1109/TVT.2017.2728806 ing algorithm under diversified drive cycles. Energy. 2018;162:
106. Meng J, Luo G, Gao F. Lithium polymer battery state-of- 871-882. https://doi.org/10.1016/j.energy.2018.08.071
charge estimation based on adaptive unscented Kalman 119. Charkhgard M, Farrokhi M. State-of-charge estimation for
Filter and support vector machine. IEEE Trans Power Elec- lithium-ion batteries using neural networks and EKF. IEEE
tron. 2016;31:2226-2238. https://doi.org/10.1109/TPEL.2015. Trans Ind Electron. 2010;57(12):4178-4187. https://doi.org/10.
2439578 1109/TIE.2010.2043035
107. Peng S, Chen C, Shi H, Yao Z. State of charge estimation of 120. Zou Y, Hu X, Ma H, Li SE. Combined state of charge and state
battery energy storage systems based on adaptive unscented of health estimation over lithium-ion battery cell cycle
Kalman filter with a noise statistics estimator. IEEE Access. lifespan for electric vehicles. J Power Sources. 2015;273:793-
2017;5:13202–13212. https://doi.org/10.1109/ACCESS.2017. 803. https://doi.org/10.1016/j.jpowsour.2014.09.146
2725301 121. Song Y, Liu D, Liao H, Peng Y. A hybrid statistical data-driven
108. Peng S, Zhu X, Xing Y, Shi H, Cai X, Pecht M. An adaptive method for on-line joint state estimation of lithium-ion batte-
state of charge estimation approach for lithium-ion series- ries. Appl Energy. 2020;261:114408. https://doi.org/10.1016/j.
connected battery system. J Power Sources. 2018;392(1):48-59. apenergy.2019.114408
https://doi.org/10.1016/j.jpowsour.2018.04.101 122. Li J, Ye M, Gao K, Xu X, Wei M, Jiao S. Joint estimation of
109. Dineva A, Csom os B, Kocsis S, Vajda I. Investigation of the state of charge and state of health for lithium-ion battery
performance of direct forecasting strategy using machine based on dual adaptive extended Kalman filter. Int J Energy
learning in state-of-charge prediction of Li-ion batteries Res. 2021;45(9):13307–13322. https://doi.org/10.1002/ER.6658
exposed to dynamic loads. J Energy Storage. 2021;36:1-16. 123. Xie J. Available power prediction limited by multiple con-
110. Ng M-F, Zhao J, Yan Q, Conduit GJ, Seh ZW. Predicting the straints for LiFePO 4 batteries based on central difference
state of charge and health of batteries using data-driven Kalman filter. Int J Energy Res. 2018;42:4730-4745. https://doi.
machine learning. Nat Mach Intell. 2020;2(3):161-170. https:// org/10.1002/er.4227
doi.org/10.1038/s42256-020-0156-7 124. Zhang Y, Xiong R, He H, Shen W. Lithium-ion battery pack
111. Singh P, Vinjamuri R, Wang X, Reisner D. Design and imple- state of charge and state of energy estimation algorithms
mentation of a fuzzy logic-based state-of-charge meter for Li- using a hardware-in-the-loop validation. IEEE Trans Power
ion batteries used in portable defibrillators. J Power Sources. Electron. 2017;32:4421-4431. https://doi.org/10.1109/TPEL.
2006;162:829-836. https://doi.org/10.1016/j.jpowsour.2005. 2016.2603229
04.039 125. Xie J, Ma J, Chen J. Available power prediction limited by
112. Avadhanula AK, Kulkarni SS. Comparative study of mathe- multiple constraints for LiFePO4 batteries based on central
matical models and data driven models for battery perfor- difference Kalman filter. Int J Energy Res. 2018;42(15):4730-
mance parameter estimation. Paper presented at: Proc. 2020 4745. https://doi.org/10.1002/ER.4227
3rd Int. Conf. Adv. Electron. Comput. Commun. ICAECC 126. Hu X, Xiong R, Egardt B. Model-based dynamic power assess-
2020, pp. 4–8, 2020. https://doi.org/10.1109/ICAECC50550. ment of lithium-ion batteries considering different operating
2020.9339480 conditions. IEEE Trans Ind Inform. 2014;10(3):1948-1959.
113. Lyu C, Han Y, Guo Q, Wang L, Song Y. State-of-Charge Esti- https://doi.org/10.1109/TII.2013.2284713
mation of Lithium-ion Batteries Based on Deep Neural Net- 127. Xiong R, Sun F, He H, Nguyen TD. A data-driven adaptive
work. Paper presented at: 2020 Global reliability and state of charge and power capability joint estimator of
prognostics and health management, PHM-Shanghai 2020, lithium-ion polymer battery used in electric vehicles. Energy.
2020. https://doi.org/10.1109/PHM-Shanghai49105.2020. 2013;63:295-308. https://doi.org/10.1016/j.energy.2013.10.027
9280940 128. Verma MKS, Basu S, Patil RS, et al. On-board state estimation
114. Alvarez Anton JC, Garcia Nieto PJ, Blanco Viejo C, Vilan in electrical vehicles: achieving accuracy and computational
Vilan JA. Support vector machines used to estimate the bat- efficiency through an electrochemical model. IEEE Trans Veh
