1 s2.0 S095965262100264X Main

Journal of Cleaner Production 292 (2021) 126044
Contents lists available at ScienceDirect
Journal of Cleaner Production

journal homepage: www.elsevier.com/locate/jclepro
Review
Intelligent algorithms and control strategies for battery management

system in electric vehicles: Progress, challenges and future outlook
M.S. Hossain Lipu a, *, M.A. Hannan b, Tahia F. Karim c, Aini Hussain a,
Mohamad Hanif Md Saad d, Afida Ayob a, Md. Sazal Miah e, T.M. Indra Mahlia f
a
Department of Electrical, Electronic and Systems Engineering, Universiti Kebangsaan Malaysia, Bangi 43600, Malaysia
b
Department of Electrical Power Engineering, College of Engineering, Universiti Tenaga Nasional, Kajang 43000, Malaysia
c
Department of Electrical and Electronic Engineering, American International University-Bangladesh, Dhaka 1229, Bangladesh
d
Institute of IR 4.0, Universiti Kebangsaan Malaysia, Bangi 43600, Malaysia
e
School of Engineering and Technology, Asian Institute of Technology, Pathumthani 12120, Thailand
f
School of Information, Systems and Modelling, University of Technology, Sydney, NSW 2007, Australia
a r t i c l e i n f o a b s t r a c t
Article history: Globally, the research on battery technology in electric vehicle applications is advancing tremendously to
Received 24 September 2020 address the carbon emissions and global warming issues. The effectiveness of electric vehicles depends
Received in revised form on the accurate assessment of key parameters as well as proper functionality and diagnosis of the battery
17 December 2020
storage system. However, poor monitoring and safety strategies of the battery storage system can lead to
Accepted 15 January 2021
Available online 20 January 2021
critical issues such as battery overcharging, over-discharging, overheating, cell unbalancing, thermal
runaway, and fire hazards. To address these concerns, an effective battery management system plays a
Handling editor: Bin Chen crucial role in enhancing battery performance including precise monitoring, charging-discharging con-
trol, heat management, battery safety, and protection. The goal of this paper is to deliver a compre-
Keywords: hensive review of different intelligent approaches and control schemes of the battery management
Battery management system system in electric vehicle applications. In line with that, the review evaluates the intelligent algorithms in
Lithium-ion batteries battery state estimation concerning their features, structure, configuration, accuracy, advantages, and
Electric vehicle disadvantages. Moreover, the review explores the various controllers in battery heating, cooling,
Thermal management
equalization, and protection highlighting categories, characteristics, targets, achievements, benefits, and
Fault diagnosis
shortcomings. The key issues and challenges in terms of computation complexity, execution problems
Battery equalization
along with various internal and external factors are identified. Finally, future opportunities and directions
are delivered to design an efficient intelligent algorithm and controller toward the development of an
advanced battery management system for future sustainable electric vehicle applications.
© 2021 Elsevier Ltd. All rights reserved.
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Reviewing methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1. Selection process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2. Results of the review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3. Progress of intelligent algorithms and control schemes for BMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.1. Battery state estimation in BMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.1.1. State of charge estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.1.2. State of health prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.1.3. Remaining useful life prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2. Battery equalization in BMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2.1. Rule-based algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
* Corresponding author.
E-mail address: [email protected] (M.S. Hossain Lipu).
https://doi.org/10.1016/j.jclepro.2021.126044
0959-6526/© 2021 Elsevier Ltd. All rights reserved.
M.S. Hossain Lipu, M.A. Hannan, T.F. Karim et al. Journal of Cleaner Production 292 (2021) 126044
3.2.2. Optimization-based algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3.3. Battery fault diagnosis in BMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3.1. Rule-based algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3.2. Machine learning-based algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3.3. Mathematical data-driven algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3.4. Entropy algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3.5. Other algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.4. Thermal management system in BMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.4.1. Controller schemes of TMS based on cooling and heating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4. Key issues and challenges of BMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.1. Algorithms issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2. Implementation issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2.1. Data abundance and data variety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2.2. Data integrity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2.3. Suitable structure and hyperparameter adjustment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2.4. Optimization integration issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2.5. Algorithm verification uncertainties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.3. Battery related issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5. Future outlook and directions for BMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.1. Development of advanced sensing equipment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.2. State monitoring for the battery pack in EVs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.3. Algorithm hybridization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.4. Training performance enhancement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.5. Real-time embedded prototype design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.6. Integrated with big data platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.7. Multiscale and co-estimation process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Declaration of competing interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1. Introduction temperature, and identify the fault (Lin et al., 2019). An efficient
BMS has the following key responsibilities: (i) estimates and eval-
Nowadays, the automotive industry has made great strides due uates the battery states accurately including state of charge (SOC),
to the various technological advancements toward the improve- state of energy (SOE), state of health (SOH) and remaining useful
ment of both passengers and pedestrians’ safety (Gohoungodji life (RUL), (ii) controls the battery temperatures within the safe
et al., 2020). However, the greater number of vehicles on the limit, (iii) operates fault diagnosis, fault prognosis, and fault
roads is responsible to increase the pollution levels dramatically in handling and (iv) balances the voltage, charge, and capacity among
urban environments (Guo et al., 2020). The European Union has battery cells (Hussain et al., 2019). A schematic diagram of an
reported that the transport sector is accountable for approximately efficient BMS for EV operation is shown in Fig. 1.
27% of the total carbon dioxide (CO2) emissions, while the vehicle BMS in EVs includes many circuits, components, power elec-
transport is responsible for over 70% of the emissions (EEA, 2017). tronics, sensors, actuators, diode, capacitor, inductor, transformer,
To tackle these problems, electric vehicles (EVs) have received switching, converter, and safety equipment which are controlled by
massive attention and popularity around the world due to their many algorithms, models, and control signals (Lelie et al., 2018). A
capability to decrease environmental pollution, reserve fossil fuels, great deal of research has been performed on developing the
minimize carbon emission and global warming issues (Lin et al., appropriate algorithms in BMS (Zhang et al., 2018b). The most
2020). EVs are the promising alternative to diesel-based vehicles common approaches employed in BMS include model-based
not only at emissions levels but also in terms of simplicity, reli- methods and intelligent methods (Lipu et al., 2020a). The intelli-
ability, comfort and efficiency (Zheng et al., 2020). However, wide gent algorithms are suitable for lithium-ion batteries to address
adoption of EV requires proper functionalities and diagnosis of the complex, dynamic, and nonlinear characteristics (Zhao et al., 2020).
battery storage system (BSS) in terms of battery cell monitoring,
charge-discharge control, cell balancing, heat management, and
power management control (Kumar and Alok, 2020).
Currently, lithium-ion batteries are dominant in the EV battery
market due to their high power and energy density, high voltage,
extended life cycles and low self-discharge rates (Nikolian et al.,
2016). Nevertheless, lithium batteries are sensitive to aging and
temperature; thus, special focus is required on their working en-
vironments to avoid any physical damage, aging, and thermal
runaways (Aaldering et al., 2019) (Xiong et al., 2020b). The battery
management system (BMS) in EV operation is necessary to monitor
battery current, voltage, temperature; examine battery charge,
energy, health, equalize the voltage among cells, control
Fig. 1. Schematic diagram of BMS for EV applications.
2
Besides, intelligent algorithms demonstrate enhanced learning new contributions with a detailed explanation of recent intelligent
capability, fast convergence speed, improved generalization and algorithms and control strategies for BMS in EVs. The main findings
high accuracy (Xiong et al., 2018b). Moreover, intelligent algorithms of this review are summarized below.
do not require explicit information and knowledge about battery
internal characteristics, chemical reactions and working principles This review comprehensively examines the various intelligent
(Z. Huang et al., 2019b). Additionally, intelligent algorithms can approaches toward SOC, SOE, SOH and RUL estimation in BMS.
operate without an added filter, mathematical model, and can The intelligent algorithms are classified according to feed-
update the network parameters by self-learning algorithm which is forward algorithms, time-series based learning, hybrid optimi-
ideal for battery state estimation (Tian et al., 2020). In line with zation algorithms, and statistical algorithms. In line with that,
that, the various optimization techniques are integrated with their structure, configuration, features, estimated error, benefits,
intelligent algorithms in BMS to improve accuracy, efficiency, weaknesses and research gaps are provided.
adaptability, and robustness (He et al., 2020; Merhy et al., 2020; The role of various controllers in the battery pack equalization,
Tran et al., 2020). An efficient controller of BMS ensures battery diagnosis of faults and thermal management are outlined
safety as well as protects the battery pack from hazard conditions highlighting their types, characteristics, target, contributions,
(Ringbeck et al., 2020). Besides, the controller maintains the battery strength, and weakness.
cooling and heating temperature within a safe limit (Hannan et al., The various existing issues and challenges with respect to al-
2019). Moreover, the controller in BMS helps to equalize the gorithm complexity, implementation issues as well as data va-
imbalance of battery packs (Zun et al., 2020). riety, data integrity, structure, uncertainties are discussed
To date, many BMS related articles have been reported in the rigorously.
literature. Shen and Gao (2019) conducted a survey on model-based Finally, the review offers useful suggestions for future oppor-
BMS highlighting state estimation, temperature control, energy tunities and developments of BMS.
equalization and fault diagnosis. Although the authors listed a large
pool of articles related to model-based approaches, they did not The rest of the paper is structured into six sections. Section 2
provide the classification of methods and outline the intelligent explains the reviewing methodology. Section 3 covers the
algorithms and control strategies in BMS. Lu et al. (2013) focused on comprehensive exploration of intelligent algorithms in battery
the key issues of BMS for lithium-ion batteries in EV applications. state estimation. Besides, the various controller schemes in battery
The authors examined the methods of SOC, SOH, battery equal- equalization, fault diagnosis and thermal management are out-
ization and faults. However, the explanation was limited to only a lined. The section 4 narrates the existing research gaps, key issues
few intelligent approaches. Lelie et al. (2018) reviewed the hard- and challenges. The prospects and recommendations for BMS are
ware features of BMS in EVs concerning battery pack topology and delivered in section 5. The concluding remarks are depicted in
associated safety, reliability and complexity. Nonetheless, the au- section 6.
thors did not cover intelligent methods and control techniques of
BMS. Balasingam et al. (2020) explored the various parameters of
2. Reviewing methodology
BMS including SOC, SOH, RUL, power fade, capacity, impedance and
optimal charging based on model-based approaches. Nevertheless,
This review process was performed based on content analysis.
intelligent algorithms and control schemes were not investigated in
The exploration of the relevant literature was carried out using the
detail. Wang et al. (2020b) conducted a systematic review on BMS
Scopus databases. The proper references were collected and cited
highlighting SOC, SOH and RUL estimation as well as battery
accordingly from Google Scholar, Scopus and Web of Science plat-
modeling including data-driven models, electrochemical models
forms. The related articles are searched using the important key-
and equivalent circuit models. However, the authors did not outline
words within the scope such as battery management system,
the controller scheme in thermal management, fault diagnosis and
lithium-ion batteries, electric vehicle, state estimation, thermal man-
battery equalization. Chaturvedi et al. (2010) developed the
agement, fault diagnosis, battery equalization. The authors have
framework for lithium-ion battery models. Nonetheless, the au-
found several journals and conference papers: however, appro-
thors did not discuss the state estimation algorithms and the role of
priate articles are chosen by assessing the paper title, abstract,
controller schemes in BMS. Xiong (2020) provided a detailed
outlines and contributions. After that, the list of articles was final-
description of model-based SOC, SOH and state of power (SOP)
ized using journal impact factor, citation and review process.
estimation of BMS. Nevertheless, the intelligent algorithms and
The findings of this reviewing approach were structured into
controller schemes in BMS were not reported. Xiong et al. (2020a)
four parts. Firstly, the intelligent approaches in battery state esti-
discussed the lithium-ion battery safety protection concerning
mation were studied comprehensively. Secondly, the role of con-
diagnosing and prognosticating short circuit in BMS. However,
trollers in battery equalization, fault diagnosis and thermal
battery state estimation, thermal management, battery equaliza-
management were reviewed. Thirdly, the review highlighted the
tion were not studied. Although Hussain et al. (2019) and Xiong
key limitations and challenges of different intelligent algorithms
et al., (2018b) presented a detailed classification of battery SOC
and controllers of BMS. Fourthly, the review suggested some se-
and SOH in BMS, respectively, but fault diagnosis, temperature
lective proposals for the future advancement of BMS in EVs. The
control, charge equalization strategies in BMS were not covered.
schematic diagram of the reviewing methodology is presented in
Hannan et al. (2018a) carried out a comprehensive study on the
Fig. 2. The method of this review was split into two stages, as
fundamental, structure and performance of lithium-ion battery as
illustrated below.
well as state estimation, temperature and heat management, fault
diagnosis, protection and equalization. However, the intelligent
methods and control techniques in BMS were not described 2.1. Selection process
extensively. Xing et al. (2011) explained the concerns of BMS in EVs
along with battery modeling, state evaluation and cell balancing. A total of 778 articles was identified after the primary screening.
Nevertheless, the authors did not investigate various intelligent The second screening was performed using the proper key-
approaches and control strategies in depth. words, title, abstract, subject, contributions and therefore 423
To bridge the existing research limitations, this survey unveils articles were selected.
3
(a) Screening Method electrochemical reaction, polarization, solid-phase diffusion con-

3. Final selection
taining heavy computation and functional relationships are needed
200 papers were identified and analyzed based on to be studied (D. Huang et al., 2019a). Furthermore, extensive
the impact factor, citation and review process. 6% Journals (181)
8%
2%
research, domain experts and a considerable amount of duration
90.5% Conferences (12)
1. Initial screening and 1.5 %
are required to conduct the experimental tests (Lai et al., 2020). The
Web pages (4)
Assessment
200 Books (3) model-based SOC estimation methods have shorter offline opera-
After initial search, a total
of 778 papers was selected tional time and obtain reasonable solutions if the parameters of
778
423 2. Second screening battery model are assessed accurately (Bi and Choe, 2020). How-
Relevant 423 articles were selected
ever, they have a lengthy and expensive parameterization process
based on keywords, title, abstract,
contents and contributions as well as slow online execution time because of the computa-
tionally intensive computation. Besides, the performance of model-
(b) Review Results
based methods declines under uncertainties in battery model pa-
rameters as well as noise and aging impacts (Li et al., 2020e). Wei
Intelligent algorithms in battery Controller schemes in battery
state estimation equalization, fault diagnosis et al. (2016) suggested a novel method for SOC estimation and
Review of intelligent - Assessment of battery SOC, SOE, and temperature control
algorithms and SOH and RUL - Types and classification. model identification with noise effect based on Frisch scheme-
controllers in BMS - Classification and configuration - Objectives and variables. based bias compensating recursive least squares (FBCRLS). The
- Benefits and shortcomings - Strength and weakness.
- Validation and error rates analysis - Outcomes and limitations. experimental results demonstrated the effectiveness of the pro-
- Contributions and research gaps.
posed approach in reducing the bias noise on model identification
Future directions Discussion of key the issues of intelligent algorithms and
Effective future suggestions for further controllers in BMS
which elevated the SOC estimation accuracy. Wei et al. (2017)
enhancement of BMS toward sustainable Exploration of various limitations such as algorithm and designed a dual estimation and multi-time scale based SOC and
EV expansion. controller issues, optimization integration issues, algorithm
validation uncertainties and hyperparameters adjustment. capacity estimation framework in BMS. The online battery model
parameters were assessed using vector-type recursive least squares
Fig. 2. Schematic diagram of the reviewing methodology. (VRLS). The dual estimation was carried out through the integration
of RLS-based capacity estimator and Kalman filter (KF) based SOC
estimation. The experiments report demonstrated the high accu-
The third screening was carried out based on the impact factor, racy and fast convergence in SOC and capacity estimation in real-
the number of citations and the review process. time. Wei et al. (2019) estimated online SOC and power capacity
Finally, a total of 200 articles published in prominent scientific under the impact of noise corruption. The adaptive forgetting
conference proceedings and notable journals were identified for recursive total least squares (AF-RTLS) was employed to lessen the
analysis and discussion. bias and noise effect in model parameters. AF-RTLS was incorpo-
rated with Luenberger observer (OB) to online compute SOC. The
multi-constraint analytical method together with OB based model
2.2. Results of the review parameters and SOC were used to assess the capacity online. The
simulation and experiments illustrated the superiority of the pro-
The value of the intelligent algorithms toward accurate battery posed approach with regard to accuracy and robustness.
state estimation including SOC, SOH, RUL and SOH was broadly On the other hand, Intelligent algorithms have received huge
surveyed. attention in battery state estimation because they need limited
Various controller schemes in battery equalization, fault diag- knowledge and less development time to build complex battery
nosis and thermal management were explicitly reviewed. systems compared to model-based methods (Y. Song et al., 2020b).
The key issues and challenges of intelligent algorithms and Intelligent algorithms and controller schemes do not require
controllers for BMS were discussed. comprehensive domain knowledge and detailed mathematical
Selective future directions for further improvement of BMS to model rather only requires a large pool of data and high computing
elevate accuracy, adaptability and robustness were provided. power (L. Song et al., 2020a). Also, intelligent methods achieve
outstanding performance if a sufficient amount of data is available
3. Progress of intelligent algorithms and control schemes for and are very efficient under uncertainties including noise, varying
BMS temperatures and aging effects (Singh et al., 2020). Moreover, they
have fast online execution and self-learning operations to execute
This section provides a comprehensive exploration of BMS the parameterization (Yao et al., 2021). Nevertheless, they have
operation including state estimation, battery equalization, fault shortcomings in terms of long training time (Deng et al., 2019).
diagnosis and thermal management. The recent progress of intel-
ligent algorithms and control strategies are highlighted to explain 3.1.1. State of charge estimation
the BMS operation. SOC is defined as how much charge is available inside a battery
cell to drive a vehicle (Lipu et al., 2020b). The precise evaluation of
3.1. Battery state estimation in BMS SOC is very significant in achieving better charging, discharging
profile and prolonged battery lifespan (Shu et al., 2020). Numerous
The accurate evaluation of battery states enhances battery aging articles have been published on SOC estimation for lithium-ion
performance, extends battery life, and confirms a secure and reli- batteries. Based on the literature survey, SOC algorithms generally
able driving of EV. Commonly, battery state estimation approaches have four categories: feed-forward algorithms, regression and
in BMS is classified into two groups; model-based and intelligent probabilistic algorithms, time-series algorithms and hybrid intel-
methods (Xiong et al., 2018a). The model-based state estimation ligent optimization algorithms.
approaches require detailed domain knowledge, a deeper under-
standing of battery and a moderate amount of data to develop 3.1.1.1. Feed-forward neural network algorithm-based SOC estima-
robust rules and mathematical model (Shrivastava et al., 2019). tion. Machine learning-based algorithms can map the battery dy-
Moreover, the comprehensive theoretical knowledge related to namics without the battery model, as well as complex
physics and chemistry such as anode, cathode material features, mathematical rules and relationships. He et al. (2014) proposed the
4
feedforward neural network (FNN) algorithm with the back- introduced FNN to determine SOE for lithium-ion batteries under
propagation learning rule using three input features, and two the current influence and different temperature settings. FNN
hidden layers (HLs) for accurate SOC estimation of LiFePO4 batte- achieved acceptable solutions with RMSE of 1.314% against dy-
ries. FNN obtained maximum SOC error of 1.9% and 3.5% under namic loading conditions. Dong et al. (2015) proposed WNN-based
Federal Urban Driving Cycles (FUDS) and US06 cycle, respectively. SOE estimation method for LiFePO4 batteries. WNN demonstrated
Although FNN method has simple execution and acceptable accu- accurate and robust performance under different discharge current
racy, it needs long data training duration. Kang et al. (2014) con- rates and temperature conditions. The particle filter estimator is
structed an enhanced radial basis function neural network (RBFNN) further applied to eliminate the effects of measurement noises in
model with adjusted center, weighted and activation function. current and voltage measurements. WNN obtained RMSE of 1.89%
RBFNN achieved reasonable results under different aging profiles and 5.93% under constant current and variant current, respectively.
and EV drive cycles with mean absolute error (MAE) under 5%.
RBFNN method can achieve global approximation as well as 3.1.1.2. Regression and probabilistic algorithm-based SOC estimation.
reasonable accuracy with incomplete information; however,
Alvarez n et al. (2013) built a support vector machine (SVM)
Anto
training operation is lengthy and can be trapped in a local mini- model for SOC estimation of LiFeMnPO4 batteries. SVM delivered a
mum. Lipu et al. (2019b) developed an ELM structure with one reasonable solution with respect to accuracy and execution time,
hidden layer and sigmoid activation function. ELM delivered ac- indicating root mean square error (RMSE) of 0.71%. SVM can
curate SOC solutions with low computational cost, reducing MAE address non-linear and high dimension models with less-
by 35.3% and 23.6% compared to RBFNN and FNN methods, overfitting issues, however, it has complex computation and
respectively under different drive cycles. Although ELM method time-consuming process. Li et al. (2014a) evaluated random forest
exhibits better generalization, scalability and fast computation (RF) based SOC estimation accuracy under battery discharge cur-
process, the suitable number of HLs and hidden neurons (HNs) is rents. RF obtained better results than FNN under steady-state and
required to achieve better results. Cui et al. (2018) introduced a dynamic state, reporting RMSE of 0.59% and 0.14%, respectively. RF
wavelet neural network (WNN) to monitor SOC for EV batteries. is accurate and robust to over-fitting and noise impacts; never-
WNN obtained better accuracy and computation time than FNN theless, its performance deviates under irregular and dynamic
and EKF algorithms, estimating MAE and maximum SOC error of systems. Sahinoglu et al. (2018) developed Gaussian process
0.59%, and 3.13%, respectively. WNN features easy training execu- regression (GPR) based SOC estimation approach in electric vehicle
tion but it needs several hidden layer units to obtain reasonable batteries. The structure of GPR was enhanced with optimal hyper-
solutions. Chemali et al. (2018) suggested a deep FNN (DFNN) to parameters and kernel function. GPR outperformed FNN and SVM,
compute SOC against Gaussian noise, offset, gains, temperatures, attaining MAE and RMSE below 0.81%, and 0.24% respectively in EV
and diverse EV drive cycles. DFNN achieved the highest SOC error drive cycles. Although GPR has easy interpretability of various
and MAE of 8% and 4% respectively, at varying temperature con- features and is excellent in capturing the model uncertainty; it
ditions. The configurations of feed-forward neural network algo- suffers from low accuracy in high dimensional spaces.
rithms in battery SOC estimation is shown in Fig. 3.
State of energy (SOE) is another assessment index of BMS which 3.1.1.3. Time series algorithm-based SOC estimation.
is computed in adjacent with SOC. SOE is calculated using the re- Time-series based algorithms are computationally rich and very
sidual energy storage capacity divided by the maximum energy efficient since they predict future SOC estimation based on the
storage capacity in a battery cell (Hu et al., 2019). Liu et al. (2014b) present information as well as historical information of input,
hidden, and output states. Chaoui et al. (2017) proposed a time-
series algorithm for SOC estimation for LiFePO4 batteries based
on input time-delay neural network (TDNN). TDNN was validated
under four calendar aging cycles and obtained RMSE below 0.33% at
10 A constant discharge test. The recurrent neural network (RNN)
algorithms are the basic methods for exploring time-series based
algorithms. Chaoui and Ibe-Ekeocha (2017) designed a RNN with
the nonlinear autoregressive network with exogenous inputs
(NARX) architecture to determine SOC for Lithium Iron Phosphate
(LiFP) and Lithium Titanate (LiTO) batteries. The proposed model
delivered high accuracy and low computational cost under
numerous charge-discharge profiles and aging cycles. The NARX is
computationally intelligent in controlling the dynamic and com-
plex system, but the performance depends on the number of his-
torical data points in the input layer, hidden layer, and output layer.
Due to the availability of a huge amount of quality data, large
storage capacity, and powerful computation processor, the deeper
architectures of RNN have become prevalent. However, deep
learning-based SOC estimation is still very limited. Yang et al.
(2019b) introduced a stacked long short-term memory (LSTM)
network to estimate SOC for LiFePO4 batteries. LSTM illustrated low
execution time, and excellent tracking accuracy under erroneous
initial SOC values, achieving MAE and RMSE within 1% and 2%,
respectively. Bian et al. (2020b) suggested a bidirectional LSTM
framework with encoder-decoder to estimate SOC under various
Fig. 3. Structure of feedforward neural network algorithms based SOC estimation (a)
temperature settings. The encoder-decoder supported LSTM algo-
FNN (He et al., 2014), (b) RBFNN (Kang et al., 2014), (c) ELM (Lipu et al., 2019b), and (d) rithm in mapping the long-term dependencies in past and future
GPR (Sahinoglu et al., 2018). resulting in a rise in SOC estimation accuracy. The experimental
5
dataset demonstrated the accurate solutions in SOC estimation, an easy execution process but suffers from slow convergence speed.
indicating MAE of 1.07% at variable temperature conditions. Yang Zhang et al. (2018e) introduced an intelligent method using particle
et al. (2020) introduced LSTM network for battery SOC estimation swarm optimization (PSO) to optimize weights and thresholds
under FUDS and US06 dive cycles under varying temperature value of FNN. The optimized FNN improved SOC precision by 2%
conditions. The results indicated MAE and RMSE 1% and 1.1% points in comparison to conventional FNN under dynamic EV drive
respectively. Though LSTM exhibits powerful prediction ability and cycle. Although PSO features good convergence speed and high
can deliver satisfactory results in dynamic load profiles, it has efficiency, it can converge prematurely. Hannan et al. (2018b) in-
shortcomings of having slow and complex training operations. Xiao tegrated the backtracking search algorithm (BSA) and FNN to
et al. (2019) employed Adam and Ensemble optimizer to build a determine the suitable number of HNs and learning rate. BSA-based
gated recurrent unit (GRU). The authors found that GRU- Ensemble FNN presented better results than conventional FNN, indicating
demonstrated better accuracy and training time than GRU-Adam, MAE and RMSE of 0.48% and 0.81%, respectively under dynamic
estimating MAE and RMSE of 3.28%, 0.84%, and 1.13%, respec- stress test (DST) cycle at 25 C. BSA has a better search exploration
tively. GRU can tackle the gating mechanism problems of LSTM; but process with the mutation and crossover technique; however, it has
it has weaknesses in terms of a huge amount of data training and lengthy computation due to the dual population algorithm. Hannan
large storage device. et al. (2020b) designed an intelligent hybrid optimization model
In recent times, the deep learning approaches are combined to with lighting search algorithm (LSA) based NARX. The accuracy of
build a hybrid algorithm based SOC estimation model. For instance, SOC was improved by optimizing input delays (IDs), feedback de-
Song et al. (2019) merged LSTM and convolutional neural network lays (FDs), and HNs. An extensive comparative analysis with
(CNN) to estimate SOC for EV batteries. CNN was used to extract prominent optimized machine learning algorithms was performed
features while the LSTM was employed to evaluate SOC perfor- and results were excellent with RMSE and MAE below 1%. LSA of-
mance. CNN-LSTM illustrated better performance than alone LSTM fers high exploration and exploitation search capability, neverthe-
and CNN network achieving RMSE and MAE inside 1.5% and 1%, less, it has a complex execution process. Lipu et al. (2019)
respectively. The different structures of time-series based SOC incorporated the gravitational search algorithm (GSA) and ELM to
estimation are depicted in Fig. 4. The mathematical equations of the determine the optimal numbers of the hidden layer (HLs) and HNs.
execution process for SOC estimation using different intelligent GSA-based ELM was outstanding in delivering SOC in a quick period
algorithms are presented in Table 1. under US06 cycle, indicating a computational cost of 0.22 s and
0.27 s at 25 C and 45 C, respectively. Although GSA features fast
learning speed and small training error, it has poor searching
3.1.1.4. Hybrid intelligent optimization algorithm-based SOC estima-
capability. Lipu et al. (2020b) introduced an optimized TDNN
tion. Recently, the integration of heuristic optimization techniques
framework for SOC estimation in lithium-ion batteries based on the
and intelligent algorithms has achieved considerable contributions
firefly algorithm (FA). The improved FA (iFA) was employed to find
toward SOC accuracy enhancement. A hybrid intelligent optimiza-
the optimal hyperparameters including HNs and input time delay.
tion algorithm for SOC estimation can be established through the
The iFA based RNN delivered excellent solutions in achieving a
optimization of hyperparameters which eventually improves SOC
lower RMSE of 0.8512% and 0.7937% in LiNiCoAlO2 and LiNiMn-
estimation accuracy under various operating conditions. For
CoO2 battery, respectively under the Hybrid Pulse Power Charac-
instance, Chang (2013) employed orthogonal least-squares (OLS)
terization (HPPC) test. Although FA is flexible and has an easy
and genetic algorithm (GA) to optimize HNs and center, width
implementation, it has shortcomings of slow convergence. The
respectively of RBFNN. The enhanced RBFNN algorithm was
evaluation of intelligent algorithms based SOC estimation is
dominant to FNN with a maximum absolute error of 1.82%. GA has
denoted in Table 2.
3.1.2. State of health prediction

SOH is defined as the charge capacity of an aged battery cell to
fulfill the operating requirements compared with the charge ca-
pacity of a brand-new battery cell (Li et al., 2020b,d). SOH is crucial
to assess the current health condition of the battery. Various
methods have been published in the literature to predict SOH for
lithium-ion batteries. This study divides SOH methods into four
groups: feed-forward algorithms, regression and probabilistic al-
gorithms, and recurrent neural network algorithms, entropy algo-
rithms, and other algorithms.
3.1.2.1. Feed-forward neural network algorithm-based SOH predic-

tion. You et al. (2016) developed a real-time FNN based SOH esti-
mation method for EV batteries. FNN was validated by diverse EV
drive cycles and attained accurate results, with RMSE and MAE of
1.06% and MAE 3.81% respectively under different temperature
settings. Pan et al. (2018) built a novel SOH framework with a
multiple correlation analysis-based ELM network. ELM performed
better than FNN with MAE and RMSE of 1.09% and 1.72%, respec-
tively. Klass et al. (2014) used on-board available battery data to
predict SOH using SVM where SVM achieved RMSE of 0.63% under
changing temperature conditions.
Fig. 4. The structures of time-series algorithm-based SOC estimation (a) TDNN algo-
rithm (Chaoui et al., 2017), (b) Recurrent NARX algorithm (Chaoui and Ibe-Ekeocha, 3.1.2.2. Regression and probabilistic algorithm-based SOH prediction.
2017), (c) LSTM layer (Yang et al., 2019b), and (d) GRU layer (Xiao et al., 2019). Li et al. (2020b) proposed a data-driven algorithm-based battery
6
Table 1
Mathematical expressions of different intelligent algorithms utilized in SOC estimation.
Refs. Methods Algorithms execution formulas Description
He et al. (2014) FNN 1 netj, is hidden layer output; netk is output layer output; xi is the input vector; wi,j wk,j and qi,j qk,j
f ðnetÞ ¼
1 þ eðnetÞ
X present the weight and bias, respectively; and Oj represents the output of the hidden layer.
8
> netj ¼ wi;j xi þ qi;j
<
j
X
>
: netk ¼ wj;k Oj þ qj;k
k
Kang et al. RBFNN kxðnÞ lm ðnÞk lm and sm denote the center and width of mth HNs in the Gaussian function, respectively; wkm
PM
(2014) yk ¼ s2m ðnÞ denotes weight; yk represents the output in the output layer;
m¼1 wkm e
Lipu et al. ELM PL xj is input weight vector; ai is hidden weight vector; bi is hidden layer bias; Bi is output weight; L is
i¼1 Bi Gðai ; bi ; xj Þ ¼ tj
(2019) HNs; G(x) activation function; tj is the output matrix of the hidden layer; Hþ is the Moore
Gðai ; bi ; xj Þ ¼ ð1 þ eðai ;bi ;xj Þ Þ1
ePenrose generalized inverse of H.
b ¼ Hþ T, b ¼ ½b1 ; :::; bL T ,T ¼ ½t1 ; :::; tN T