tery state of charge. IEEE Trans Power Electron. 2013;28(12): Technol. 2020;69(3):2563-2575. https://doi.org/10.1109/TVT.
5919-5926. https://doi.org/10.1109/TPEL.2013.2243918 2020.2966266
129. Weng C, Sun J, Peng H. An open-circuit-voltage model of 143. Lee CH, Chen MY, Hsu SH, Jiang JA. Implementation of an
lithium-ion batteries for effective incremental capacity analy- SOC-based four-stage constant current charger for Li-ion bat-
sis. Paper presented at: ASME 2013 Dyn. Syst. Control Conf. teries. J Energy Storage. 2018;18:528-537. https://doi.org/10.
DSCC 2013, 1, pp. 1–8, 2013. https://doi.org/10.1115/ 1016/j.est.2018.06.010
DSCC2013-3979 144. Liu YH, Teng JH, Lin YC. Search for an optimal rapid charg-
130. Barai A, Widanage WD, Marco J, Mcgordon A, Jennings P. A ing pattern for lithium-ion batteries using ant colony system
study of the open circuit voltage characterization technique algorithm. IEEE Trans Ind Electron. 2005;52(5):1328-1336.
and hysteresis assessment of lithium-ion cells. J Power https://doi.org/10.1109/TIE.2005.855670
Sources. 2015;295:99-107. https://doi.org/10.1016/j.jpowsour. 145. Min H, Sun W, Li X, et al. Research on the optimal charging
2015.06.140 strategy for Li-ion batteries based on multi-objective optimization.
131. Zheng F, Xing Y, Jiang J, Sun B, Kim J, Pecht M. Influence of Energies. 2017;10(709):1-15. https://doi.org/10.3390/en10050709
different open circuit voltage tests on state of charge online 146. Sun J, Ma Q, Liu R, Wang T, Tang C. A novel multiobjective
estimation for lithium-ion batteries. Appl Energy. 2016;183: charging optimization method of power lithium-ion batteries
513-525. https://doi.org/10.1016/j.apenergy.2016.09.010 based on charging time and temperature rise. Int J Energy Res.
132. Dong G, Wei J, Zhang C, Chen Z. Online state of charge esti- 2019;43(13):7672-7681. https://doi.org/10.1002/ER.4727
mation and open circuit voltage hysteresis modeling of 147. Li Y, Li K, Xie Y, Liu J, Fu C, Liu B. Optimized charging of
LiFePO4 battery using invariant imbedding method. Appl lithium-ion battery for electric vehicles: adaptive multistage
Energy. 2016;162:163-171. https://doi.org/10.1016/j.apenergy. constant current–constant voltage charging strategy. Renew
2015.10.092 Energy. 2020;146:2688-2699. https://doi.org/10.1016/j.renene.
133. Feng F, Lu R, Zhu C. A combined state of charge estimation 2019.08.077
method for lithium-ion batteries used in a wide ambient tem- 148. Zhang C, Jiang J, Gao Y, Zhang W, Liu Q, Hu X. Charging
perature range. Energies. 2014;7:3004-3032. https://doi.org/10. optimization in lithium-ion batteries based on temperature
3390/en7053004 rise and charge time. Appl Energy. 2017;194:569-577. https://
134. Xie S, Xiong R, Zhang Y, He H. The estimation of state of doi.org/10.1016/j.apenergy.2016.10.059
charge for power battery packs used in hybrid electric vehicle. 149. Li Y, Li K, Xie Y, et al. Optimization of charging strategy for
Energy Procedia. 2017;105:2678-2683. https://doi.org/10.1016/ lithium-ion battery packs based on complete battery pack
j.egypro.2017.03.774 model. J Energy Storage. 2021;37:102466. https://doi.org/10.
135. Roscher MA, Bohlen OS, Sauer DU. Reliable state estimation 1016/J.EST.2021.102466
of multicell Lithium-ion battery systems. IEEE Trans Energy 150. Wu X, Shi W, Du J. Multi-objective optimal charging method
Convers. 2011;26(3):737-743. https://doi.org/10.1109/TEC. for lithium-ion batteries. Energies. 2017;10(9):1271. https://
2011.2155657 doi.org/10.3390/en10091271
136. Sun ZD, Wang S, Qin Y, Liu JG, Yan CW. Research of pulse 151. Liu K, Zou C, Li K, Wik T. Charging pattern optimization for
current charge method for all vanadium redox battery. Adv lithium-ion batteries with an electrothermal-aging model.
Mater Res. 2014;971–973:1121-1124. https://doi.org/10.4028/ IEEE Trans Ind Inform. 2018;14(12):5463-5474. https://doi.
www.scientific.net/AMR.971-973.1121 org/10.1109/TII.2018.2866493
137. Ye M, Gong H, Xiong R, Mu H. Research on the battery 152. Liu J, Duan Q, Chen H, Sun J, Wang Q. An optimal multi-
charging strategy with charging and temperature rising con- stage charge strategy for commercial lithium ion batteries.
trol awareness. IEEE Access. 2018;6:64193–64201. https://doi. Sustain Energy Fuels. 2018;2(8):1726-1736. https://doi.org/10.
org/10.1109/ACCESS.2018.2876359 1039/c8se00235e
138. Pandžic H, Bobanac V. An accurate charging model of battery 153. Wu X, Hu C, Du J, Sun J. Multistage CC-CV charge method