Alvarez n SVM
Anto y ¼ aT :bðxÞ þ s a is the weight; b(x) is a mapping to dimensional feature space; s is the expression of biasness.
et al. (2013)
Li et al. (2014a) RF 1 XN b MSEOBB is the out-of-bag (OBB) error which represents the mean square error (MSE) of RF; YXi are
MSEzMSEOBB ¼ ð Y ðXi Þ Yi Þ2
N i¼1 the predicted output; Yi is the observed output; N is the total number of samples.
0 1
Cui et al. WNN xl’ is the output of the hidden layer; j (x) is generating function; al and bl are the wavelet dilation
(2018) BPK w x b C and translator parameters respectively; wkl and wlm’ are weight vector in the hidden layer and
B l C
x0l ¼ jal ;bl B k¼1 kl k C output layer, respectively; xk is input layer data; xm’’ is the output of the output layer.
@ al A
00
0 ¼
PL 0 0
xm ¼ netm l¼1 wlm xl
Hannan et al. NARX yðn þ 1Þ ¼ wihwih,who who, wjhwjh are the weight vector; b0b0, bh are the bias vector; fh(.)f h ð:Þfh., f0 ð:Þ are
(2020b) 2 0 X
du 13 the activation function; du and dy are number of input delay and feedback delay, respectively.
6 B bh þ wih uðn du Þ C7
6 PNh B C7
f0 6 B C7
i¼0
6b0 h¼1 who :fh B X dy C7
4 @ A5
þ wjh uðn dy Þ
i¼0
Yang et al. LSTM fk ¼ sg ðWf xk þ Uf hk1 þ bf Þ xk is the input of the LSTM unit at time k; hk is the hidden state; ck is the unit memory; W, U, are the
(2019b) ik ¼ sg ðWi xk þ Ui hk1 þ bi Þ weight matrices; b is bias parameter; ik , fk and ok are the activation function of the input gate,
ok ¼ sg ðWo xk þ U0 hk1 þ b0 Þ forget gate, and output gate respectively; sg , sc , and sh are sigmoid, hyperbolic tangent, and
ck ¼ fk +ck1 þ ik sc ðWc xk þ UC hk1 þ bc Þ hyperbolic tangent, respectively.
hk ¼ ok +sh ðck Þ
Xiao et al. GRU zk ¼ sðWzx xk þ Wzh hk1 þ bz Þ zk and rk denote the gate updating and resetting vector, respectively; sstands for sigmoid
(2019) ~ define the state vector and candidate state, respectively; The parameters
rk ¼ sðWrx xk þ Wrh hk1 þ br Þ function; h and h
k k
~ ¼ tanhðW x x þ W h ðh during gate updating operation are denoted by Wzx , Wzh , and bk ;while the parameters during gate
h k h k h k1 +rk Þ þ bh Þ
~ þ ð1 z Þ+h resetting operation are represented by Wrx , Wrh, and br; the candidate state parameters are
h ¼ z +h
k k k k k1
depicted by Whx, Whh and bh .
degradation model to predict SOH using support vector regression lithium-ion batteries. GPR method was applied to achieve the
(SVR) and ICA. The improved filter method was used to smooth smoothing incremental capacity curves. The input features were
incremental capacity curves followed by the peak fitting technique extracted from the partial incremental capacity curves. The pre-
to decompose the smooth curves. The training dataset was devel- dictive capability of the proposed model was verified using the
oped using the extracted features from the decomposed incre- aging datasets from NASA along with the tested batteries under the
mental capacity curves. The proposed model was validated using different initial health conditions. The validation report illustrated
the aging datasets and achieved RMSE and MAE under 2%. Li et al. high accuracy and robustness in SOH estimation.
(2014b) designed a reliable prognostic model to examine SOH using
a relevance vector machine (RVM). The experimental results indi- 3.1.2.3. Recurrent neural network algorithm-based SOH prediction.
cated an accuracy of 90% with an acceptable limit of error tolerance. Eddahech et al. (2012) designed a RNN network with current,
Peng et al. (2018) examined battery capacity and SOH based on GPR temperature, SOC variation, and constant-phase-element (CPE) to
using the raw data collected from the Center for Advanced Life predict SOH by taking into account electrochemical impedance
Cycle Engineering (CALCE) at the University of Maryland and the spectroscopy (EIS) measurements. RNN obtained MSE of 0.462%
Department of Engineering Science (DES) at the University of Ox- and 0.296% in capacity and resistance prediction. Qu et al. (2019)
ford. The results illustrated the applicability and stability of GPR established a hybrid intelligent method with LSTM and PSO to
with an error of 2.2%. Li et al. (2020c) developed a novel multi-time- determine SOH of lithium-ion batteries. PSO based LSTM was
scale framework to predict SOH for lithium-ion batteries based on validated by the National Aeronautics and Space Administration
GRP and partial incremental capacity (PIC). First, PIC curves were (NASA) Prognostics Center of Excellence (PCoE) dataset at RT and
employed to extract significant features from different degrees reported RMSE below 2%. Ungurean et al. (2020) presented a deep
such as intercept, peak and slope. Then, kernel-modified GPR was learning-based SOH estimation model for lithium-ion batteries
used to build a short-term battery degradation model and using GRU. GRU outperforms LSTM with regard to estimation error
accordingly autoregressive long-term battery prediction model was and few parameter requirements.
designed. The four-dataset consisting of different health levels and
cycling test conditions were applied to verify the effectiveness and 3.1.2.4. Entropy algorithm-based SOH prediction. Hu et al. (2014)
robustness of the proposed model and results were promising with proposed sample entropy-based SOH prediction using an aging
regard to battery health situations forecasting. Li et al. (2020d) dataset at three different temperatures. The results were excellent
introduced a novel ICA based GPR model to predict SOH for under the HPPC test with an average relative error of 2%. Hu et al.
7
Table 2
The evaluation results of the intelligent algorithms for SOC estimation in BMS.
Algorithms Refs. Input features and output Structure/functions/ Verification profiles Error rate Research gaps/future works
hyperparameters
FNN He et al. Variables ¼ [I (t), I(t-4),…, I(t- -HLs 2 10 C, 25 C, 40 C, -RMSE 2.5% -The complexity of FNN was increased with
(2014) 4k), V (t), V (t-4),…,V(t-4k), Tem -HNs 5 and 50 C (US06) the inclusion of Kalman filter (KF).
(t), T (t-4),…, T (t-4k), -Sigmoid function temperatures -RMSE 1.4% -Aging effect was not examined.
Output ¼ [SOC (t)] - US06 and FUDS drive (FUDS)
cycles
RBFNN Kang et al. Variables ¼ [I(t), V (t), Cap (t)] -Single HL 10 C, 25 C, and -MAE 5% -The network parameters were not optimized.
(2014) Output ¼ [Voltage (t)] -Gaussian function 40 C temperatures -Computational cost was not mentioned.
- ECE and UDDS drive
cycles under 100, 400,
and 700 cycles.
ELM Lipu et al. Variables ¼ [I(t), V (t), T (t)] -Single HL 25 C, and 45 C -RMSE 0.76% -The proposed model was tested with only two
(2019b) Output ¼ [SOC (t)] -HNs 147/158/220/328. temperatures (BJDST) drive cycle data and two temperatures.
-Sigmoid function -BJDST and US06 drive -RMSE 1.56%
cycles (BJDST)
SVM
Alvarez Variables ¼ [I(t), V (t), T (t)] -Kernel parameter 0.5 -Different charge and -R2 0.97 -The model was not validated under different
Anton Output ¼ [SOC (t)] -SVM type regression discharge current environmental settings and drive cycles.
et al. 0.25 rates.
(2013) -Room temperature
(RT)
RF Li et al. Variables ¼ [I (t), V (t), T (t), DT -Number of trees 500 - NASA database -RMSE 0.14% -The outcomes lacked in detailed comparative
(2014a) (t), VC (t), TC (t)] -Number of variables 2 - RT (dynamic state) analysis with other intelligent methods under
Output ¼ [SOC (t)] -RMSE 0.59% different operating conditions.
(steady state)
WNN Cui et al. Variables ¼ [I(t), V (t)] -Multi-hidden layer -NEDC, UDDS drive -MAE < 1% - Temperature impact was not investigated.
(2018) Output ¼ [SOC (t)] -HNs 10 cycle -SOC ErrorMAX -Scaled conjugate gradient and Bayesian
-Morlet wavelet 25 C temperature. < 3.5% regularization can be employed in the future.
function
GPR Sahinoglu Inputs ¼ [I(t), I (t-1), V(t), V (t- -Single HL 10 C, and 25 C -RMSE < 0.3%
- The validation was performed without
et al. 1), T (t), T(t-1), SOC (t-1)] -Kernel function temperatures. -MAE < 1%temperature effects.
(2018) Output ¼ [SOC (t)] -EV load profiles. -The influence of battery aging can be
considered in the future.
TDNN Chaoui Inputs ¼ [I(t), I (t-1),…, I (t-20), -HLs 2 6 A, 10 A, and 20 A -RMSE 0.33% -Validation was limited to only specific charge
et al. V (t), V (t-1),…,V (t-20), T (t), -HNs 3/2 (first/second discharge currents. discharge current rates.
(2017) Output ¼ [SOC (t)] hidden layer) -Ambient temperature -Detailed comparative analysis was not carried
-Sigmoid activation out.
function.
LSTM Chemali Variables ¼ [I(t), V (t), T (t)] -HNs 50 -HWFET, UDDS, -SOC ErrorMAX -The hyperparameters of LSTM were chosen
et al. Output ¼ [SOC (t)] -Batch size 250, 500, LA92 and US06 drive 5.38% randomly.
(2017) and 1000 cycles. -High configuration graphical processing units
-Sigmoid and tanh -RT (GPU) are required for fast training execution.
GRU Xiao et al. Variables ¼ [I(t), V (t), T (t)] -GRU units 200 10 C, 30 C, and -RMSE 2.45% -The hyperparameter selection of the GRU was
(2019) Output ¼ [SOC (t)] -Batch size 230 50 C temperatures. (FUDS) based on experience.
-Learning rate (LR) -FUDS, DST and US06 -RMSE 1.97% -Robustness validation under unknown initial
0.001 and 0.0005 drive cycles. (DST) SOC was not performed.
-Epochs 100 -RMSE 1.40% - High processing power and memory
-Ensemble optimizer (US06) footprint are required.
-Sigmoid and tanh
GA Chang, Variables ¼ [I(t), V (t), T (t)] - RBFNN - 0.3C, 0.5C and 1 C -Maximum -The model became complex with three
(2013) Output ¼ [SOC (t)] - HNs 24 discharge currents absolute error algorithms: RBFNN, GA and orthogonal least
-The center, and width - RT of 1.82%. square.
of the activation -EV load profiles and varying temperatures
function are optimized. were not investigated.
BSA Hannan Variables ¼ [I(t), V (t), T (t)] -FNN configuration 0 C, 25 C, and 45 C -RMSE 1.74% - The training process was slow due to the
et al. Output ¼ [SOC (t)] -The HNs and LR are temperatures. (FUDS) selection of a dual population of BSA.
(2018b) optimized -FUDS and DST drive -RMSE 1.47% -Comprehensive experiments are required on
cycles. (DST) different types of lithium-ion batteries.
GSA Lipu et al. Variables ¼ [I(t), V (t), T (t)] -ELM structure 25 C, and 45 C -RMSE 0.76% - GSA can be trapped in local minima.
(2019b) Output ¼ [SOC (t)] -The HNs are optimized. temperatures. (BJDST) - The proposed model requires real-time
-BJDST and US06 drive -RMSE 1.56% evaluation through a prototype embedded
cycles. (US06) system.
LSA Hannan Variables ¼ [I(t), V (t), T (t)] -NARX model 0 C, 25 C, and 45 C -MAE 0.58% -Implementation of LSA into NARX was not
et al. Output ¼ [SOC (t)] -The IDs, FDs and HNs temperatures. (FUDS) easy due to several parameters and three
(2020b) are optimized - FUDS and US06 drive -MAE 0.38% projectiles.
cycles. (DST) - The proposed model needs a prototype
hardware verification using modularized
design.
FA Lipu et al. Variables ¼ [I(t), V (t), T (t)] -TDNN configuration -HPPC load profile. -RMSE 0.8512% - FA suffers from slow convergence speed.
(2020b) Output ¼ [SOC (t)] -The HNs and input (LiNiCoAlO2) - The model robustness needs to be checked
time delay are -RMSE 0.7937% under temperature and aging impacts.
optimized. (LiNiMnCoO2)
I: Current, V: Voltage, T: Temperature, Cap: Capacity, DT: Discharge time, VC: Voltage change, TC: Temperature change, NEDC: New European Driving Cycle; HWFET: Highway
Fuel Economy Test, UDDS: Urban Dynamometer Driving Schedule, BJDST: Beijing Dynamic Stress Test, SDT: Static Discharge Test.
8
(2016a) designed a SOH estimation approach using sparse Bayesian demonstrated effectiveness in terms of uncertainty management
predictive modeling (SBPM) based sample entropy algorithm. The and prediction performance. Li et al. (2019b) combined the LSTM
verification and comparison of the proposed model were carried with Elman neural network (ENN) and EMD to develop a novel
out through an experimental platform using lithium-ion batteries hybrid approach to predict RUL. The EMD was applied to decom-
at three different temperatures. The results indicated an improve- pose the battery capacity curve into some sub-layers while the
ment of 15.6%, 6.7%, and 0.7% in achieving average relative capacity LSTM and ENN were employed to forecast the high and low-
errors compared to the polynomial model at 10 C, 22 C, and 35 C, frequency sub-layers, respectively. A comprehensive experimental
respectively. The entropy method can assess the data fluctuation test was conducted to evaluate the performance and results were
and predictability but has drawbacks with regard to high compu- promising, indicating relative error under 3.5%. Ma et al. (2019)
tational complexity. proposed false nearest neighbors method-based hybrid neural
network method to predict RUL for lithium-ion batteries. The
3.1.2.5. Other algorithms. Yang et al. (2017) presented GA based hybrid neural network was structured using CNN and LSTM algo-
SOH prediction model to identify capacity and initial SOC accu- rithms. The false nearest neighbors approach was employed to
rately. The experimental cases proved the stability, accuracy, and determine the optimal window size while hybrid neural network
robustness of the proposed method under different aging states, was used to conduct the model training and testing. The compar-
and results indicated maximum capacity error and SOC error below ative analysis with other state-of-the-art methods proved the
5% and 2.1%, respectively. Zhu et al. (2016) considered the battery effectiveness of the proposed method in forecasting RUL with lower
internal resistance to monitor SOH based on Grey-Markov Chain error rate.
(GMC) algorithm. GMC was validated through life-cycle experi-
mental tests using 16 battery cells with 3,200 groups of data. The 3.1.3.3. Statistical data-driven algorithm-based RUL prediction.
results demonstrated the effectiveness of GMC in predicting ca- Li et al. (2016) proposed a Gaussian process mixture (GPM) based
pacity and SOH for LiFePO4 battery under different state levels. novel RUL prediction technique of lithium-ion battery. The results
Comparative performance analysis of intelligent algorithms-based illustrated that GPM was more accurate and robust than SVM with
SOH estimation is depicted in Table 3. regard to RMSE, indicating 1.58% and 1.3% at 60th and 80th cycles,
respectively. Tang et al. (2014) proposed RUL prognostic strategy
3.1.3. Remaining useful life prediction based on the Wiener process with measurement error (WPME). The
RUL is the remaining number of load cycles or time of a battery practical case studies, as well as numerous examples, illustrated the
cell to meet its performance requirement (Zhou et al., 2020). RUL is superiority of WPME over the traditional maximum likelihood
a critical parameter that supports online decision-making activities estimation (MLE) in terms of accuracy and robustness.
such as fault identification and diagnosis (Li et al., 2020f). The ac-
curate estimation of RUL is rarely feasible due to the numerous 3.1.3.4. Monte Carlo algorithm-based RUL prediction. Xiong et al.
uncertainties which affect such future prediction (Zhang et al., (2019) employed a moving-window-based method to indicate
2019b). Hence, the researchers have been trying to develop an lithium-ion battery RUL using Monte Carlo (MC) simulation. The
efficient RUL estimation method considering the uncertainties experiments were conducted using 1 C and 2 C current rates under
related to prediction and prognostics (Zhou et al., 2020). In this 25 C and 40 C temperatures and results reported capacity esti-
study, RUL prediction methods are divided into machine learning- mation errors within 1.5%. Zhang et al. (2019a) employed MC
based algorithms, time-series based algorithms, statistical data- simulation to develop an enhanced RUL prognostic model for
driven algorithms, Monte Carlo algorithms, and other algorithms. lithium-ion batteries using the Box-Cox transformation (BCT). An
accurate RUL prediction was obtained with standard deviations and
3.1.3.1. Regression algorithm-based RUL prediction. Wang et al. error of [1.8, 7] cycles, and [20, 10] cycles, respectively. MC can
(2014a) designed RVM based SOH prediction model utilizing the map the non-linear, complex problems and provide satisfactory
capacity degradation model of lithium-ion batteries. RVM was solutions in probability distributions but it needs high computation
examined by randomly selected aging cycles using three lithium- powerful processor.
ion battery cells at ambient temperature. RVM provided prog-
nostic solutions with reasonable absolute percentage error. Wang 3.1.3.5. Other algorithms. Ng et al. (2014) established a naive bayes
et al. (2013) employed PSO to find the optimal parameters of the (NB) based degradation model of lithium-ion battery to evaluate
autoregressive (AR) model toward accurate RUL prediction of RUL at different temperatures and discharge current rates. NB
lithium-ion batteries. The absolute prediction errors of 9 cycles, 7 delivered better predictions than SVM with an average RMSE of
cycles, 5 cycles, 0 cycles, 8 cycles and 1 cycle were noted in 104, 128, 16.1 cycles (0.17%), and SD of 10.7 cycles (0.15%). NB has an easy
146, 157, 170, and 183 inspection cycles. Xue et al. (2020) proposed execution process and short computation time; nonetheless, it has
an improved SVR model optimized by GA toward accurate predic- dependencies among variables. He et al. (2011) introduced the
tion of RUL of lithium-ion batteries. GA was used to find the optimal DempstereShafer Theory (DST) and Bayesian Monte Carlo (BMC)
key parameters of SVR. The GA optimized SVR was tested by NASA techniques to examine RUL using the double-index degradation
battery dataset and achieved better RUL prediction outcomes in model. DST based BMC predicted RUL accurately with SD of 12 and
comparison to other methods. 63 cycles in case one and case two respectively. The performance
analysis of intelligent methods for RUL prediction is presented in
3.1.3.2. Recurrent neural network algorithm-based RUL prediction. Table 4. The performance analysis of state estimation in BMS under
Zhang et al. (2018d) proposed LSTM to predict RUL under different different case studies including temperature, noise, unknown
current rates and temperature settings. LSTM was examined by four initial SOC and aging cycles is depicted in Table 5. The advantages
lithium-ion cells under different current rates (1 C, 2 C, and 3.5 C) at and disadvantages of intelligent algorithms in battery state esti-
25 C and 40 C temperatures. The validation reports indicated the mation are shown in Table 6.
superiority of LSTM over SVM and simple RNN. Liu et al. (2019)
introduced a data-driven ensemble prognostic approach based on 3.2. Battery equalization in BMS
combined Bayesian model averaging (BMA) and LSTM. The CALCE
dataset was used for validation and the proposed model The battery pack in EV is typically configured using multiple
9
Table 3
The assessment results of intelligent approaches for SOH prediction in BMS.
Algorithms Refs. Input features and output Structure/ Verification profiles accuracy Research Gaps/Future Works
functions
FNN You et al. Variables ¼ [I (t), V (t), T (t)] -HLs 2 10 EV drive cycles -RMSE 1.06% -Rest impact on the aging process need to be
(2016) Output ¼ [SOH (t)] -HNs 80 10 C, 25 C, 45 C, -MAE 3.81% considered.
and 60 C -The proposed method can be extended to evaluate
temperatures. SOH for the lithium-ion battery packs.
ELM Pan et al. Variables ¼ [I (t), V(t), T (t), (R -Single hidden -NEDC, UDDS, and -RMSE 1.09% -The effect of low temperature on the resistance can be
(2018) (t), PR (t)] layer Japan Drive Cycle -MAE 1.72% considered further.
Output ¼ [SOH (t)] -HNs 20 (JP1015) -SOH prediction for battery pack comprising series-
-Sigmoid - 25 C ± 2 C parallel combination needs to be studied.
function temperature.
SVM Klass et al. Variables ¼ [I(t), V (t), T (t), SOC -SVM Kernel -EV current profiles -RMSE 0.85% -SVM faces difficulties in calculating the loss function
(2014) (t)] option 14 0 , 10 , 20 , 30 (temperature- and penalty factor.
Output ¼ [SOH (t)] -Radial based and 40 C independent) -The diffusion resistance factor for drive cycles need to
kernel temperatures. -RMSE 0.63% be explored.
function (temperature
-dependent)
RVM Li et al. Variables ¼ [I (t), V (t), Cap (t)] -Polynomial 500,600,700, and -RMSE 0.357% -Complexity of the nonlinear mapping was raised with
(2014b) Output ¼ [SOH (t)] kernel 800 aging cycles -R2 0.98 the forecast horizon.
function -Ambient -The battery failure and preventive maintenance need
temperature (25 C) to be considered to enhance the reliability of the
battery system.
GPR Peng et al. Variables ¼ [I (t), V (t), Cap (t)] -Single hidden -CALCE dataset -SOC error <7% -The proposed model needs to be updated under
(2018) Output ¼ [SOH (t)] layer -DES dynamic conditions.
-Kernel 40 C temperature. -The outcome of this model can be compared with the
function RVM and model-based approaches.
RNN Eddahech Variables ¼ [I (t), Temperature - HNs 6. -ECE 15 urban -MSE 0.462% -The real-time automated system for observing battery
et al. (t), SOC (t), R/C (t), R/C (t-1), R/C - Prediction driving cycle (capacity) performance and predicting SOH needs to be
(2012) (t-2)] window 3 - 40 C, 45 C, and -MSE 0.296% developed.
Output ¼ [SOH (t)] -Sigmoid 50 C temperatures. (resistance)
activation
function.
LSTM Qu et al. Variables ¼ [I (t), V (t), Cap (t)] -HNs 64 -NASA PCoE -RMSE 2% -The proposed model needs to be assessed for a higher
(2019) -Output ¼ [SOH (t)] -Batch size 20, -RT number of cycles.
29, and 35 -The adaptability of the method needs to be verified
-Window size using multiple cells and dynamic load conditions.
9, 10, and 11
-Sigmoid and
tanh
GRU Ungurean Variables ¼ [Cap (t)] -GRU layers 2 -NASA PCoE -SOC -Key parameters including, hidden units, batch size,
et al. -Output ¼ [SOH (t)] -GRU unit 50 4 C, 25 C, and ErrorMAX < ±6% weight and bias are not optimized.

(2020) -Window size 44 C temperatures. -The embedded implementations need to be carried
5 out.
-Epoch 150
-Dropout 0.5
-Sigmoid and
tanh
Entropy Hu et al. Variables ¼ [I (t), Voltage (t), T -Window -HPPC load profile -SOC ErrorMAX -Validation can be compared other optimization
(2014) (t), Cap (t)] length 1, and 2 10 C, 22 C and 2% algorithm for better result.
Output ¼ [SOH (t)] -Tolerance 35 C temperatures.
between 0.1
and 0.25
-Length 100 to
5,000 data
points
R: Ohmic resistance, C: Capacitance, PR: Polarization resistance.
batteries connected either in series or parallel to fulfill the power consumption, heat dissipation, and slow equalization speed
and energy supplies of EVs. Nevertheless, due to the aging influ- (Baronti et al., 2014). In contrast, the non-dissipation equalization
ence, ambient temperature variation, and accordingly change in approach is designed using energy transfer circuits and features
battery internal resistance and capacitance result in battery high efficiency, low energy consumption, and fast equalization
inconsistency issues which reduce the safety, lifespan, effectiveness (Shang et al., 2017b). Generally, the battery voltage, capacity, and
of battery pack (Shen and Gao, 2019) (Huang and Abu Qahouq, SOC are employed as the equalization variables, as shown in Fig. 5
2015). Hence, the development of an efficient and reasonable (Feng et al., 2019). Battery voltage-based equalization has easy
equalization control strategy is an urgent necessity to address these implementation but it is affected by the internal parameters (Shang
battery inconsistency problems. The equalization control method et al., 2018). Battery capacity-based equalization has the benefits of
of the battery pack is divided into groups; dissipation equalization utilizing the maximum capacity, however, it cannot be appropriate
and non-dissipation equalization approach (Y. Chen et al., 2016; for online operation (Diao et al., 2018). Battery SOC-based equal-
Shang et al., 2017a) The dissipation equalization approach is ization has positives with regard to battery pack safety improve-
structured using parallel resistors in every single battery and ex- ment, and energy optimization but requires complex algorithms
hibits shortcomings in terms of low efficiency, large power and a powerful controller to achieve precise solutions (Li et al.,
10
Table 4
The comparative study of intelligent methods for RUL prediction in BMS.
Algorithms Refs. Input features Structure/functions Validation Profiles Accuracy Research Gaps/Future Works
and output
RVM Wang Inputs ¼ [Cap -Polynomial kernel function -Randomly selected -Absolute prediction error -Further research can be conducted under
et al. (t)] inspection cycles. 4,3,7,2 cycles at 10, 117, 176, a higher number of aging cycles.
(2013) Output ¼ [RUL - Ambient temperature 190, cycles respectively
(t)] (25 C)
LSTM Zhang Inputs ¼ [Cap -LSTM layer 2 1 C, 2 C and 3.5 C -SD 28 (253 cycle for cell 1) -The proposed model needs to be
et al. (t)] -HNs 50, and 100 discharge current under -SD 24 (285 cycle for cell 2) evaluated under EV drive cycle data.
(2018d) Output ¼ [RUL -Retained probability 0.2 maximum 600 aging
(t)] -Sigmoid and tanh cycles
25 C, and 40 C
temperatures.
NB Ng et al. Inputs ¼ [Cap -Discriminant function -NASA test data -Average RMSE of 16.1 cycles -The accuracy of the proposed model
(2014) (t)] -Regression Probability -1C and 2C discharge (0.17%) needs to be checked under dynamic tests
Output ¼ [RUL distribution current -SD of 10.7 cycles (0.15%). and varying temperature settings.
(t)] 10-fold cross-validation -RT (25 C) temperature.
(CV) in training
GPM Li et al. Inputs ¼ [Cap -Gaussian covariance -NASA PCoE -RMSE 1.58% at 60th cycles -Future research work can be conducted
(2016) (t)] function -RT (25 C) temperature. -RMSE 1.3% at 80th cycles on improving training speed.
Output ¼ [RUL -Kernel function
(t)] -Maximum likelihood
algorithm
-Fixed mixing proportions as
the gating function
WPME Tang Inputs ¼ [Cap -Inverse Gaussian -NASA test data MSE <0.1% -Nonlinear degradation trend and
et al. (t)] distributions -RT (25 C) temperature. relaxation effect can be take into
(2014) Output ¼ [RUL consideration for future research.
(t)]
MC Zhang Inputs ¼ [Cap -Transformation 0.5C, 1C, and 2C -SD 4.2 cycles (Cell A) -The acceleration of aging time using three
et al. (t)] parameter 6.53 and 8.72 discharge current -SD 7.2 cycles (Cell C) or more battery cells under test conditions
(2019a) Output ¼ [RUL for Cell C and Cell D 10 C, 25 C, and 40 C -SD 3.9 cycles (Cell E) need to be monitored.
(t)] respectively. temperatures.
-Pearson correlation
coefficient 0.9888
2019a). This review categorizes the battery equalization methods algorithm. The authors used a two-stage DC/DC converter topology
into two parts; rule-based algorithms and optimization-based and FL control to achieve the static SOC. Then, NN was used to
algorithms. obtain the dynamic SOC on the basis of static SOC. The proposed
method illustrated enhanced performance with regard to energy
efficiency and time efficiency.
3.2.1. Rule-based algorithms
Usually, various kinds of converter circuits such as buck-boost,
3.2.2. Optimization-based algorithms
Cuk converters, fly-back converter along with controller; fuzzy
Zhang et al. (2015) established an intelligent battery equaliza-
logic, proportional integral derivative (PID) are deployed for battery
tion control method with fuzzy logic controleneural network (FL-
equalization to achieve better scalability, shorten equalization time
NN) integrated with GA optimization. The results reported that the
and high efficiency. Wang et al. (2020a) designed a buck-boost
GA based equalization scheme improved the equalization time and
equalization circuit to minimize the cell inconsistency of series-
efficiency by 1002.5 s and 93.1%, respectively, compared to the
connected lithium-ion batteries using voltage and SOC as equal-
mean-difference algorithm. Sun et al. (2015) designed a novel SOC
ization variables. FLC was employed to minimize equalization time
based equalization algorithm for lithium-ion batteries using PSO.
and enhance efficiency. The proposed method improved the
The experimental results denoted that SOC based equalizer ach-
equalization time and efficiency by 49% and 4.88% under static,
ieved maximum SOC gap within 2% and improved the capacity by
charging, and discharging situations. Ma et al. (2018) proposed FLC
13.2%. Liu et al. (2017b) proposed an equalization strategy for the
based equalization scheme to address the inconsistency in series-
series-connected battery cells based on ant colony optimization
connected battery cells using a two-stage bidirectional equaliza-
(ACO). ACO achieved satisfactory solutions in terms of minimizing
tion circuit. Simulation results demonstrated the superiority of FLC
energy loss as well as achieving high balancing efficiency and high
over the mean-difference method, reducing SOC deviation and
speed. Table 7 denotes the detailed comparative study of the
equalization duration by 18.5% and 23%, respectively under NEDC
equalization methods concerning variables, objectives, strengths,
drive cycle. Nguyen et al. (2014) focused on designing individual
and weaknesses.
cell equalizers (ICEs) based on the adaptive neuro-fuzzy algorithm.
The proposed model is executed in a 0.13- mm CMOS technology
and obtained a reasonable learning accuracy error of 1.8 e5. Liu 3.3. Battery fault diagnosis in BMS
et al. (2017a) presented model predictive control (MPC) to mini-
mize the SOC imbalance and energy dissipation of the lithium-ion Battery safety is of great importance to guarantee the secure,
battery pack. The MPC terminated the balancing when SOC differ- reliable, and safe operation of EVs. Thus, the development of a
ence and energy dissipation reached 4%, and 7% respectively, suitable diagnostics and fault handling mechanism is very signifi-
indicating 93% efficiency. Zhang et al. (2015) introduced a novel cant since a minor fault could ultimately cause severe problems to
intelligent control scheme for battery equalization based on FL-NN battery health (Tran and Fowler, 2020). Based on the literature
11
Table 5
Evaluation of case studies of state estimation in BMS.
Cases Refs. Settings Battery state Intelligent Battery type Dataset Results
estimation (SOC/ algorithm
SOH/RUL)
Temperature Chemali 20 C, 10 C, 0 C, 10 C, and SOC DFNN 2.9 Ah Panasonic 18650 PF -US06 -MAE 3.4%
et al. (2018) 25 C lithium-ion battery -HWFET (HWFET)
-MAE 2.2% (US06)
Bian et al. 0 C, 10 C, and 25 C SOC LSTM 2.9 Ah Panasonic 18650 PF -HWFET -SOC
(2020) lithium-ion battery errorMAX 4.31%
-MAE 1.09%
Ungurean 4 C, 25 C, and 44 C SOH GRU 1.85 Ah 18650 lithium-ion -NASA PCoE -SOC
et al. (2020) battery errorMAX < ±6%
Zhang et al. 10 C, 25 C, and 40 C RUL MC 2.7 Ah LiNiCoAlO2 -Capacity -MAE 20.7 cycle
(2019a) lithium-ion battery degradation dataset (Cell A)
-MAE 53.7 cycle
(Cell C)
-MAE 10.6 cycle
(Cell E)
Noise Cui et al. -The random noises are added SOC WNN 2.15 Ah Samsung -UDDS -MAE 0.92%
(2018) -Positive and negative bias noises ICR-18650e22P lithium-ion -NEDC (random noise)
are included battery -UKBC -MAE 1.16% (bias
noise)
You et al. -Inclusion of noise in current and SOH LSTM 3.1 Ah 18650 lithium-ion - UDDS -Average SOC error
(2017) voltage with the aging cycles battery - LA92DDS 2.46%
Unknown Song et al. - 100%, 80%, 60%, 40% and 20% SOC CNN 1.1 Ah 18650 lithium-ion -DST -MAE 0.87%
initial SOC (2019) initial SOC battery -FUDS -RMSE 1.38%
-US06
Sahinoglu 50% initial SOC SOC GPR 4.93 Ah LiMn2O4 lithium- -Constant charge/ -MAE 0.54%
et al. (2018) ion battery discharge test -RMSE 0.27%
Aging Hannan 50, 100, 150, and 200 aging SOC NARX - 2.0 Ah Samsung ICR18650- -HPPC -MAE 0.96%
et al. (2020) cycles 26F lithium-ion battery -RMSE 1.28%
Chaoui et al. 0 h, 352 h, 544 h and 650 h SOC TDNN - 20 Ah LiFePO4 lithium-ion -Charge and -RMSE 0.33%
(2017) calendar aging profiles battery discharge current
profiles
Li et al. 50, 100 and 150 aging cycles SOH GPR 1.85 Ah 18650 lithium-ion -NASA PCoE -MAE 1.2%
(2020d) battery -RMSE 1.8%
Li et al. 1,30,60,90,120 and 150 aging SOH SVR - LiNiMnCO2 lithium-ion -Aging cycle test -MAE 1.003%
(2020b) cycles (LiNiMnCO2) battery (LiNiMnCO2-Model
1,80,160, 240, 320 and 400 aging - LiFePO4 lithium-ion 1)
cycles (LiFePO4) battery -MAE 1.642%
(LiFePO4-Model 1)
Li et al. 100, 200, 300, 400, 500, 600 and RUL LSTM 1.1 Ah LiCoO2 lithium-ion -CALCE dataset -RMSE 0.915 (0.5C)
(2019b) 700 aging cycles battery -RMSE 0.974 (1C)
Li et al. 50, 100 and150 aging cycles RUL GPR 1.85 Ah 18650 lithium-ion -NASA PCoE -MAE 8.51 cycle
(2020c) battery -RMSE 11.62 cycle
review (Chao et al., 2017; Hendricks et al., 2015; Lyu et al., 2019), The battery faults are considered as the critical faults in BMS
The fault mechanisms in BMS are usually very complicated and can which take place due to overheating, overcharging, over-
be classified into three groups; sensor fault, actuator fault, and discharging, internal short circuit (ISC), external short circuit
battery fault, as depicted in Fig. 6. Sensor faults in BMS cover the (ESC), battery swelling, electrolyte leakage, and thermal runaway.
faults in the current, voltage, and temperature sensors (Lombardi The overcharging and over-discharging could cause side reactions,
et al., 2014). The current sensors faults can deviate the estimation health degradation which eventually result in battery swelling and
accuracy of SOC, SOE, SOH and RUL. Moreover, the battery must be electrolyte leakage. Battery overcharge and over-discharge could
functioned within the safe limit of voltage and temperature rec- lead to various undesirable battery side reactions, resulting in
ommended by the manufacturer (Lu et al., 2013). If the measured accelerated degradation. These side reactions and gases generated
values exceed the boundary, the performance of the battery may by the chain reactions during the thermal runway may eventually
deteriorate or even lead to accidents. Besides, sensor faults in introduce battery swelling and electrolyte leakage (Bandhauer
voltage and temperate could result in TMS error or battery equal- et al., 2011). The ISC and ESC are occurred due to mechanical col-
ization error in BMS. lisions, and separator failure caused by manufacturing defects
Actuators faults in BMS include fuse fault, high voltage contactor (Yang et al., 2019c). Battery overheating is the direct cause to trigger
fault, controller area network (CAN), bus fault, terminal connector thermal runway which damages the battery with regard to the
fault, cooling system fault (Liu et al., 2015). Battery connection fault melting of the separator, and decomposition of cathode, solid
may result in a rise in resistance, produce extreme irregular heat, electrolyte interface (SEI), and electrolyte (Wang et al., 2012). Lamb
which further accelerates temperature (Feng et al., 2014; Han et al., et al. (2015) explored the failure propagation and thermal runway
2014; Liu et al., 2014a). Besides, the poor connection between of the multi-cell lithium-ion battery pack. Feng et al. (2016) adop-
batteries may deliver insufficient power supply which leads to a ted four preventive measures of thermal runway such as reduction
melting of the battery terminals and risk of accidents (Yao et al., of accumulated energy, improvement of heat dissipation, the in-
2015). If the cooling system does not work satisfactorily, the bat- clusion of thermal resistance between batteries, and the rise of
tery cannot operate within the temperature range, hence resulting onset temperature. Hofmann et al. (2017) revealed that the
in thermal runaway. reduction of battery pressure could eventually delay thermal
12
Table 6
The benefits and drawbacks of different intelligent algorithms in battery state estimation of BMS.
Methods Advantages Disadvantages Refs.
FNN - This method has an easy execution. - It has a lengthy data training process. He et al. (2014); Liu et al. (2014b);
- It has acceptable accuracy. You et al. (2016)
RBFNN - This method can achieve global approximation. - The training operation has local minimum Haddad Zarif et al. (2014); Chen et al.
- It obtains reasonable accuracy with incomplete information. trapping issues. (2016a)
- It requires a long training duration.
ELM - This method has a quick computation process. - The performance of ELM relies on the optimal Pan et al. (2018)
- It has better generalization and scalability. number of HNs.
SVM - This method has less-overfitting issues. - It has complex computation
Li et al. (2014b); Alvarez n et al.
Anto
- It has excellent outcomes in non-linear and multi-dimension - It has a time-consuming operation. (2013)
models.
RF - This method has quick and accurate execution under huge datasets. - Unable to provide satisfactory solutions in a highly Li et al. (2014a); Chen et al. (2018)
- It is accurate and robust to over-fitting and noise impacts. irregular system.
- The values of trees and leaves influence the
estimation and prediction outcomes.
WNN - This method has less complexity in selecting hidden nodes and - Several hidden layer units are needed to obtain Cui et al. (2018); Dong et al. (2015)
weights. reasonable solutions.
- It has easy training execution.
GPR - This method is excellent in capturing the model uncertainty. - Poor accuracy in multi-dimensional spaces. Sahinoglu et al. (2018); Peng et al.
- It has simple interpretation of various features. (2018)
NARX - This method is very effective to address time series problems. - The effectiveness relies on the appropriate Chaoui et al. (2017); Eddahech et al.
- It has intelligent computation in controlling the dynamic system. selection of IDs, FDs and HNs. (2012)
LSTM - This method has satisfactory outcomes under long-term - It needs long and complex training execution. Chemali et al. (2017); Qu et al.
dependencies. - It requires a costly processor to accelerate training (2019); Zhang et al. (2018d)
- It needs a lower computational effort during the online stage. operations.
GRU - This method overcomes the gating mechanism issues of LSTM. - It wants a huge amount of training data. Xiao et a. (2019); Ungurean et al.
- It can map the long-term sequential dependencies. - It needs a high-performance processor. (2020)
Entropy - This method can quantify the predictability, complexity, and - It has a high computational complexity. Hu et al. (2014)
fluctuation of the measured signals.
NB - This method has a short computation time. - It has dependencies among variables. Ng et al. (2014)
- It has an easy implementation process. - It needs a large number of records to achieve
acceptable solutions.
WPME - This method can handle degradation function and noise. - It has a high mathematical computation process. Tang et al. (2014)
- It has slow convergence.
MC - This method can model complex problems. - This method is costly. Zhang et al. (2019a)
- It has an excellent solution in probability distributions. - It needs high computation power.
- It is intuitive and relatively easy to execute.
BSA - This algorithm has enhanced the exploration capability process - It has a lengthy computation due to the dual Hannan et al. (2018b)
with the mutation and crossover technique. population algorithm.
- The search direction and crossover process are
complex.
PSO - This method has a simple execution. - This method can converge prematurely. Zhang et al. (2018e); Xia et al.
- It needs fewer parameters to adjust. - Difficult to outline the initial design parameters. (2018a)
- It has high efficiency and fast convergence speed.
GSA - This algorithm has easy execution and fast learning speed. - This algorithm has a poor searching capability. Lipu et al. (2019)
- It has a small training error and better generalization performance.
LSA - This algorithm has high exploration and exploitation search - It has a complex computation process. Hannan et al. (2020b)
capability.
- It has high efficiency and good convergence characteristics.
FA - This algorithm is flexible and has easy execution. - It has a limitation of slow convergence. Lipu et al. (2020b)
- It can provide correct results using a fewer number of iterations.
runaway propagation. learning-based algorithms, mathematical data-driven algorithms,

The methodological framework of fault diagnosis in BMS is entropy algorithm and other algorithms.
illustrated in Fig. 7 which includes data acquisition, feature
extraction, fault diagnosis, fault prognosis, and fault handling. First,
3.3.1. Rule-based algorithms
the data acquisition is employed to collect and store battery data
Xiong et al. (2012) suggested a probabilistic and a rule-based
including current, voltage, and temperature from experimental
technique to detect the failure in lithium-ion batteries. The Bool-
arrangements or test bench models. After, the feature extraction is
ean expression was formulated based on rules established while
carried out for data preprocessing, extraction, and selection. Then,
the voltage was dropped and the temperature was increased during
the selected data is processed for fault diagnosis and fault prognosis
the battery over-discharge stage. The proposed method was veri-
phases. Fault detection, isolation, and estimation are covered under
fied by experiments and executed in real-time to identify the faults
the fault diagnosis process. The fault prognosis stage includes
in over-charged batteries. Muddappa and Anwar (2014) proposed a
battery state algorithm execution and validation to deliver early
fault diagnosis method using fuzzy logic for high power battery
detection or prediction of battery faults. Finally, the fault handling
packs in EV applications. First, residuals with respect to SOC,
stage examines and assesses the outcomes from fault diagnosis and
voltage, and temperature were generated using a model-based
fault prognosis, and accordingly delivers effective decisions
observer. Then SOC, voltage, temperature change together with
including alarm generation, faulty batteries isolation, and power
the residuals were integrated into the fuzzy logic algorithm to
supply shut down. This study divides the fault diagnosis ap-
identify the various faults such as battery aging, overcharge, over-
proaches into five sections; rule-based algorithms, machine
discharge, thus providing a reliable, fast, and effective strategy in
13
connection faults. The developed method detected the faults

accurately with two different sensors and employed the enhanced
correlation coefficient to address the cell inconsistencies and
measurement error. Kong et al. (2018) suggested a quantitative
diagnosis method to identify the micro-short circuit fault of
lithium-ion batteries based on variation in remaining charging
capacity. The performance of the proposed model was examined
against different charging, aging conditions, and results were
satisfactory in terms of computing micro-short circuit resistance
and the leakage current with computational efforts. Xia and Mi
(2016) introduced a fault-tolerant voltage measurement approach
based on matrix interpretation for series battery packs. An elec-
tronic circuit was designed to separate sensor and cell faults by
locating the abnormal signals. The experimental results proved the
practicability and robustness of the proposed method concerning
diagnostic accuracy, cost, and diagnosis confidence. Xia et al. (2016)
concentrated to distinguish the faults between sensor and battery
cells using an enhanced interleaved voltage measurement method.
An improved circuit topology was designed and the model showed
improvements in reducing noise levels compared to the basic
interleaved topology.
3.3.4. Entropy algorithms

Fig. 5. Battery pack equalization during the charging/discharging control based on (a) Wang et al. (2017b) presented Shannon entropy-based a novel
voltage, (b) capacity and (c) SOC (Feng et al., 2019). fault diagnosis method to diagnose the voltage faults in battery
systems. The proposed diagnosis method was evaluated under real-
time voltage monitoring data and showed excellent forecast results
diagnosing battery faults. in terms of the time and location of the voltage fault in battery
packs. Liu et al. (2018) applied entropy theory to detect the voltage
3.3.2. Machine learning-based algorithms abnormality for a multiple-cell battery system. The entropy weight
Yang et al. (2018) presented RF-based classification method and was introduced to further improve reliability by identifying the
three steps model-based diagnosis algorithm to diagnose and abnormal cells and predict faults. Li et al. (2020a) proposed a novel
prognosis the electrolyte leakage and ESC fault. The pre-trained RF technique employing sample entropy and empirical mode
classifier used the RMSE indicator to confirm the leakage condition. decomposition (EMD) to detect faults in BMS under numerous
The several cases demonstrated the effectiveness of the proposed operational settings. The EMD was used to extract the fault features
method in diagnosing ECS battery cells. Zhao et al. (2017) focused experimentally using the voltage signal decomposition and noise
on developing a novel fault diagnosis approach for EV batteries interference cancellation. After, sample entropy detected the bat-
using 3s multi-level screening strategy (3s-MSS) and machine tery faults accurately based on extracted fault features. The real-
learning algorithm. The 3s-MSS was utilized to build the criteria for world EV data was utilized to validate the performance of the
the identification of the abnormal changes of battery terminal proposed method under normal and sudden faults.
voltages while the machine learning algorithm was employed to
execute the diagnosis process using the big data statistical regula- 3.3.5. Other algorithms
tion. Hong et al. (2019) developed a novel LSTM algorithm for Kim et al. (2018) introduced a real-time fault diagnosis strategy
battery fault diagnosis by taking into account the battery perfor- utilizing the outlier detection algorithm for multi-cell lithium-ion
mance concerning weather and driver’s behavior in EV operation. batteries. The Z-score standardized pre-processing method was
All hyperparameters of LSTM were pre-optimized offline using a used to detect the abnormalities in battery cells and identify faults
dual-model-cooperation prediction technique. The prognosis sta- in healthy battery cells, aged battery cells and shorted battery cells.
bility and feasibility of the method were examined using real-time Table 8 shows the evaluation of different kinds of faults analyzed by
data under various voltage abnormities. The results showcased the intelligent algorithms and controller schemes in BMS.
successful execution of the proposed method in battery protection
and safety toward the identification of advent battery faults. 3.4. Thermal management system in BMS
3.3.3. Mathematical data-driven algorithms The high temperature of the battery cell is a serious concern that
Xia et al. (2017) established a novel detection method based on needs to be addressed properly to improve EV performance. A
the correlation coefficient to identify the short circuit faults in thermal runaway of battery usually occurs when high temperatures
battery packs. The proposed method was executed without any activate the exothermic reactions which accelerate the temperature
hardware or analytical redundancy and worked effectively under further and leads to more damaging reactions (Hannan et al., 2020).
the inconsistencies in SOC and SOH. Li and Wang (2018) adopted a The impacts of thermal runway under different categories of
novel correlation coefficient-based fault diagnosis approach for the lithium-ion batteries are shown in Fig. 8. It is observed that the
lithium-ion battery pack in EVs. The proposed method not only electrolyte, cathode and solid electrolyte interface (SEI) layer start
amplified the voltage difference among batteries but also extended decomposing when the temperature increases beyond 90 C,
the fault memory by assigning a suitable size of moving windows. However, LiFePO4 exhibits limited exothermic heat discharge, thus
Kang et al. (2019) proposed a multi-fault diagnostic approach based providing improved thermal stability in comparison to other
on interleaved voltage measurement and recursive correlation co- lithium-ion batteries (Lu et al., 2013). Wang et al. (2016) found that
efficient to identify sensor faults, short circuit faults, and the lifespan of the battery had become shortened by two months if
14
M.S. Hossain Lipu, M.A. Hannan, T.F. Karim et al.
Table 7
Analysis of control strategies used for battery equalization in BMS.
Refs Equalization Type of converter Variables Objectives Strength Weakness Contributions Research Gaps/Future
method Works
Wang et al. FLC -Buck-Boost -Voltage - To minimize -FLC has a high fault -FLC has poor - Improves the -The temperature impact
(2020a) Converter -SOC equalization time. tolerance capability. flexibility. equalization time and and capacity degradation
- To enhance efficiency. - FLC has high -FLC rule generation efficiency by 49% and were not considered in SOC
equalization time and relies on knowledge 4.88%, respectively. estimation. -The path
efficiency. and human expertise. optimization algorithm is
required to reduce energy
loss and increase
equalization speed.
Ma et al. (2018) FLC -The two-stage -SOC - To reduce SOC -Minimizes SOC - The proposed equalization
bidirectional deviation. deviation and method was not validated
equalization circuit - To minimize equalization duration under aging impacts.
equalization time. by 18.5% and 23%, - The equalization effect
respectively. under the battery module
can be studied.
Nguyen et al. Neuro-fuzzy -Flyback converter -Current -To improve learning -This method can -This method needs the -Obtains a learning -The membership function
(2014) -Voltage ability from various address nonlinear, influx of data to achieve accuracy error of of the input variables was
working conditions. complex, and dynamic high precision. 1.8 e5. not optimized.
systems.
-It has an easy
implementation
process both in
hardware and software.
McCurlie et al. MPC -Flyback converter -SOC -To enhance the -This method is -This method is -Balancing terminates - The efficiency could be
(2017) efficiency of power appropriate for time- computationally when SOC difference is hampered due to the
electronics switching. varying, nonlinear, and demanding. 2%. switching loss and
15
-To extend battery life. multivariable -The solution is not conduction loss.
Liu et al. MPC -Bi-directional -SOC -To minimize energy processes. guaranteed. -Achieves 93% - This research could be
(2017a) Buck-Boost -Energy dissipation during -It can intrinsically efficiency in balancing. extended for a series-
converter balancing. handle the constraints. parallel combination of
battery packs to balance the
cell capacities.
Zhang et al. GA -Two-stage DC/DC -SOC -To improve energy -The concept of GA is -GA takes a long time to -GA scheme reduces - GA takes a substantial
(2015) converter efficiency. easy to understand. obtain near-optimal equalization time by amount of computing time.
-To minimize -GA is inherently solutions. 1002.5 s. -The superiority of GA over
equalization time. parallel, easily -GA is sensitive to input -GA scheme improves other optimization
distributed. parameters. energy efficiency by algorithms can be assessed.
93.1%.
Sun et al. PSO -Bidirectional DC/ -SOC -To reduce battery pack -PSO requires fewer -PSO can converge -Achieves the - A detailed comparative
(2015) DC converter imbalance and energy parameters to adjust. prematurely. maximum SOC gap analysis is required to prove
loss. -PSO has a simple -Hard to define the within 2% the PSO effectiveness.
Journal of Cleaner Production 292 (2021) 126044