energy storage. IEEE Trans Power Syst. 2019;34(2):1416-1426. for Li-ion battery. Math Probl Eng. 2015;2015:1-10. https://doi.
https://doi.org/10.1109/TPWRS.2018.2876466 org/10.1155/2015/294793
139. Liu K, Li K, Yang Z, Zhang C, Deng J. Battery optimal charg- 154. Liu Y, Xu M, Xu Z, Wang X. A study of fast charging of Li-ion
ing strategy based on a coupled thermoelectric model. Paper battery with pulsed current. Paper presented at: ASME Inter-
presented at: 2016 IEEE Congress on Evolutionary Computa- national Mechanical Engineering Congress and Exposition,
tion, CEC 2016, 2016. https://doi.org/10.1109/CEC.2016. Proceedings (IMECE), 2019. https://doi.org/10.1115/
7748334 IMECE2019-10375
140. Liu K, Li K, Yang Z, Zhang C, Deng J. An advanced Lithium- 155. Fang H, Depcik C, Lvovich V. Optimal pulse-modulated
ion battery optimal charging strategy based on a coupled ther- Lithium-ion battery charging: algorithms and simulation.
moelectric model. Electrochim Acta. 2017;225:330-344. https:// J Energy Storage. 2018;15:359-367. https://doi.org/10.1016/j.
doi.org/10.1016/j.electacta.2016.12.129 est.2017.11.007
141. Liu K, Li K, Ma H, Zhang J, Peng Q. Multi-objective optimiza- 156. Wang S-C, Chen G-J, Liu Y-H. Adaptive charging strategy
tion of charging patterns for lithium-ion battery management. with temperature rise mitigation and cycle life extension for
Energy Convers Manag. 2018;159:151-162. https://doi.org/10. Li-ion batteries. CPSS Trans Power Electron Appl. 2018;3(3):
1016/j.enconman.2017.12.092 202-212. https://doi.org/10.24295/cpsstpea.2018.00020
142. Chen Z, Shu X, Xiao R, Yan W, Liu Y, Shen J. Optimal charg- 157. Aliev RA, Aliev RA, Guirimov BG, Uyar K. Recurrent fuzzy
ing strategy design for lithium-ion batteries considering mini- neural network based system for battery charging. Paper pres-
mization of temperature rise and energy loss. Int J Energy Res. ented at: International Symposium on Neural Networks, 2007,
2019;43(9):4344-4358. https://doi.org/10.1002/ER.4560 pp. 307–316.
158. Petchjatuporn P, Khaehintung N, Sunat K, Sirisuk P, 164. Patnaik L, Praneeth AVJS, Williamson SS. A closed-loop
Kiranon W. Implementation of GA-trained GRNN for intelli- constant-temperature constant-voltage charging technique to
gent fast charger for Ni-cd batteries. Paper presented at: Conf. reduce charge time of lithium-ion batteries. IEEE Trans Ind
Proc. - IPEMC 2006 CES/IEEE 5th Int. Power Electron Electron. 2019;66(2):1059-1067. https://doi.org/10.1109/TIE.
Motion Control Conf, 1, pp. 137–141, 2007. https://doi.org/10. 2018.2833038
1109/IPEMC.2006.297060 165. Attia PM, Grover A, Jin N, et al. Closed-loop optimization of
159. Tian J, Xiong R, Shen W, Lu J, Yang XG. Deep neural net- fast-charging protocols for batteries with machine learning.
work battery charging curve prediction using 30 points col- Nature. 2020;578:397-402. https://doi.org/10.1038/s41586-020-
lected in 10 min. Joule. 2021;5:1521-1534. https://doi.org/10. 1994-5
1016/j.joule.2021.05.012 166. Shen P, Ouyang M, Lu L, Li J, Feng X. The co-estimation of
160. Ullah Z, Burford B, Rahman S, Corporation NS, Clara S, state of charge, state of health, and state of function for
Fuuy N. Fast intelligent battery charging: neural-fuzzy lithium-ion batteries in electric vehicles. IEEE Trans Veh
approach. IEEE Aerosp Electron Syst Mag. 1996;11(6):26-34. Technol. 2018;67(1):92-103. https://doi.org/10.1109/TVT.2017.
https://doi.org/10.1109/62.500207 2751613
161. Liu CL, Chiu YS, Liu YH, Ho YH, Huang SS. Optimization of 167. Baure G, Dubarry M. Synthetic vs. real driving cycles: a com-
a fuzzy-logic-control-based five-stage battery charger using a parison of electric vehicle battery degradation. Batteries. 2019;
fuzzy-based taguchi method. Energies. 2013;6(7):3528-3547. 5(2):42. https://doi.org/10.3390/batteries5020042
https://doi.org/10.3390/en6073528
162. Park S, Pozzi A, Whitmeyer M, et al. Reinforcement learning-
based fast charging control strategy for li-ion batteries. Paper
presented at: CCTA 2020 - 4th IEEE Conf. Control Technol. How to cite this article: Adaikkappan M,
Appl. pp. 100–107, 2020. https://doi.org/10.1109/CCTA41146.
Sathiyamoorthy N. Modeling, state of charge
2020.9206314
estimation, and charging of lithium-ion battery in
163. Suresh R, Rengaswamy R. Modeling and control of battery
systems. Part II: a model predictive controller for optimal electric vehicle: A review. Int J Energy Res. 2022;46
charging. Comput Chem Eng. 2018;119:326-335. https://doi. (3):2141-2165. doi:10.1002/er.7339
org/10.1016/j.compchemeng.2018.08.017