-To improve battery execution. initial design -Improves the capacity -The battery cell SOC
pack performance. -PSO has high efficiency parameters. by 13.2% variation and temperature
and fast convergence influences can be
speed. considered.
Liu et al. ACO -Bidirectional DC/ -Voltage - To reduce the energy -ACO has guaranteed -ACO has a complex - Obtains total energy - The formulation of multi-
(2017b) DC converter loss convergence. theoretical study. loss up to 1.53 in the objective function and
-To achieve high -ACO has the -The convergence entire balancing other key factors need to be
balancing efficiency adaptability to changes duration of ACO is process. considered to achieve
and high speed. in new solutions. uncertain. better equalizer efficiency.
-Comprehensive
experimental verification is
required.
the battery temperature rises 1 C in the temperature range of

30 Ce40 C. Therefore, the heat management in the battery system
for EV applications is very crucial and further exploration is
required to design a suitable cooling and heating system for BMS.
An effective thermal management system (TMS) is essential to
maintain the battery temperature within the boundary toward the
optimum performance of BMS (Zhang et al., 2018a). The operating
temperature of the battery is controlled by the TMS so that each
battery cell must be operated inside a threshold temperature range.
When the temperature of the battery reaches outside the safe limit,
Fig. 7. General fault diagnosis framework for BMS (Niu, 2016).
TMS provides immediate response through the heating and cooling
management and control system thus ensuring safe operation and
protecting the battery from harmful occurrences (Jacob et al., 2019).
regulating the temperature and balancing SOC during the charging
Battery TMS either employs air as the heating/cooling ventilation or
and discharging processes. The FLC demonstrated strength in
liquid for the cooling/heating insulation layer, as illustrated in
reducing SD and equalization time of SOC by 18.5% and 23%
Fig. 9. The air TMS covers passive cooling, passive cooling/heating,
respectively. Umair Ali et al. (2018) demonstrated FLC based fast-
and active cooling/heating, as depicted in Fig. 9 (a) while liquid TMS
charging approach and temperature control unit for a lithium-ion
includes passive cooling, active cooling, and active cooling and
battery to protect the battery from overvoltage and overheating.
heating, as shown in Fig. 9 (b). The liquid TMS delivers better so-
Rahman et al. (2015) presented an advanced evaporative cooling (EC)
lutions with regard to heat capacity and thermal conductivity. An
TMS based on FLC to regulate the temperature of the battery in the
electronic control unit is used for the control operation of TMS
limit of 20e40 C during charging and discharging mode. The Proton
(Hong et al., 2017).
Saga EV was employed to conduct the experiments and results were
promising with regard to energy, saving 17.69% compared to air
3.4.1. Controller schemes of TMS based on cooling and heating cooling battery thermal management. Xia et al. (2018b) presented
Hannan et al. (2019) proposed a robust control system for TMS FLC based TMS to control the battery temperature toward the
using PSO based fuzzy logic controller (FLC). The effectiveness of enhancement of battery life, and performance. FLC can operate un-
fuzzy was compared with the PID controller with respect to over- der various uncertainties, nonetheless, it lacks flexibility and needs
shoot, undershoot, and settling time. The proposed method detailed knowledge and experience to develop fuzzy rules. Afzal and
improved the battery performance by controlling the temperature Ramis (2020) introduced a hybrid multi-objective optimization
and consequently achieved overshoot and undershoot of 0.497% and method based on GA and FL to develop a thermal management
0.975% during the heating and cooling subroutine, respectively. Ma system for a lithium-ion battery system. The maximum temperature,
et al. (2014) designed a novel TMS based on FLC to handle the in- Nusselt number and friction coefficient were selected as the objec-
consistencies of series-connected lithium-ion batteries. The devel- tive function. The results indicated that the proposed model ach-
oped model increased the safety of the battery system effectively by ieved the desired temperature range with the improved Nusselt
Fig. 6. The outline of various faults in the lithium-ion battery systems.
16
Table 8
Analysis of various fault detection methods based on intelligent algorithms and controller schemes in BMS.
Ref. Methods Type of Faults Benefits Shortcomings Contributions Research Gaps/Future Works
faults characteristics
Xiong et al. Rule-based -Overcharge - The rise of -Easy process and -Difficult to -Real-time execution -This method can be employed for
(2012) method Temperature. implementations. estimate the -Performs early warning and series and parallel connected
-The drop-in suitable parameters failure detection. lithium-ion battery pack.
temperature. in the rules - The proposed method can be
tested with other types of lithium-
ion batteries.
Muddappa Fuzzy logic -Aged -Derived from -Easy to operate with - Poor self-learning -The reliable and quick -The experimental validation is
and battery voltage, uncertainty. Skills. diagnosis method required for real-time battery
Anwar, -Overcharge temperature, and -High - Detects various faults applications.
(2014) -Over- SOC parameters. computational cost. including aged battery,
discharge overcharge, and over-
discharge
Yang et al. RF -Electrolyte -Maximum -Improved -Accuracy depends -Diagnosis of the ESC faults and -The accuracy and robustness of RF
(2018) leakage temperature rise. classification on a large number of leakage faults accurately under can be further checked with other
-ESC fault -Discharge capability. training datasets. several cases. regression methods such as SVM,
capacity. -Fast execution -Abundant fault -Low computational efforts. RVM.
process. data is not easily
accessible.
Zhao et al. FNN -Battery - Abnormal -Reliable and -A large amount of -Performs the fault diagnosis -Further research can be
(2017) fault changes in battery Accurate. training dataset is and prognosis with big data performed on battery fault taking
terminal voltages. -Not sensitive to needed statistical regulation. place in winter.
uncertainty -Data abundance for
training operation is
not easily
obtainable.
Hong et al. LSTM -Battery -Voltage -Powerful prediction -Hyperparameters -Detects the advent of battery -Performance of LSTM can be
(2019) fault abnormities. ability. adjustments is a faults. improved with parameter
laborious task. -Mitigates runway risks. optimization.
-Time-consuming - The dataset and model need to be
training operation. updated periodically to obtain
better results.
Wang et al. Shanon -Voltage -Irregular -Wide Range -The entropy -Assesses the time and location - Future work can be performed
(2017b) Entropy fault during changes in battery application. outcomes are of the voltage fault accurately. using a larger battery pack under a
EV operation voltages. -It can predict affected by the complex environment.
fluctuations in the computation -Data acquisition time interval
chaotic system. window could be expanded to 30 s in the
-It has a high future research
computational
burden.
Kong et al. Wavelet -Short -Remaining - It has low -Poor directional -Determines the micro-short -Further research could be
(2018) transform circuit faults charging capacity computational selectivity circuit resistance and the extended using an aging impact on
variation. complexity. -unpredictable leakage current with diagnostic results.
-It can map the variation in output computational efforts. -Experimental tests are required
signals more coefficients. for validation under short circuit
efficiently than the operation.
Fourier transform.
Kang et al. Correlation -Sensor -Voltage -Easy to apply and -It has a complex -Robust to cell irregularities -This topology can be redesigned
(2019) coefficient faults inconsistencies. understand. matrix and measurement noise. with a lower number of voltage
-Short -Accurate and interpretation. -Detect the location and type of sensors without affecting
circuit faults Robust. -Difficult to faults precisely. diagnosis results.
-Connection formulate a
faults mathematical
correlation
coefficient.
-Time-consuming
method.
Xia et al. Sensor -Sensor -Voltage -It works efficiently -High mathematical -It has a lower noise level. -This research could be extended
(2016) topology faults variations. without hardware or computation. - It has broader application by for multiple same subjects where
-Battery cell software -Complex circuit removing the coprime the fault tolerance level is critical
faults redundancy. design. constraint
number and lower friction coefficient. Although the integration of model predictive control (MPC) method to prevent non-uniform
FLC with heuristic algorithms can deliver satisfactory outcomes, it aging as well as thermal and SOC imbalances. MPC can extend to
has limitations in terms of fitness function selection and computa- time-varying, multivariable process and constraints; but it has
tional time. Lopez-Sanz et al. (2017) introduced a data-driven non- weaknesses concerning model construction and guaranteed solu-
linear model predictive (NMPC) control strategy to control the tions. Table 9 shows the performance of different controller schemes
temperature of high voltage battery cells. The proposed method for TMS of BMS.
maintained healthier temperatures of the battery and reduced the
cooling electrical consumption and temperature by around 38%. Altaf
et al. (2016) built a robust electro thermal control using a data-driven
17
some limitations. FNN-based approaches can achieve reasonable

results; however, performance is constrained due to storage ca-
pacity and execution time. The regression and probabilistic ap-
proaches are robust against noise, uncertainty and data-overfitting
issues; however, they fail to provide accurate solutions in high
dimensional and non-linear systems. The time series-based
methods can predict future predictions accurately; nevertheless,
the selection of past information and feedback steps is difficult to
determine. Although deep learning algorithms have provided
excellent results in SOC, SOH, and RUL prediction, they need good
quality and a vast amount of data as well as fast and expensive
computing processors. The optimization strategies can be inte-
grated with machine learning and deep learning algorithms;
nonetheless, accuracy could deviate due to the local minima trap,
poor search capability, and wrong parameter settings. The entropy
algorithm can measure the complexity and variation in the
measured signals, nevertheless, it has a shortcoming of the high
computational process. MC algorithm is outstanding in probability
and complex distributions, nonetheless it is expensive and needs
heavy computation power. The statistical data-driven based algo-
rithms can operate under noise and degradation function; however,
they have a high mathematical computation process which leads to
slow convergence. The fuzzy controller is reliable, customizable,
Fig. 8. Effects on high temperatures on different lithium-ion battery materials (Lu efficient; however, it requires lots of data and needs human
et al., 2013).
expertise. MPC has improved transient response and can control
multiple variables within the boundary, but it suffers from the high
4. Key issues and challenges of BMS computational load, high algorithm complexity, and the high
number of control parameters.
The implementation of intelligent approaches employed in BMS
for EV applications has become a major concern due to the algo- 4.2. Implementation issues
rithm complexity as well as various internal and external factors.
The key issues and challenges are divided into three groups; algo- This section covers the various implementation barriers of
rithms issues, implementation issues and battery-related issues. intelligent approaches and control schemes of BMS in terms of data
variety, data integrity, algorithm complexity and uncertainties.
4.1. Algorithms issues 4.2.1. Data abundance and data variety

The data abundance and data variety are the key issues in
Although intelligent techniques have demonstrated promising executing intelligent algorithms. The precision of intelligent
solutions toward the development of an efficient BMS, they have methods relies on sufficient quantity and quality data.
Fig. 9. Schematic diagram of battery TMS based on air and liquid (a) Air TMS, and (b) Liquid TMS (Hong et al., 2017).
18
Table 9
Literature survey of the controller schemes for TMS of BMS.
Intelligent Ref. Thermal Type Target Pros Cons Achievements Research Gaps/
methods Future Works
FLC Hannan et al. Heating and Cooling -To control -The membership - The method is -Obtains an -This topology can
(2019) temperature during functions (MFs) of overshoot of 0.497%
validated with only be validated using a
the high rate of FLC are optimized 10 series- during the heating prototype for a
charging and using PSO. connected battery subroutine. lithium-ion battery
discharging process -Can maintain a cells. -Achieves storage system.
-To minimize consistent - The fitness undershoot of
irreversible temperature function of 0.975% during the
capacity loss and among the battery optimized MFs is cooling subroutine.
degradation of pack. -Attains settling
not assessed in the
battery convergence curve.time of 32 min 13 s
performance. and 28 min 46 s
during heating and
heating.
FLC Ma et al. (2014) Cooling -To regulate the - Can balance SOC - The results are -Decreases the -Appropriate
temperature. and temperature verified through equalization time hardware
-To balance SOC. among cells under simulation study by 23% validation is
both charging and only. -Reduces the required to
discharging standard deviation demonstrate the
operation. of the final SOC by suitability of the
18.5%. proposed method.
FLC Umair Ali et al. Heating and Cooling -To develop an -Protects the -Unsatisfactory -Takes 9.76% less -The MFs of FLC can
(2018) efficient, real-time, battery from performance in time in complete be tuned properly
fast-charging overvoltage and over-cooling battery charging for better charging
approach under overheating. control. compared to the efficiency.
temperature -Fast-charging - Improves the SOC conventional - The methodology
feedback. operation. and charging rate method. can be
-Keeps the battery while avoiding - Charge capacity implemented using
within SOC limits. carbon emission can reach to 99.26% multi-cell battery
cost. SOC without cells.
causing any
performance
degradation.
FLC Rahman et al. Cooling - To increase the - Can control the -MFs of FLC are not -Saves energy by -Performance
(2015) battery life span temperature of the optimized. 17.69% compared investigation at
battery pack within -Generation of to air cooling higher
20e40 C during fuzzy rules is a battery thermal temperatures can
both charging and tedious task. management. be carried out.
discharging mode. -The proposed
model can be tested
under different
drive cycles.
MPC Lopez-Sanz et al. Cooling -To control the -Fast response. - SOC, level of -Minimizes the -The costs,
(2017) temperature of -Ability to reduce charge and life high temperature durability,
high voltage thermal stress and cycle of batteries efficiently. complexity and
battery cells. energy are not considered. -Keeps the battery performance of the
consumption of - The battery pack temperature at a proposed method
components. under low voltage safer range. in TMS need to be
is not examined. - Decreases cooling investigated.
energy by 5%.
MPC Altaf et al. (2016) Cooling -To prevent non- -Can control the - The balancing -Maintains the - Experimental
uniform aging. temperature with between SOC and balance between verification is
-To reduce thermal incomplete driving over-cooling is not over-heating and required for large
and SOC information carried out. SOC. battery packs
imbalances. -Both voltage -The controller is - Has negligible - Further studies
control and suitable for hot voltage error can be conducted
balancing control environment only. (<3.2%) under on battery module
can be achieved. model size and electronic
uncertainties and component losses.
cell temperature
variation.
Observer-based He and Ma, (2015) Cooling - To minimize the -Ability to track the -Suffers from the -Non-uniformity in -A comprehensive
controller temperature and cell core slow response in temperature experimental
maintain the temperature comparison to among cells is demonstration is
colling flow among accurately. fuzzy. decreased from 4.2 needed to verify the
cells. to 1.0 C controller’s
-Cooling flow effectiveness.
consumption is
reduced by 38%.
19
Nevertheless, the collection of large amounts of data with enough prediction, they have some negatives concerning computational
variety is time consuming and laborious task. In general, the data is complexity and long training interval. The integration of optimi-
collected through experiments at a sampling frequency of 1 Hz. The zation may lead to unsatisfactory outcomes if the convergence,
data length among EV drive cycles varies with different shapes of search capability, dimension settings are not addressed properly.
voltage and current values (Hannan et al., 2020b). For instance, the Therefore, the selection of suitable optimization technique is vital
one complete cycle for FUDS, DST, BJDST and US06 is computed to to attain acceptable results in prediction and control operation in
be 1372 s, 360 s, 916 s and 600 s, respectively (Zheng et al., 2016). BMS.
Since the training accuracy of intelligent algorithms needs a large
pool of data, thus multiple repetitions of EV drive cycles are 4.2.5. Algorithm verification uncertainties
employed for data preparation (Yang et al., 2020). A large amount of The performance of intelligent algorithms could diverge due to
data can support to achieve better solutions, however, it may result the presence of uncertainties such as erroneous initial state value,
in computational burden and extend the training time leading to equipment error, electromagnetic interference (EMI) noise, EV
over-fitting issues (Najafabadi et al., 2015). Thus, special attention is drive cycle patterns, temperatures and many unknown factors. In
required on data abundance and data variety issues. practical cases, the initial SOC may not be 100%. Hence, the accuracy
of the state estimation in BMS can be evaluated under different
4.2.2. Data integrity initial SOC values. For instance, the performance of GRU was
Data integrity is another barrier in implementing intelligent checked under 100%, 80%, 60%, 40% and 20% initial SOC (Yang et al.,
methods in a real-world environment. A few of the recognized 2019a). Moreover, the battery measurements may be affected by
automotive research groups (CALCE, 2017; Kollmeyer, 2018; low sensor precision, EMI and noise effects (Kang et al., 2014). Thus,
Kollmeyer et al., 2020; Y. Wang et al., 2017a) have released the high- different types of noises can be added to check the performance
quality publicly available battery dataset which contains a fixed accuracy of intelligent algorithms. For example, Cui et al. (2018)
pattern of charge-discharge current ensuring the various protocols verified the adaptability and robustness of WNN based SOC esti-
of EV drive cycles. The experiments to collect the various EV drive mation approach with the presence of bias noises and random
cycle data are carried out in the laboratory environment using the noises. Furthermore, numerous EV drive cycles are employed to
recommended charge/discharge current rates and temperature assess the adaptability and robustness of the WNN algorithm. It is
conditions. The obtained EV drive cycles are the simulated data and reported that the accuracy of SOC was varying due to different
their current and voltage profiles would not be as similar as data patterns of current and voltage measurements in EV drive cycle.
under the real-world environment. Hence, the verification of Another study by Li et al. (2019a) validated the GRU algorithm for
intelligent algorithms under practical cases needs further SOC estimation under various EV drive cycles, indicating MAE of
investigation. 2.12%, 1.13%, 0.71% and 1.01% in HWFET, LA92, UDDS and US06,
respectively. Moreover, the temperature has a huge impact on
4.2.3. Suitable structure and hyperparameter adjustment battery state estimation. Leng et al. (2015) found that the capacity
The effectiveness of intelligent algorithms relies heavily on of lithium-ion battery started decreasing when the temperature
input features, network architecture, training algorithms, activation increased above 45 C. Chemali et al. (2018) applied the DFNN in
functions and hyperparameter adjustment. So far, lots of efforts battery SOC estimation of BMS under wide-ranging temperatures
have been made by different researchers to design a suitable at 20 C, 10 C, 0 C, 10 C, and 25 C. The results illustrated the
structure for intelligent algorithms. Accordingly, the several varying MAE between 1.4% and 2.2% in HWFET drive cycle. There-
hyperparameters are adjusted to develop an improved framework fore, the verification of intelligent algorithms in BMS under diverse
for intelligent algorithms including HLs, HNs, learning rate, weight, EV drive cycles and temperatures is essential toward accurate and
bias, membership function, timestep, iterations, batch size, sam- robust state estimation in BMS.
pling interval etc. Besides, various training and activation functions
are employed to execute intelligent methods including Sigmoid, 4.3. Battery related issues
Tangent, Gaussian, Radial basis kernel, Gradient descent,
Levenberg-Marquardt, Logistic regression, backpropagation etc. The accuracy of state estimation in BMS could be deviated due to
The appropriate selection of hyperparameters and functions de- the various battery related issues such as battery chemistry, aging
creases the computational complexity resulting in the reduction in cycle, thermal runway, capacity fading, charging and balancing
data underfitting and overfitting issues. In the majority cases, the problems. The accuracy of battery state estimation under different
intelligent models are configured by tuning the hyperparameters battery chemistries could be different even if the same intelligent
and functions based on trial and error method, which leads to loss algorithm is applied. For instance, Lithium Iron Phosphate (LiFP)
of human energy and substantial time. Therefore, suitable design batteries exhibit cost-effectiveness and outstanding safety features
and hyperparameter adjustment of intelligent algorithms are the while Lithium Titanate (LiTO) batteries offer a long lifespan (Zhang
key issues to be explored. et al., 2018c). Chaoui and Ibe-Ekeocha (2017) assessed the perfor-
mance of recurrent NARX algorithm in SOC estimation and the
4.2.4. Optimization integration issues results indicated RMSE of 0.53% and 0.7% in LiFP and LiTO batteries,
The optimization integration into intelligent methods and con- respectively.
trol schemes is a challenging and laborious task. Numerous opti- Battery aging is another cause to degrade the accuracy of
mization methods could be combined with intelligent approaches. intelligent algorithms and controller strategies in BMS. Leng et al.
However, the performance of optimization varies from one algo- (2015) reported that the rise of aging cycles resulted in various is-
rithm to another in terms of execution time and convergence sues such as degradation of internal resistance, capacitance,
speed. Besides, optimization algorithms are structured using a lot of structure variation of the anode, cathode material and growth of
parameters and operational stages. The parameter initialization solid electrolyte interphase (SEI) thickness, leading to a variation in
and operational loop execution in optimization need depth battery SOC, SOH and RUL prediction in BMS. For instance, Li et al.
knowledge and human expertise. Although the inclusion of opti- (2016) validated the GPM method for RUL prediction using NASA
mizations into intelligent methods in BMS has demonstrated sub- PCoE data and computed RMSE of 1.58% and 1.3% in 60 and 80
stantial contributions in terms of accuracy, robustness and efficient cycles, respectively. Another study by Li et al. (2014b) predicted
20
SOH using RVM method and reported RMSE of 0.17%, 0.24%, 0.36% noise caused by EMI and sensor flaws. Thus, the advanced sensing
and 0.22% in 500, 600, 700 and 800 aging cycles, respectively. technology-based BTS is required to reduce the influence of the
The performance of intelligent algorithms and control schemes measurement noise and accordingly elevate the estimation accu-
in EVs could hamper due to the high battery temperature. The racy. Various methods are introduced to reduce noise impacts
thermal runaway is a serious concern that can be occurred by including wavelength transform, Butterworth filter, bias compen-
mechanical and electrical abuse resulting in lithium plating, over- sating method, adaptive forgetting variant and recursive total least
charge, internal short circuit, exothermic reactions and collision squares method. Wei et al. (2021) integrated bilinear principle and
problems (Zhang et al., 2018a). The growth of SOC and high aging instrumental variable (IV) to evaluate the impact of measurement
cycles are the root cause of thermal runaway (Galushkin et al., noises on model parameters and SOC estimation. The IV estimator
2018). The negative electrode, SEI layer and electrolyte begin to was employed to co-estimate the model parameters and noise
decompose when the temperature is over 90 C (Lipu et al., 2020a). variances. A two-stage least squares algorithm was utilized to solve
Hence, various studies are carried out on thermal runway and the co-estimation problem. The Luenberger observer was com-
modeling including analytical model (An et al., 2019), 2D model bined with the parameterization method to determine SOC online.
(Bugryniec et al., 2020), 3D model (Feng et al., 2016) and kinetics The simulations and experiments results confirmed the model
analysis (Ren et al., 2018). superiority with high noise immunity. Another study by (Wei et al.,
The battery capacity and power fading are other responsible 2020) concentrated on the noise-immune and unbiased model
factors in deteriorating the performance of intelligent algorithms parameter determination for lithium-ion batteries. A novel two-
and controller schemes in BMS. The capacity loss of the battery step method integrating variable projection algorithm (VPA) and
starts when the transformation of the battery active material be- least squares (LS) was employed to co-estimate the noise variances
gins under any discharge rates (Ramadass et al., 2002). In contrast, and unbiased model parameters. The computational efficiency of
the power rate capability is noticed when the rise of battery the proposed model was further enhanced using a numerical
impedance and decrease in operating voltage are occurred (Panchal solver. Both simulations and experiments demonstrated excellent
et al., 2017). A study by (Amine et al., 2005) reported that the ca- outcomes in term of parameterization accuracy and noise
pacity of the battery is reduced when the temperature rises above tolerance.
45 C under high aging cycles. Another study mentioned that,
battery capacity loss takes place when the voltage climbs up above 5.2. State monitoring for the battery pack in EVs
the maximum value, as suggested by the battery manufacturer
(Takei et al., 2001). In the majority studies, the single lithium-ion battery is
The charging method of lithium-ion batteries is crucial to obtain considered for SOC, SOE, SOH, and RUL prediction as well as fault
a good balance among heat, efficiency, lifespan and health. A slow diagnosis and TMS. However, EV is designed using a battery pack
charging method could hinder the widespread acceptance of EVs consisting of hundreds of lithium-ion battery cells. The accuracy of
while the fast-changing approach generates heat and reduces a single battery would not be as same as the accuracy of the battery
lifespan (Yin et al., 2019). On the other hand, charge imbalance pack due to the unbalancing issues that occur after repeated
among batteries is noted due to the variation in physical property, charge/discharge cycles. Although various controllers have been
degradation in battery aging and capacity, material defects, developed to address the above issues, further exploration is
manufacturing technology and tolerances (Diao et al., 2019). The required in improving their performance with regard to cost, size,
charge balancing issues of the battery results in divergence in power loss, efficiency, execution, equalization time, and voltage/
battery state estimation as well as affect the safety mechanism of current stress. Awadallah and Venkatesh (2016) proposed an
BMS (Turksoy et al., 2020). Hence, an effective charge balancing adaptive neuro-fuzzy inference system (ANFIS) algorithm to
approach is required to elevate the performance of intelligent compute SOC for a lithium-ion battery pack consisting of 36 cells
algorithms. connected through a series-parallel combination. The battery SOC
Aside from the abovementioned issues, the battery state esti- was evaluated at the cell level and then combined to calculate SOC
mation, temperature control and fault diagnosis in BMS can be for the battery pack. Sun et al. (2016) proposed RBFNN based SOC
influenced by the battery model, battery self-discharge, hysteresis, algorithm framework to determine SOC for a battery pack with
and charge/discharge current rate. twelve lithium-ion polymer batteries connected in series. Deng
et al. (2020) introduced a data-driven technique for SOC estima-
5. Future outlook and directions for BMS tion of lithium-ion battery pack addressing the cell inconsistencies
and data complexity through correlation and principal component
Based on the comprehensive study and analysis of available analysis.
literature of intelligent approaches and various control schemes of
BMS toward state estimation, thermal management, fault diagnosis 5.3. Algorithm hybridization
and battery balancing, the specific opportunities for future works
are suggested which are discussed in the following sections. Presently, the hybridization of intelligent algorithms has illus-
trated superiority over a single intelligent algorithm in improving
5.1. Development of advanced sensing equipment the prediction and control performance. Generally, hybridization is
formed by combining intelligent methods with the optimization
The monitoring of battery SOC, SOE, SOH, and RUL is carried out technique. For instance, Hannan et al. (2020b) combined recurrent
using different battery testing systems (BTSs) such as Digatron NARX algorithms with LSA to enhance the accuracy of SOC esti-
(Chemali et al., 2017), Arbin BT 2000 (Sahinoglu et al., 2018), DAQ mation. Chang (2013) applied GA and orthogonal least-squares al-
(Chaoui et al., 2017) and NEWARE BTS 4000 (Wang et al., 2014b). gorithms to optimize the RBFNN model toward accurate and robust
Nevertheless, the measurements by BTS may not be accurate due to SOC estimation. PSO was integrated with BPNN (Zhang et al., 2018e)
EMI, noise impacts, and equipment error (B. Xia et al., 2018b). In and WNN (Xia et al., 2018a), respectively to optimize the hyper-
addition, the majority of state estimation intelligent algorithms are parameters and improve SOC estimation accuracy. Nevertheless,
validated under laboratory environments which may not achieve the inappropriate combination of hybridization contributes to
the desired outcomes in real applications due to a large amount of complex configuration and hence leads to mathematical
21
complexity and undesirable outcomes. Therefore, further studies SOH, and RUL prediction along with thermal runway and fault
are required and practicability issues need to be considered to diagnosis throughout the entire life can be monitored and saved in
develop an efficient hybrid algorithm. the cloud storage. Subsequently, the battery monitoring and control
center will pre-process the data, carry out the analysis and deliver
5.4. Training performance enhancement effective decisions for future performance improvement. Haldar
et al. (2020) focused on real-time battery monitoring and man-
One of the major challenges to execute the intelligent methods agement in EVs based on the internet of things (IOT). The authors
is the training accuracy and duration specifically for the deep developed a real-time wireless battery management system to
learning algorithm. Chemali et al. (2018) carried out the training monitor the battery discharging behavior and health status in EVs.
operation of deep learning algorithms offline with thousands of Sivaraman and Sharmeela (2020) developed an IOT based BMS to
battery data in around 50e60 h. In another study (Chemali et al., monitor SOC, SOH as well as control the charge imbalance and fault
2017), the computational time to train the battery data was esti- diagnosis. Kim et al. (2018b) introduced a cloud-based battery
mated at around 4e9 h. Yang et al. (2019a) evaluated the training monitoring system for the lithium-ion battery storage system. The
performance of the GRU network in SOC estimation under different proposed system was validated by cyber-physical testbench using
epoch numbers and results indicated training time of 10.03 and Raspberry Pi3 IOT boards. The results illustrated that SOC and ca-
79.14 min in 1000 and 7000 epochs, respectively. The training time pacity were monitored accurately by the condition monitoring al-
could be extended if the intelligent algorithm validation is con- gorithm while the fault diagnosis was assessed by the data mining
ducted under more HLs, HNs and different temperatures. Therefore, algorithm.
further attention is essential to the selection of suitable activation
functions, training algorithms, and hyperparameters to expedite 5.7. Multiscale and co-estimation process
the training process. In recent times, deep learning-based SOC
estimation has received wide attention due to the high configura- The battery state estimation using multi-scale and co-
tion processor with advanced graphics processing units (GPUs) and estimations plays a vital role in improving the efficiency and
matured frameworks such as TensorFlow which has helped to reducing the computation of BMS. For example, battery SOC and
reduce the training time (Sahinoglu et al., 2018). For instance, Song SOE vary quickly with current fluctuation while the SOH and RUL
et al. (2019) used two GeForce GTX 1080 Ti GPUs to train the CNN- change with a capacity variation. Thus, different time scales are
LSTM framework for SOC estimation of lithium-ion batteries. Li needed to monitor the different states of the battery. Besides, the
et al. (2019a) applied NVIDIA GeForce GTX 1070Ti GPU and 16 GB majority of the literature has suggested the methodology for single
RAM to execute the training operation of GRU in SOC estimation. state estimation. The co-estimation or joint estimation combing
two or three states has been very limited so far. A few investigations
5.5. Real-time embedded prototype design have been performed to shed some light on utilizing the intelligent
algorithms for joint estimation. Hence, further studies are required
To date, different experimental tests have been performed to on the co-estimation approach under different spatial and temporal
validate the effectiveness of intelligent approaches toward state scales. Chaoui et al. (2017) proposed an intelligent algorithm for
estimation, thermal management, and fault diagnosis. However, two states (SOC and SOH) joint estimation for lithium-ion batteries.
the implementation of intelligent methods in real-time BMS with The battery SOC was estimated using the TDNN algorithm while the
small memory units and low computational cost has not been SOH was assessed using the normalized data in SOC learning.
explored comprehensively. Hence, further investigation is needed Another work by (Chaoui and Ibe-Ekeocha, 2017) presented RNN
to develop an embedded prototype system for real-time state based co-estimation method for SOC and SOH estimation in
estimation and control operation of BMS. There have been a few lithium-ion batteries. SOH was predicted using the present and past
prominent researches on real-time state estimation in BMS using values of current and voltage in SOC learning along with four
intelligent algorithms. Chen et al. (2019) validated the FNN based milestones of battery lifespan under 100, 200, 1600 and 1800 cy-
SOC estimation method on the hardware-in-the Loop (HIL) exper- cles. The future outlook and opportunities for advanced BMS are
imental platform in real-time. The HIL test bench was designed shown in Fig. 10.
using a DC source, a current senor, battery monitoring unit, a host
computer, a battery control unit and a CAN analyzer. The SOC error 6. Conclusion
and capacity error of the proposed method in HIL test was obtained
to be 2% and 19.7%, respectively. Singh et al. (2020) developed a HIL In recent decades, numerous analysis and evolution have been
experimental setup to validate the ANFIS based SOC estimation in carried out on intelligent algorithms and control strategies of BMS
real-time using dSpace MicroLabBox hardware controller. The HIL in EVs due to the big data availability, powerful computation pro-
results were very close to the simulated results which demon- cessor, and high capacity data storage devices. As a first contribu-
strated the suitability of the proposed model in real-time EV tion, this review explores the current progress of intelligent
applications. algorithms toward SOC, SOE, SOH, and RUL prediction, concen-
trating on structure, input features, pros, cons, and estimation error.
5.6. Integrated with big data platform The review has revealed that the intelligent algorithms have illus-
trated better performance in battery state estimation in terms of
The accuracy and robustness of intelligent algorithms of BMS accuracy, adaptability, robustness, and efficiency. As a second
under real-world environments can be further enhanced through contribution, this survey examines the various control techniques
cloud storage, cloud computing and big data platforms. The big data in temperature control, fault identification and prognosis and bat-
technology helps in executing intelligent methods with a large tery charge equalization, given importance on type, objectives,
memory device, computation and analysis. The voltage, current, outcomes, advantages, disadvantages and research limitations. It is
temperature, and other information of EVs are continuously found that FLC and MPC can map a time-varying and complex
transferred to the big data platform and accordingly, intelligent battery system under constraints and uncertainties. As a third
approaches can be trained under real-time tests and deliver more contribution, the review highlights the key concerns of intelligent
precise outcomes. The battery state estimation such as SOC, SOE, algorithms and controller executions in BMS including data variety,
22
Fig. 10. Future trends for advanced BMS in EVs.
appropriate structure, selection of input features, hyperparameter Acknowledgment