Modeling, State of Charge Estimation, and Charging of Lithium-Ion Battery in Electric Vehicle: A Review

Uploaded by

Copyright:

Available Formats

Modeling, State of Charge Estimation, and Charging of Lithium-Ion Battery in Electric Vehicle: A Review

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Modeling, State of Charge Estimation, and Charging of Lithium-Ion Battery in Electric Vehicle: A Review

Uploaded by

Copyright:

Available Formats

1099114x, 2022, 3, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/er.7339 by Nanyang Technological Uni, Wiley Online Library on [16/02/2023].

Modeling, state of charge estimation, and charging of

Maheshwari Adaikkappan | Nageswari Sathiyamoorthy

Department of Electrical and Electronics

1 | INTRODUCTION battery and recharged by the plugging vehicle. Among

a data acquisition system. Finally, the system controller

Equivalent Circuit Model

Battery Management System

Direct (Ampere hour, OCV method)

Model based (Adaptive filter methods,

State Estimation Charge and

Constant Current Charging

F I G U R E 1 Overview of the software and hardware

TABLE 1 Comparison of various types of Electrochemical models of battery

TABLE 2 RC network based Equivalent Circuit models

Circuit diagram Model equations Model parameters

Two RC network: V Bat ¼ OCVðSoCÞ I Bat R0 V 1 V 2 OCV,R0 , R1 ,C 1 ,R2 ,C2

Three RC network: V Bat ¼ OCVðSoCÞ I Bat R0 V 1 V 2 V 3 OCV ,R0 , R1 , C1 , R2 ,

TABLE 3 Comparison of data-driven model of battery

TABLE 4 Combination of models and identification algorithms used in literature

TABLE 5 Comparison of various battery models

2 describes the computational

(ECM) in real-time characteristics of

4-Very high • Low accuracy

TABLE 6 Summary of the SoC estimation at different combinations

Authors Algorithm execution Remarks Accuracy Complexity Technical challenges

Authors Algorithm execution Remarks Accuracy Complexity Technical challenges

Authors Algorithm execution Remarks Accuracy Complexity Technical challenges

Authors Algorithm execution Remarks Accuracy Complexity Technical challenges

TABLE 7 Keynotes of different charging methodologies

Core idea Advantages Disadvantages

TABLE 8 Challenges for recommended work

Recommended work Remarks Challenges

complexity. Careful attention is needed to develop an J Energy Res. 2020;44(4):2535-2548. https://doi.org/10.1002/er.

You might also like