optimization, uncertainties, computation complexity. As a fourth
contribution, the review discusses various opportunities for future This research was funded by Universiti Kebangsaan Malaysia
research works toward further technological development of under research grant number GGPM-2020-006. This research also
intelligent approaches and control schemes of BMS in EV applica- received partial financial support from the Ministry of Higher Ed-
tions. This review proposes the future trends of BMS emphasizing ucation, Malaysia under grant number 20190101LRGS provided by
advanced sensing equipment, embedded system, high- Universiti Tenaga Nasional, Malaysia.
performance processor, hybridizing algorithms, multi-state esti-
mation, model validation using battery module/pack and cloud
References
computing technology.
In summary, the critical analysis and key information attained Aaldering, L.J., Leker, J., Song, C.H., 2019. Analysis of technological knowledge stock
from this review would be beneficial for the automobile engineers and prediction of its future development potential: the case of lithium-ion
batteries. J. Clean. Prod. 223, 301e311.
as well as the EV industry in developing and executing advanced
Afzal, A., Ramis, M.K., 2020. Multi-objective optimization of thermal performance in
BMS for EV applications. Thus, further research works on BMS by battery system using genetic and particle swarm algorithm combined with
intelligent algorithms and controller schemes will not only elevate fuzzy logics. J. Energy Storage 32, 101815.
Altaf, F., Egardt, B., Mårdh, L.J., 2016. Load management of modular battery using
the battery performance, lifespan but also ensure safe, reliable
model predictive control: thermal and state-of-charge balancing. IEEE Trans.
operation of EV resulting in substantial growth of the battery and Contr. Syst. Technol. 25, 47e62.
EV market. The expansion of the EV and related battery market can
Alvarez Anto n, J.C., García Nieto, P.J., de Cos Juez, F.J., Sa
nchez Lasheras, F., Gonz
alez
Vega, M., Roquen ~ í Gutie
rrez, M.N., 2013. Battery state-of-charge estimator using
help to achieve a pathway for sustainable development goals
the SVM technique. Appl. Math. Model. 37, 6244e6253.
relating to clean energy, emission reduction, employment oppor- Amine, K., Liu, J., Belharouak, I., 2005. High-temperature storage and cycling of C-
tunities and economic development. Nevertheless, several short- LiFePO4/graphite Li-ion cells. Electrochem. Commun. 7, 669e673.
comings could hamper the wide adoption of EVs such as battery An, Z., Shah, K., Jia, L., Ma, Y., 2019. Modeling and analysis of thermal runaway in Li-
ion cell. Appl. Therm. Eng. 160, 113960.
price and performance degradation, low driving mileage, poor Awadallah, M.A., Venkatesh, B., 2016. Accuracy improvement of SOC estimation in
vehicle features and ineffective marketing policies. Therefore, a lithium-ion batteries. J. Energy Storage 6, 95e104.
long-term future initiative is necessary to enhance the EV perfor- Balasingam, B., Ahmed, M., Pattipati, K., 2020. Battery management system-
sdchallenges and some solutions. Energies 13, 2825.
mance with regard to accurate battery monitoring and controlling Bandhauer, T.M., Garimella, S., Fuller, T.F., 2011. A critical review of thermal issues in
schemes development, investments, incentives, effective policies, lithium-ion batteries. J. Electrochem. Soc. 158, R1eR25.
carbon emission reduction, global collaboration and sustainable Baronti, F., Roncella, R., Saletti, R., 2014. Performance comparison of active balancing
techniques for lithium-ion batteries. J. Power Sources 267, 603e609.
development.
Bi, Y., Choe, S.Y., 2020. An adaptive sigma-point Kalman filter with state equality
constraints for online state-of-charge estimation of a Li(NiMnCo)O2/Carbon
battery using a reduced-order electrochemical model. Appl. Energy 258,
Declaration of competing interest 113925.
Bian, C., He, H., Yang, S., 2020a. Stacked bidirectional long short-term memory
The authors declare that they have no known competing networks for state-of-charge estimation of lithium-ion batteries. Energy 191,
116538.
financial interests or personal relationships that could have Bian, C., He, H., Yang, S., Huang, T., 2020b. State-of-charge sequence estimation of
appeared to influence the work reported in this paper. lithium-ion battery based on bidirectional long short-term memory encoder-
23
decoder architecture. J. Power Sources 449, 227558. using state of charge estimation: an IOT based approach. In: Proceedings of the
Bugryniec, P.J., Davidson, J.N., Brown, S.F., 2020. Computational modelling of ther- IEEE International Conference on Emerging Trends on Sustainable Technology
mal runaway propagation potential in lithium iron phosphate battery packs. and Engineering Applications, pp. 1e5.
Energy Rep. 6, 189e197. Han, X., Ouyang, M., Lu, L., Li, J., 2014. Cycle life of commercial lithium-ion batteries
CALCE, 2017. Lithium-ion battery experimental data, 8.23.18. https://web.calce.umd. with lithium titanium oxide anodes in electric vehicles. Energies 7, 4895e4909.
edu/batteries/data.htm. Hannan, M.A., Al-Shetwi, A., Begum, R.A., Young, S.E., Hoque, M.M., Ker, P.,
Chang, W.-Y., 2013. Estimation of the state of charge for a LFP battery using a hybrid Mansur, M., Alzaareer, K., 2020a. The value of thermal management control
method that combines a RBF neural network, an OLS algorithm and AGA. Int. J. strategies for battery energy storage in grid decarbonization: issues and rec-
Electr. Power Energy Syst. 53, 603e611. ommendations. J. Clean. Prod. 276, 124223.
Chao, W., Chunbo, Z., Jinlei, S., Jianhu, J., 2017. Fault mechanism study on Li-ion Hannan, M.A., Hoque, M.M., Hussain, A., Yusof, Y., Ker, P., 2018a. State-of-the-Art
battery at over-discharge and its diagnosis approach. IET Electr. Syst. Transp. and energy management system of lithium-ion batteries in electric vehicle
7, 48e54. applications: issues and recommendations. IEEE Access 6, 19362e19378.
Chaoui, H., Ibe-Ekeocha, C.C., 2017. State of charge and state of health estimation for Hannan, M.A., Lipu, M.S.H., Hussain, A., Ker, P.J., Mahlia, T.M.I., Mansor, M., Ayob, A.,
lithium batteries using recurrent neural networks. IEEE Trans. Veh. Technol. 66, Saad, M.H., Dong, Z.Y., 2020b. Toward enhanced state of charge estimation of
8773e8783. lithium-ion batteries using optimized machine learning techniques. Sci. Rep. 10,
Chaoui, H., Ibe-Ekeocha, C.C., Gualous, H., 2017. Aging prediction and state of charge 4687.
estimation of a LiFePO4 battery using input time-delayed neural networks. Elec. Hannan, M.A., Lipu, M.S.H., Hussain, A., Saad, M.H., Ayob, A., 2018b. Neural network
Power Syst. Res. 146, 189e197. approach for estimating state of charge of lithium-ion battery using back-
Chaturvedi, N.A., Klein, R., Christensen, J., Ahmed, J., Kojic, A., 2010. Algorithms for tracking search algorithm. IEEE Access 6, 10069e10079.
Advanced Battery-Management Systems: modeling, estimation, and control Hannan, M.A., Young, Y.S., Hoque, M.M., Ker, P.J., Uddin, M.N., 2019. Lithium ion
challenges for lithium-ion batteries. IEEE Control Syst 30, 49e68. battery thermal management system using optimized fuzzy controller. In:
Chemali, E., Kollmeyer, P., Preindl, M., Ahmed, R., Emadi, A., 2017. Long short-term Proceedings of the IEEE Industry Applications Society Annual Meeting, pp. 1e9.
memory-networks for accurate state of charge estimation of Li-ion batteries. He, F., Ma, L., 2015. Thermal management of batteries employing active temperature
IEEE Trans. Ind. Electron. 65, 6730e6739. control and reciprocating cooling flow. Int. J. Heat Mass Tran. 83, 164e172.
Chemali, E., Kollmeyer, P.J., Preindl, M., Emadi, A., 2018. State-of-charge estimation He, H., Cao, J., Cui, X., 2020. Energy optimization of electric vehicle’s acceleration
of Li-ion batteries using deep neural networks: a machine learning approach. process based on reinforcement learning. J. Clean. Prod. 248, 119302.
J. Power Sources 400, 242e255. He, W., Williard, N., Chen, C., Pecht, M., 2014. State of charge estimation for Li-ion
Chen, C., Xiong, R., Yang, R., Shen, W., Sun, F., 2019. State-of-charge estimation of batteries using neural network modeling and unscented Kalman filter-based
Lithium-ion battery using an improved neural network model and extended error cancellation. Int. J. Electr. Power Energy Syst. 62, 783e791.
Kalman filter. J. Clean. Prod. 234, 1153e1164. He, W., Williard, N., Osterman, M., Pecht, M., 2011. Prognostics of lithium-ion bat-
Chen, X., Shen, W., Dai, M., Cao, Z., Jin, J., Kapoor, A., 2016a. Robust adaptive sliding- teries based on Dempster-Shafer theory and the Bayesian Monte Carlo method.
mode observer using RBF neural network for lithium-ion battery state of charge J. Power Sources 196, 10314e10321.
estimation in electric vehicles. IEEE Trans. Veh. Technol. 65, 1936e1947. Hendricks, C., Williard, N., Mathew, S., Pecht, M., 2015. A failure modes, mecha-
Chen, Y., Liu, X., Cui, Y., Zou, J., Yang, S., 2016b. A MultiWinding transformer cell-to- nisms, and effects analysis (FMMEA) of lithium-ion batteries. J. Power Sources
cell active equalization method for lithium-ion batteries with reduced number 297, 113e120.
of driving circuits. IEEE Trans. Power Electron. 31, 4916e4929. Hofmann, A., Uhlmann, N., Ziebert, C., Wiegand, O., Schmidt, A., Hanemann, T., 2017.
Chen, Z., Sun, M., Shu, X., Shen, J., Xiao, R., 2018. On-board state of health estimation Preventing Li-ion cell explosion during thermal runaway with reduced pres-
for lithium-ion batteries based on random forest. In: Proceedings of the IEEE sure. Appl. Therm. Eng. 124, 539e544.
International Conference on Industrial Technology, pp. 1754e1759. Hong, J., Wang, Z., Liu, P., 2017. Big-data-based thermal runaway prognosis of bat-
Cui, D., Xia, B., Zhang, R., Sun, Z., Lao, Z., Wang, W., Sun, W., Lai, Y., Wang, M., Cui, D., tery systems for electric vehicles. Energies 10, 919.
Xia, B., Zhang, R., Sun, Z., Lao, Z., Wang, W., Sun, W., Lai, Y., Wang, M., 2018. Hong, J., Wang, Z., Yao, Y., 2019. Fault prognosis of battery system based on accurate
A novel intelligent method for the state of charge estimation of lithium-ion voltage abnormity prognosis using long short-term memory neural networks.
batteries using a discrete wavelet transform-based wavelet neural network. Appl. Energy 251, 113381.
Energies 11, 995. Hu, X., Jiang, J., Cao, D., Egardt, B., 2016. Battery health prognosis for electric vehicles
Deng, Y., Ying, H., E, J., Zhu, H., Wei, K., Chen, J., Zhang, F., Liao, G., 2019. Feature using sample entropy and sparse Bayesian predictive modeling. IEEE Trans. Ind.
parameter extraction and intelligent estimation of the State-of-Health of Electron. 63, 2645e2656.
lithium-ion batteries. Energy 176, 91e102. Hu, X., Li, S.E., Jia, Z., Egardt, B., 2014. Enhanced sample entropy-based health
Deng, Z., Hu, X., Lin, X., Che, Y., Xu, L., Guo, W., 2020. Data-driven state of charge management of Li-ion battery for electrified vehicles. Energy 64, 953e960.
estimation for lithium-ion battery packs based on Gaussian process regression. Huang, D., Chen, Z., Zheng, C., Li, H., 2019a. A model-based state-of-charge esti-
Energy 205, 118000. mation method for series-connected lithium-ion battery pack considering fast-
Diao, W., Pecht, M., Liu, T., 2019. Management of imbalances in parallel-connected varying cell temperature. Energy 185, 847e861.
lithium-ion battery packs. J. Energy Storage 24, 100781. Huang, W., Abu Qahouq, J.A., 2015. Energy sharing control scheme for state-of-
Diao, W., Xue, N., Bhattacharjee, V., Jiang, J., Karabasoglu, O., Pecht, M., 2018. Active charge balancing of distributed battery energy storage system. IEEE Trans.
battery cell equalization based on residual available energy maximization. Appl. Ind. Electron. 62, 2764e2776.
Energy 210, 690e698. Huang, Z., Yang, F., Xu, F., Song, X., Tsui, K.-L., 2019b. Convolutional gated recurrent
Dong, G., Zhang, X., Zhang, C., Chen, Z., 2015. A method for state of energy esti- uniterecurrent neural network for state-of-charge estimation of lithium-ion
mation of lithium-ion batteries based on neural network model. Energy 90, batteries. IEEE Access 7, 93139e93149.
879e888. Hussain, S., Nengroo, S.H., Zafar, A., Kim, H.-J., Alvi, M.J., Ali, M.U., 2019. Towards a
Eddahech, A., Briat, O., Bertrand, N., Dele tage, J.Y., Vinassa, J.M., 2012. Behavior and smarter battery management system for electric vehicle applications: a critical
state-of-health monitoring of Li-ion batteries using impedance spectroscopy review of lithium-ion battery state of charge estimation. Energies 12, 446.
and recurrent neural networks. Int. J. Electr. Power Energy Syst. 42, 487e494. Jacob, R., Liu, M., Sun, Y., Belusko, M., Bruno, F., 2019. Characterisation of promising
EEA, 2017. Greenhouse gas emissions from transport in europe d European envi- phase change materials for high temperature thermal energy storage. J. Energy
ronment agency, 5.15.20. https://www.eea.europa.eu/data-and-maps/ Storage 24, 100801.
indicators/transport-emissions-of-greenhouse-gases/transport-emissions-of- Kang, L., Zhao, X., Ma, J., 2014. A new neural network model for the state-of-charge
greenhouse-gases-12. estimation in the battery degradation process. Appl. Energy 121, 20e27.
Feng, F., Hu, X., Liu, J., Lin, X., Liu, B., 2019. A review of equalization strategies for Kang, Y., Duan, B., Zhou, Z., Shang, Y., Zhang, C., 2019. A multi-fault diagnostic
series battery packs: variables, objectives, and algorithms. Renew. Sustain. method based on an interleaved voltage measurement topology for series
Energy Rev. 116, 109464. connected battery packs. J. Power Sources 417, 132e144.
Feng, X., Lu, L., Ouyang, M., Li, J., He, X., 2016. A 3D thermal runaway propagation Kim, T., Adhikaree, A., Pandey, R., Kang, D., Kim, M., Oh, C.Y., Back, J., 2018a. Outlier
model for a large format lithium ion battery module. Energy 115, 194e208. mining-based fault diagnosis for multiceli lithium-ion batteries using a low-
Feng, X., Sun, J., Ouyang, M., He, X., Lu, L., Han, X., Fang, M., Peng, H., 2014. Char- priced microcontroller. In: Proceedings of the IEEE Applied Power Electronics
acterization of large format lithium ion battery exposed to extremely high Conference and Exposition, pp. 3365e3369.
temperature. J. Power Sources 272, 457e467. Kim, T., Makwana, D., Adhikaree, A., Vagdoda, J., Lee, Y., 2018b. Cloud-based battery
Galushkin, N.E., Yazvinskaya, N.N., Galushkin, D.N., 2018. Mechanism of thermal condition monitoring and fault diagnosis platform for large-scale lithium-ion
runaway in lithium-ion cells. J. Electrochem. Soc. 165, A1303eA1308. battery energy storage systems. Energies 11, 125.
Gohoungodji, P., N’Dri, A.B., Latulippe, J.M., Matos, A.L.B., 2020. What is stopping the Klass, V., Behm, M., Lindbergh, G., 2014. A support vector machine-based state-of-
automotive industry from going green? A systematic review of barriers to green health estimation method for lithium-ion batteries under electric vehicle
innovation in the automotive industry. J. Clean. Prod. 277, 123524. operation. J. Power Sources 270, 262e272.
Guo, J., Zhang, X., Gu, F., Zhang, H., Fan, Y., 2020. Does air pollution stimulate electric Kollmeyer, P., 2018. Panasonic 18650PF Li-ion battery data, 4.13.20. https://doi.org/
vehicle sales? Empirical evidence from twenty major cities in China. J. Clean. 10.17632/wykht8y7tg.1.
Prod. 249, 119372. Kollmeyer, P., Vidal, C., Naguib, M., Skells, M., 2020. LG 18650HG2 Li-ion battery
Haddad Zarif, M., Charkhgard, M., Alfi, A., 2014. Hybrid state of charge estimation data, 4.13.20. https://data.mendeley.com/datasets/cp3473x7xv/3#folder-
for lithium-ion batteries: design and implementation. IET Power Electron. 7, 337ff5af-9640-43a4-9658-464a4478b02d.
2758e2764. Kong, X., Zheng, Y., Ouyang, M., Lu, L., Li, J., Zhang, Z., 2018. Fault diagnosis and
Haldar, S., Mondal, S., Mondal, A., Banerjee, R., 2020. Battery management system quantitative analysis of micro-short circuits for lithium-ion batteries in battery
24
packs. J. Power Sources 395, 358e368. Transportation Electrification Conference and Expo, Asia-Pacific (ITEC Asia-Pa-
Kumar, R.R., Alok, K., 2020. Adoption of electric vehicle: a literature review and cific), pp. 1e6.
prospects for sustainability. J. Clean. Prod. 253, 119911. Lombardi, W., Zarudniev, M., Lesecq, S., Bacquet, S., 2014. Sensors fault diagnosis for
Lai, X., He, L., Wang, S., Zhou, L., Zhang, Y., Sun, T., Zheng, Y., 2020. Co-estimation of a BMS. In: Proceedings of the European Control Conference, pp. 952e957.
state of charge and state of power for lithium-ion batteries based on fractional Lopez-Sanz, J., Ocampo-Martinez, C., Alvarez-Florez, J., Moreno-Eguilaz, M., Ruiz-
variable-order model. J. Clean. Prod. 255, 120203. Mansilla, R., Kalmus, J., Gr€ aeber, M., Lux, G., 2017. Nonlinear model predictive
Lamb, J., Orendorff, C.J., Steele, L.A.M., Spangler, S.W., 2015. Failure propagation in control for thermal management in plug-in hybrid electric vehicles. IEEE Trans.
multi-cell lithium ion batteries. J. Power Sources 283, 517e523. Veh. Technol. 66, 3632e3644.
Lelie, M., Braun, T., Knips, M., Nordmann, H., Ringbeck, F., Zappen, H., Sauer, D.U., Lu, L., Han, X., Li, J., Hua, J., Ouyang, M., 2013. A review on the key issues for lithium-
2018. Battery management system hardware concepts: an overview. Appl. Sci. ion battery management in electric vehicles. J. Power Sources 226, 272e288.
8, 534. Lyu, D., Ren, B., Li, S., 2019. Failure modes and mechanisms for rechargeable
Leng, F., Tan, C.M., Pecht, M., 2015. Effect of temperature on the aging rate of Li ion Lithium-based batteries: a state-of-the-art review. Acta Mech. 230, 701e727.
battery operating above room temperature. Sci. Rep. 5, 1e12. Ma, G., Zhang, Y., Cheng, C., Zhou, B., Hu, P., Yuan, Y., 2019. Remaining useful life
Li, C., Chen, Z., Cui, J., Wang, Y., Zou, F., 2014a. The lithium-ion battery state-of- prediction of lithium-ion batteries based on false nearest neighbors and a
charge estimation using random forest regression. In: Proceedings of the hybrid neural network. Appl. Energy 253, 113626.
Prognostics and System Health Management Conference, pp. 336e339. Ma, T., Serrano, B., Mohammed, O., 2014. Fuzzy logic based power and thermal
Li, C., Xiao, F., Fan, Y., 2019a. An approach to state of charge estimation of lithium- management system design for multi-cell lithium-ion battery bank protection
ion batteries based on recurrent neural networks with gated recurrent unit. and operation. In: Proceedings of the Clemson University Power Systems
Energies 12, 1592. Conference, pp. 1e5.
Li, H., Pan, D., Chen, C.L.P., 2014b. Intelligent prognostics for battery health moni- Ma, Y., Duan, P., Sun, Y., Chen, H., 2018. Equalization of lithium-ion battery pack
toring using the mean entropy and relevance vector machine. IEEE Trans. Syst. based on fuzzy logic control in electric vehicle. IEEE Trans. Ind. Electron. 65,
Man, Cybern. Syst. 44, 851e862. 6762e6771.
Li, L., Wang, P., Chao, K.-H., Zhou, Y., Xie, Y., 2016. Remaining useful life prediction McCurlie, L., Preindl, M., Emadi, A., 2017. Fast model predictive control for redis-
for lithium-ion batteries based on Gaussian processes mixture. PloS One 11, tributive lithium-ion battery balancing. IEEE Trans. Ind. Electron. 64,
1e13. 1350e1357.
Li, P., Zhang, Z., Xiong, Q., Ding, B., Hou, J., Luo, D., Rong, Y., Li, S., 2020f. State-of- Merhy, G., Nait-Sidi-Moh, A., Moubayed, N., 2020. Control, regulation and optimi-
health estimation and remaining useful life prediction for the lithium-ion zation of bidirectional energy flows for electric vehicles’ charging and dis-
battery based on a variant long short term memory neural network. J. Power charging. Sustain. Cities Soc. 57, 102129.
Sources 459, 228069. Muddappa, V.K.S., Anwar, S., 2014. Electrochemical model based fault diagnosis of
Li, W., Fan, Y., Ringbeck, F., Jo €st, D., Han, X., Ouyang, M., Sauer, D.U., 2020e. Elec- Li-Ion battery using fuzzy logic. In: Proceedings of the American Society of
trochemical model-based state estimation for lithium-ion batteries with Mechanical Engineers (ASME), pp. 1e9.
adaptive unscented Kalman filter. J. Power Sources 476, 228534. Najafabadi, M.M., Villanustre, F., Khoshgoftaar, T.M., Seliya, N., Wald, R.,
Li, X., Dai, K., Wang, Z., Han, W., 2020a. Lithium-ion batteries fault diagnostic for Muharemagic, E., 2015. Deep learning applications and challenges in big data
electric vehicles using sample entropy analysis method. J. Energy Storage 27, analytics. J. Big Data 2, 1e21.
101121. Ng, S.S.Y., Xing, Y., Tsui, K.L., 2014. A naive Bayes model for robust remaining useful
Li, X., Wang, Z., 2018. A novel fault diagnosis method for lithium-Ion battery packs life prediction of lithium-ion battery. Appl. Energy 118, 114e123.
of electric vehicles. Meas. J. Int. Meas. Confed. 116, 402e411. Nguyen, N., Oruganti, S.K., Na, K., Bien, F., 2014. An adaptive backward control
Li, X., Yuan, C., Wang, Z., 2020b. State of health estimation for Li-ion battery via battery equalization system for serially connected lithium-ion battery packs.
partial incremental capacity analysis based on support vector regression. En- IEEE Trans. Veh. Technol. 63, 3651e3660.
ergy 203, 117852. Nikolian, A., Firouz, Y., Gopalakrishnan, R., Timmermans, J.-M., Omar, N., van den
Li, X., Yuan, C., Wang, Z., 2020c. Multi-time-scale framework for prognostic health Bossche, P., van Mierlo, J., 2016. Lithium ion batteriesddevelopment of
condition of lithium battery using modified Gaussian process regression and advanced electrical equivalent circuit models for nickel manganese cobalt
nonlinear regression. J. Power Sources 467, 228358. lithium-ion. Energies 9, 360.
Li, X., Zhang, L., Wang, Z., Dong, P., 2019b. Remaining useful life prediction for Niu, G., 2016. Data-driven Technology for Engineering Systems Health Manage-
lithium-ion batteries based on a hybrid model combining the long short-term ment: Design Approach, Feature Construction, Fault Diagnosis, Prognosis,
memory and Elman neural networks. J. Energy Storage 21, 510e518. Fusion and Decision. Springer.
Li, X., Yuan, C., Li, Xiaohui, Wang, Z., 2020d. State of health estimation for Li-Ion Pan, H., Lü, Z., Wang, H., Wei, H., Chen, L., 2018. Novel battery state-of-health online
battery using incremental capacity analysis and Gaussian process regression. estimation method using multiple health indicators and an extreme learning
Energy 190, 116467. machine. Energy 160, 466e477.
Lin, Q., Wang, J., Xiong, R., Shen, W., He, H., 2019. Towards a smarter battery Panchal, S., Mcgrory, J., Kong, J., Fraser, R., Fowler, M., Dincer, I., Agelin-Chaab, M.,
management system: a critical review on optimal charging methods of lithium 2017. Cycling degradation testing and analysis of a LiFePO4battery at actual
ion batteries. Energy 183, 220e234. conditions. Int. J. Energy Res. 41, 2565e2575.
Lin, W.Y., Hsiao, M.C., Wu, P.C., Fu, J.S., Lai, L.W., Lai, H.C., 2020. Analysis of air quality Peng, Y., Hou, Y., Song, Y., Pang, J., Liu, D., 2018. Lithium-ion battery prognostics with
and health co-benefits regarding electric vehicle promotion coupled with po- hybrid Gaussian process function regression. Energies 11, 1420.
wer plant emissions. J. Clean. Prod. 247, 119152. Qu, J., Liu, F., Ma, Y., Fan, J., 2019. A neural-network-based method for RUL prediction
Lipu, M.S.H., Hannan, M.A., Hussain, A., Ayob, A., Saad, M.H.M., Karim, T.F., and SOH monitoring of lithium-ion battery. IEEE Access 7, 87178e87191.
How, D.N.T., 2020a. Data-driven state of charge estimation of lithium-ion bat- Rahman, A., Nur, F., Hawlader, M.N.A., Afroz, R., 2015. Fuzzy controlled evaporative
teries: algorithms, implementation factors, limitations and future trends. battery thermal management system for EV/HEV. Int. J. Electr. Hybrid Veh.
J. Clean. Prod. 277, 124110. (IJEHV) 7, 22e39.
Lipu, M.S.H., Hannan, M.A., Hussain, A., Ayob, A., Saad, M.H.M., Muttaqi, K.M., Ramadass, P., Haran, B., White, R., Popov, B.N., 2002. Capacity fade of Sony 18650
2020b. State of charge estimation in lithium-ion batteries: a neural network cells cycled at elevated temperatures: Part I. Cycling performance. J. Power
optimization approach. Electronics 9, 1546. Sources 112, 606e613.
Lipu, M.S.H., Hannan, M.A., Hussain, A., Saad, M.H.M., Ayob, A., Uddin, M., 2019. Ren, D., Liu, X., Feng, X., Lu, L., Ouyang, M., Li, J., He, X., 2018. Model-based thermal
Extreme learning machine model for state of charge estimation of lithium-ion runaway prediction of lithium-ion batteries from kinetics analysis of cell
battery using gravitational search algorithm. IEEE Trans. Ind. Appl. 55, components. Appl. Energy 228, 633e644.
4225e4234. Ringbeck, F., Garbade, M., Sauer, D.U., 2020. Uncertainty-aware state estimation for
Liu, G., Ouyang, M., Lu, L., Li, J., Han, X., 2014a. Analysis of the heat generation of electrochemical model-based fast charging control of lithium-ion batteries.
lithium-ion battery during charging and discharging considering different J. Power Sources 470, 228221.
influencing factors. J. Therm. Anal. Calorim. 116, 1001e1010. Sahinoglu, G.O., Pajovic, M., Sahinoglu, Z., Wang, Y., Orlik, P.V., Wada, T., 2018.
Liu, J., Chen, Y., Fathy, H.K., 2017a. Nonlinear model-predictive optimal control of an Battery state-of-charge estimation based on regular/recurrent Gaussian process
active cell-to-cell lithium-ion battery pack balancing circuit. IFAC-PapersOnLine regression. IEEE Trans. Ind. Electron. 65, 4311e4321.
50, 14483e14488. Shang, Y., Cui, N., Duan, B., Zhang, C., 2018. Analysis and optimization of star-
Liu, P., Sun, Z., Wang, Z., Zhang, J., 2018. Entropy-based voltage fault diagnosis of structured switched-capacitor equalizers for series-connected battery strings.
battery systems for electric vehicles. Energies 11, 136. IEEE Trans. Power Electron. 33, 9631e9646.
Liu, X., Wu, J., Zhang, C., Chen, Z., 2014b. A method for state of energy estimation of Shang, Y., Xia, B., Lu, F., Zhang, C., Cui, N., Mi, C.C., 2017a. A switched-coupling-
lithium-ion batteries at dynamic currents and temperatures. J. Power Sources capacitor equalizer for series-connected battery strings. IEEE Trans. Power
270, 151e157. Electron. 32, 7694e7706.
Liu, Y., Zhao, G., Peng, X., 2019. Deep learning prognostics for lithium-ion battery Shang, Y., Xia, B., Zhang, C., Cui, N., Yang, J., Mi, C.C., 2017b. An automatic equalizer
based on ensembled long short-term memory networks. IEEE Access 7, based on forward-flyback converter for series-connected battery strings. IEEE
155130e155142. Trans. Ind. Electron. 64, 5380e5391.
Liu, Z., He, H., Ahmed, Q., Rizzoni, G., 2015. Structural analysis based fault detection Shen, M., Gao, Q., 2019. A review on battery management system from the
and isolation applied for a lithium-ion battery pack. IFAC-PapersOnLine 28, modeling efforts to its multiapplication and integration. Int. J. Energy Res. 43,
1465e1470. 5042e5075.
Liu, Z., Liu, X., Han, J., Yang, W., 2017b. An optimization algorithm for equalization Shrivastava, P., Soon, T.K., Idris, M.Y.I.B., Mekhilef, S., 2019. Overview of model-based
scheme of series-connected energy storage cells. In: Proceedings of the IEEE online state-of-charge estimation using Kalman filter family for lithium-ion
25
batteries. Renew. Sustain. Energy Rev. 113, 109233. power capacity with noise effect attenuation for lithium-ion battery. IEEE Trans.
Shu, X., Li, G., Shen, J., Yan, W., Chen, Z., Liu, Y., 2020. An adaptive fusion estimation Ind. Electron. 66, 5724e5735.
algorithm for state of charge of lithium-ion batteries considering wide oper- Xia, B., Cui, D., Sun, Z., Lao, Z., Zhang, R., Wang, W., Sun, W., Lai, Y., Wang, M., 2018a.
ating temperature and degradation. J. Power Sources 462, 1e13. State of charge estimation of lithium-ion batteries using optimized Levenberg-
Singh, K.V., Bansal, H.O., Singh, D., 2020. Hardware-in-the-loop implementation of Marquardt wavelet neural network. Energy 153, 694e705.
ANFIS based adaptive SoC estimation of lithium-ion battery for hybrid vehicle Xia, B., Mi, C., 2016. A fault-tolerant voltage measurement method for series con-
applications. J. Energy Storage 27, 101124. nected battery packs. J. Power Sources 308, 83e96.
Sivaraman, P., Sharmeela, C., 2020. IoT-Based Battery Management System for Xia, B., Nguyen, T., Yang, J., Mi, C., 2016. The improved interleaved voltage mea-
Hybrid Electric Vehicle. Wiley. surement method for series connected battery packs. J. Power Sources 334,
Song, L., Zhang, K., Liang, T., Han, X., Zhang, Y., 2020a. Intelligent state of health 12e22.
estimation for lithium-ion battery pack based on big data analysis. J. Energy Xia, B., Shang, Y., Nguyen, T., Mi, C., 2017. A correlation based fault detection method
Storage 32, 101836. for short circuits in battery packs. J. Power Sources 337, 1e10.
Song, X., Yang, F., Wang, D., Tsui, K.L., 2019. Combined CNN-LSTM network for state- Xia, Z., Ma, X., Qiu, D., Bu, G., Xia, Y., Zhao, B., Lin, Z., Shi, Y., 2018b. A design of
of-charge estimation of lithium-ion batteries. IEEE Access 7, 88894e88902. battery thermal management system based on fuzzy control. In: Proceedings of
Song, Y., Liu, D., Liao, H., Peng, Y., 2020b. A hybrid statistical data-driven method for the International Conference on Computer Modeling. Simulation and Algo-
on-line joint state estimation of lithium-ion batteries. Appl. Energy 261, 1e13. rithm, pp. 92e94.
Sun, F., Xiong, R., He, H., 2016. A systematic state-of-charge estimation framework Xiao, Bin, Liu, Y., Xiao, Bing, 2019. Accurate state-of-charge estimation approach for
for multi-cell battery pack in electric vehicles using bias correction technique. lithium-ion batteries by gated recurrent unit with ensemble optimizer. IEEE
Appl. Energy 162, 1399e1409. Access 7, 54192e54202.
Sun, J., Zhu, C., Lu, R., Song, K., Wei, G., 2015. Development of an optimized algo- Xing, Y., Ma, E.W.M., Tsui, K.L., Pecht, M., 2011. Battery management systems in
rithm for bidirectional equalization in lithium-ion batteries. J. Power Electron. electric and hybrid vehicles. Energies 4, 1840e1857.
15, 775e785. Xiong, J., Banvait, H., Li, L., Chen, Y., Xie, J., Liu, Y., Wu, M., Chen, J., 2012. Failure
Takei, K., Kumai, K., Kobayashi, Y., Miyashiro, H., Terada, N., Iwahori, T., Tanaka, T., detection for over-discharged Li-ion batteries. In: Proceedings of the IEEE In-
2001. Cycle life estimation of lithium secondary battery by extrapolation ternational Electric Vehicle Conference, pp. 1e5.
method and accelerated aging test. In: Journal of Power Sources. Elsevier, Xiong, R., 2020. Battery Management Algorithm for Electric Vehicles. Springer.
pp. 697e701. Xiong, R., Cao, J., Yu, Q., He, H., Sun, F., 2018a. Critical review on the battery state of
Tang, S., Yu, C., Wang, X., Guo, X., Si, X., 2014. Remaining useful life prediction of charge estimation methods for electric vehicles. IEEE Access 6, 1832e1843.
lithium-ion batteries based on the wiener process with measurement error. Xiong, R., Li, L., Tian, J., 2018b. Towards a smarter battery management system: a
Energies 7, 520e547. critical review on battery state of health monitoring methods. J. Power Sources
Tian, H., Qin, P., Li, K., Zhao, Z., 2020. A review of the state of health for lithium-ion 405, 18e29.
batteries: research status and suggestions. J. Clean. Prod. 261, 120813. Xiong, R., Ma, S., Li, H., Sun, F., Li, J., 2020a. Toward a safer battery management
Tran, D.D., Vafaeipour, M., El Baghdadi, M., Barrero, R., Van Mierlo, J., Hegazy, O., system: a critical review on diagnosis and prognosis of battery short circuit.
2020. Thorough state-of-the-art analysis of electric and hybrid vehicle power- iScience 23, 101010.
trains: topologies and integrated energy management strategies. Renew. Sus- Xiong, R., Pan, Y., Shen, W., Li, H., Sun, F., 2020b. Lithium-ion battery aging mech-
tain. Energy Rev. 119, 109596. anisms and diagnosis method for automotive applications: recent advances and
Tran, M.-K., Fowler, M., 2020. A review of lithium-ion battery fault diagnostic al- perspectives. Renew. Sustain. Energy Rev. 131, 110048.
gorithms: current progress and future challenges. Algorithms 13, 62. Xiong, R., Zhang, Y., Wang, J., He, H., Peng, S., Pecht, M., 2019. Lithium-ion battery
Turksoy, A., Teke, A., Alkaya, A., 2020. A comprehensive overview of the dc-dc health prognosis based on a real battery management system used in electric
converter-based battery charge balancing methods in electric vehicles. vehicles. IEEE Trans. Veh. Technol. 68, 4110e4121.
Renew. Sustain. Energy Rev. 133, 110274. Xue, Z., Zhang, Y., Cheng, C., Ma, G., 2020. Remaining useful life prediction of
Umair Ali, M., Hussain Nengroo, S., Adil Khan, M., Zeb, K., Ahmad Kamran, M., lithium-ion batteries with adaptive unscented kalman filter and optimized
Kim, H.-J., 2018. A real-time simulink interfaced fast-charging methodology of support vector regression. Neurocomputing 376, 95e102.
lithium-ion batteries under temperature feedback with fuzzy logic control. Yang, F., Li, W., Li, C., Miao, Q., 2019a. State-of-charge estimation of lithium-ion
Energies 11, 1122. batteries based on gated recurrent neural network. Energy 175, 66e75.
Ungurean, L., Micea, M.V., C^ arstoiu, G., 2020. Online state of health prediction Yang, F., Song, X., Xu, F., Tsui, K.-L., 2019b. State-of-Charge estimation of lithium-ion
method for lithium-ion batteries, based on gated recurrent unit neural net- batteries via long short-term memory network. IEEE Access 7, 53792e53799.
works. Int. J. Energy Res. 1e11. Yang, F., Zhang, S., Li, W., Miao, Q., 2020. State-of-charge estimation of lithium-ion
Wang, B., Qin, F., Zhao, X., Ni, X., Xuan, D., 2020a. Equalization of series connected batteries using LSTM and UKF. Energy 201, 117664.
lithium-ion batteries based on back propagation neural network and fuzzy logic Yang, R., Xiong, R., He, H., Chen, Z., 2018. A fractional-order model-based battery
control. Int. J. Energy Res. 44, 4812e4826. external short circuit fault diagnosis approach for all-climate electric vehicles
Wang, D., Miao, Q., Pecht, M., 2013. Prognostics of lithium-ion batteries based on application. J. Clean. Prod. 187, 950e959.
relevance vectors and a conditional three-parameter capacity degradation Yang, R., Xiong, R., He, H., Mu, H., Wang, C., 2017. A novel method on estimating the
model. J. Power Sources 239, 253e264. degradation and state of charge of lithium-ion batteries used for electrical ve-
Wang, Q., Jiang, B., Li, B., Yan, Y., 2016. A critical review of thermal management hicles. Appl. Energy 207, 336e345.
models and solutions of lithium-ion batteries for the development of pure Yang, S., Wang, W., Lin, C., Shen, W., Li, Y., 2019c. Investigation of internal short
electric vehicles. Renew. Sustain. Energy Rev. 64, 106e128. circuits of lithium-ion batteries under mechanical abusive conditions. Energies
Wang, Q., Ping, P., Zhao, X., Chu, G., Sun, J., Chen, C., 2012. Thermal runaway caused 12, 1885.
fire and explosion of lithium ion battery. J. Power Sources 208, 210e224. Yao, L., Fang, Z., Xiao, Y., Hou, J., Fu, Z., 2021. An intelligent fault diagnosis method
Wang, S., Zhao, L., Su, X., Ma, P., 2014a. Prognostics of lithium-ion batteries based on for lithium battery systems based on grid search support vector machine. En-
battery performance analysis and flexible support vector regression. Energies 7, ergy 214, 118866.
6492e6508. Yao, L., Wang, Z., Ma, J., 2015. Fault detection of the connection of lithium-ion power
Wang, Y., Liu, C., Pan, R., Chen, Z., 2017a. Experimental data of lithium-ion battery batteries based on entropy for electric vehicles. J. Power Sources 293, 548e561.
and ultracapacitor under DST and UDDS profiles at room temperature. Data Br Yin, Y., Hu, Y., Choe, S.-Y., Cho, H., Joe, W.T., 2019. New fast charging method of
12, 161e163. lithium-ion batteries based on a reduced order electrochemical model
Wang, Y., Tian, J., Sun, Z., Wang, L., Xu, R., Li, M., Chen, Z., 2020b. A comprehensive considering side reaction. J. Power Sources 423, 367e379.
review of battery modeling and state estimation approaches for advanced You, G. won, Park, S., Oh, D., 2016. Real-time state-of-health estimation for electric
battery management systems. Renew. Sustain. Energy Rev. 131, 110015. vehicle batteries: a data-driven approach. Appl. Energy 176, 92e103.
Wang, Y., Zhang, C., Chen, Z., 2014b. A method for joint estimation of state-of- You, G.W., Park, S., Oh, D., 2017. Diagnosis of electric vehicle batteries using
charge and available energy of LiFePO4 batteries. Appl. Energy 135, 81e87. recurrent neural networks. IEEE Trans. Ind. Electron. 64, 4885e4893.
Wang, Z., Hong, J., Liu, P., Zhang, L., 2017b. Voltage fault diagnosis and prognosis of Zhang, C.-W., Chen, S.-R., Gao, H.-B., Xu, K.-J., Yang, M.-Y., Zhang, C.-W., Chen, S.-R.,
battery systems based on entropy and Z-score for electric vehicles. Appl. Energy Gao, H.-B., Xu, K.-J., Yang, M.-Y., 2018e. State of charge estimation of power
196, 289e302. battery using improved back propagation neural network. Batteries 4, 69.
Wei, Z., Dong, G., Zhang, X., Pou, J., Quan, Z., He, H., 2021. Noise-immune model Zhang, J., Zhang, L., Sun, F., Wang, Z., 2018a. An overview on thermal safety issues of
identification and state-of-charge estimation for lithium-ion battery using lithium-ion batteries for electric vehicle application. IEEE Access 6,
bilinear parameterization. IEEE Trans. Ind. Electron. 68, 312e323. 23848e23863.
Wei, Z., Meng, S., Xiong, B., Ji, D., Tseng, K.J., 2016. Enhanced online model identi- Zhang, R., Xia, B., Li, B., Cao, L., Lai, Y., Zheng, W., Wang, H., Wang, W., Zhang, R.,
fication and state of charge estimation for lithium-ion battery with a FBCRLS Xia, B., Li, B., Cao, L., Lai, Y., Zheng, W., Wang, H., Wang, W., 2018b. State of the
based observer. Appl. Energy 181, 332e341. art of lithium-ion battery SOC estimation for electrical vehicles. Energies 11,
Wei, Z., Zhao, D., He, H., Cao, W., Dong, G., 2020. A noise-tolerant model parame- 1820.
terization method for lithium-ion battery management system. Appl. Energy Zhang, S., Yang, L., Zhao, X., Qiang, J., 2015. A GA optimization for lithium-ion
268, 114932. battery equalization based on SOC estimation by NN and FLC. Int. J. Electr.
Wei, Z., Zhao, J., Ji, D., Tseng, K.J., 2017. A multi-timescale estimator for battery state Power Energy Syst. 73, 318e328.
of charge and capacity dual estimation based on an online identified model. Zhang, X., Liu, C., Rao, Z., 2018c. Experimental investigation on thermal manage-
Appl. Energy 204, 1264e1274. ment performance of electric vehicle power battery using composite phase
Wei, Z., Zhao, J., Xiong, R., Dong, G., Pou, J., Tseng, K.J., 2019. Online estimation of change material. J. Clean. Prod. 201, 916e924.
26
Zhang, Y., Xiong, R., He, H., Pecht, M.G., 2019a. Lithium-ion battery remaining useful Zheng, F., Xing, Y., Jiang, J., Sun, B., Kim, J., Pecht, M., 2016. Influence of different
life prediction with box-cox transformation and Monte Carlo simulation. IEEE open circuit voltage tests on state of charge online estimation for lithium-ion
Trans. Ind. Electron. 66, 1585e1597. batteries. Appl. Energy 183, 513e525.
Zhang, Y., Xiong, R., He, H., Pecht, M.G., 2018d. Long short-term memory recurrent Zheng, J., Sun, X., Jia, L., Zhou, Y., 2020. Electric passenger vehicles sales and carbon
neural network for remaining useful life prediction of lithium-ion batteries. dioxide emission reduction potential in China’s leading markets. J. Clean. Prod.
IEEE Trans. Veh. Technol. 67, 5695e5705. 243, 118607.
Zhang, Y.Z., Xiong, R., He, H.W., Pecht, M., 2019b. Validation and verification of a Zhou, Y., Huang, M., Pecht, M., 2020. Remaining useful life estimation of lithium-ion
hybrid method for remaining useful life prediction of lithium-ion batteries. cells based on k-nearest neighbor regression with differential evolution opti-
J. Clean. Prod. 212, 240e249. mization. J. Clean. Prod. 249, 119409.
Zhao, X., Xuan, D., Zhao, K., Li, Z., 2020. Elman neural network using ant colony Zhu, M., Hu, W., Kar, N.C., 2016. The SOH estimation of LiFePO4 battery based on
optimization algorithm for estimating of state of charge of lithium-ion battery. internal resistance with Grey Markov chain. In: Proceedings of the IEEE
J. Energy Storage 32, 101789. Transportation Electrification Conference and Expo, pp. 1e6.
Zhao, Y., Liu, P., Wang, Z., Zhang, L., Hong, J., 2017. Fault and defect diagnosis of Zun, C.Y., Park, S.U., Mok, H.S., 2020. New cell balancing charging system research
battery for electric vehicles based on big data analysis methods. Appl. Energy for lithium-ion batteries. Energies 16, 1e10.
207, 354e362.
27

1 s2.0 S095965262100264X Main

Uploaded by

Copyright:

Available Formats

1 s2.0 S095965262100264X Main

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

1 s2.0 S095965262100264X Main

Uploaded by

Copyright:

Available Formats

Journal of Cleaner Production 292 (2021) 126044

Contents lists available at ScienceDirect

Journal of Cleaner Production

Intelligent algorithms and control strategies for battery management

3.2.2. Optimization-based algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

(a) Screening Method electrochemical reaction, polarization, solid-phase diffusion con-

3.1.2. State of health prediction

3.1.2.1. Feed-forward neural network algorithm-based SOH predic-

Refs. Methods Algorithms execution formulas Description

R: Ohmic resistance, C: Capacitance, PR: Polarization resistance.

Methods Advantages Disadvantages Refs.

runaway propagation. learning-based algorithms, mathematical data-driven algorithms,

connection faults. The developed method detected the faults

3.3.4. Entropy algorithms

Journal of Cleaner Production 292 (2021) 126044

the battery temperature rises 1 C in the temperature range of

Fig. 6. The outline of various faults in the lithium-ion battery systems.

some limitations. FNN-based approaches can achieve reasonable

4.1. Algorithms issues 4.2.1. Data abundance and data variety

Fig. 10. Future trends for advanced BMS in EVs.

appropriate structure, selection of input features, hyperparameter Acknowledgment

You might also like