Remaining useful life prediction of lithium-ion battery with optimal input sequence selection and error compensation
Introduction
Lithium-ion batteries (LIBs) possess obvious advantages over high voltage, high energy density, low self-discharge rate, long cycle life and high safety performance, therefore they have become the main power sources in aircraft, electric vehicles, mobile phones, laptops and even aerospace [1]. With the increase of the charge and discharge times, LIBs will inevitably degrade and even lead to failure. If effective action cannot be taken before the battery failure, the equipment with LIBs will not operate healthily, and the casualties may happen in serious cases [2]. Battery management systems are widely adopted to solve above safety problems which utilizes the information of remaining useful life (RUL) prediction of LIBs. Therefore, the prediction of RUL by calculating the time interval from starting time to threshold point has attracted wide attentions [3]. Due to the degradation process of LIBs is affected by complex internal electrochemical reaction, external temperature and pressure, charge and discharge interval and other factors, accurate RUL prediction is very difficult and challenging.
As the development of fault diagnosis theory [4], [5], [6], typical RUL prediction methods including model-based methods [7], [8], [9], data-driven methods [10], [11], [12] and hybrid methods [13], [14], [15], which owns the function of avoiding faults and attracts many research in different fields [16], [17]. With the rapid development of machine learning theory and the reality that it is difficult to obtain accurate degradation model of LIBs, data-driven approach including machine learning and deep learning have been investigated worldwide [18]. Different from deep learning-based prediction methods [19], [20], support vector regression (SVR) is a representative machine learning method, which shows obvious superiority by handling small samples, nonlinearity and time-series analysis. The application of SVR to the RUL prediction of LIBs falls into two categories: one is the direct method by utilizing semplice the SVR to forecast the RUL [14], [21], [22], the other is indirect technique by employing intelligent optimization algorithm to optimize SVR’s parameters which influence prediction precision [15], [23], [24]. Although the optimization algorithms with search characteristics improve the accuracy of prediction, it fatally leads to the computational complexity. Therefore, it is of great significance to explore an effective processing method for real-time RUL prediction of LIBs without changing the SVR algorithm.
It is a pleasant discovery that both health indicator (HI) and feature enhancement technology can solve the above problems and ameliorate the prediction accuracy. The typical HI including capacity and voltage can be used to quantify the aging precess of LIBs, and determine the failure when capacity or voltage transcends some percentage of its nominal value [3], [25]. In terms of feature enhancement, sliding window method [19] and trend reconstruction method [14] are chosen to reflect efficiently the degrade of HI due to the fluctuations caused by charging and discharging process. Especially, an obvious but neglected problem is that both length and mode of input data sequence on SVR will affect prominently the prediction accuracy. Therefore, it is of great significance to establish effective HI, adopt proper feature enhancement techniques and search for the optimal data entry model.
In response to the above discussions, the main contributions of this paper are outlined in fourfold as follows. 1) The classical capacity is chosen as the first HI, and discharge voltage difference of equal time interval is introduced as the second HI, then two health indicators input the predictor so that more LIBs aging information can be achieved. 2) Ensemble empirical mode decomposition (EEMD) is adopted as feature enhancement technique to reconstruct the degradation trend of LIBs, so as to reduce the impact of HI fluctuations. 3) Fortunately, phase space reconstruction (PSR) method is found to be able to solve both the number of sequences and the selection pattern contained in the SVR input window, so as to achieve the optimal input sequence. 4) The regression error data of training set is collected as the input of SVR, the error SVR prediction model is established to compensate the prediction error of the test set, thus prediction accuracy is greatly improved.
Section snippets
Framework of RUL prediction
The framework of the proposed method is shown in Fig. 1, and it involves three principal parts to achieve the prediction:
1) Two HIs are established identified as H1 and H2, respectively. EEMD is used to reconstruct H1 and H2 so as to reduce the influence.
2) PSR is utilized to determine the maximum embedding dimension and delay time, then the optimal input sequence pattern of SVR training can be obtained.
3) The final RUL prediction consists of initial prediction and error compensation.
Feature extraction
Effective
Lithium-ion battery data description
The lithium-ion degradation data used in this paper was derived from Prognostics Center of Excellence (PCoE) [29]. The battery runs through 3 different operational profiles (charge, discharge and impedance) at room temperature, and it is first charged in 1.5 A constant current mode until the voltage reaches 4.2 V, and then, it continues to charge in constant-voltage mode until the charging current drops to 20 mA. During discharge phase, the battery is discharged at a constant current of 2 A
Conclusion
This paper proposes a hybrid data-driven method to improve the prediction accuracy of RUL for lithium-ion battery. Through the research, several meaningful conclusions can be found: 1) Ensemble empirical mode decomposition (EEMD) can be used to reconstruct the degradation trend of LIBs and reduce the impact of HI fluctuations. 2) Phase space reconstruction (PSR) is an efficient technique in deal with the input sequence pattern of SVR. 3) The idea of error compensation and the optimization of
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgement
This work was supported in part by the National Natural Science Foundation of China [Grant Nos. 61873197, 61873102], National key research and development program [Grant No. 2018YFB1701202]. The authors would like to thank Prof. Ye Yuan (School of Artificial Intelligence and Automation, Huazhong University of Science and Technology) for insightful discussion and continuous help.
Liaogehao Chen received the B.E. in the School of Information Science and Engineering, Wuhan University of Science and Technology, Wuhan, China. Since 2018, he has been studying for master’s degree of control engineering in the School of Information Science and Engineering, Wuhan University of Science and Technology. His current research interests include remaining useful life prediction of key equipment.
References (41)
Lithium ion secondary batteries; past 10 years and the future
J. Power Sources
(2001)- et al.
Annulus-event-based fault detection, isolation and estimation for multirate time-varying systems: applications to a three-tank system
J. Process Control
(2019) - et al.
Data driven discovery of cyber physical systems
Nat. Commun.
(2019) - et al.
A recurrent neural network approach for remaining useful life prediction utilizing a novel trend features construction method
Measurement
(2019) - et al.
Health state prediction and analysis of SOFC system based on the data-driven entire stage experiment
Appl. Energy
(2019) - et al.
An integrated unscented kalman filter and relevance vector regression approach for lithium-ion battery remaining useful life and short-term capacity prediction
Reliab. Eng. Syst. Saf.
(2015) - et al.
A new hybrid method for the prediction of the remaining useful life of a lithium-ion battery
Appl. Energy
(2017) - et al.
Remaining useful life prediction of lithium-ion batteries with adaptive unscented kalman filter and optimized support vector regression
Neurocomputing
(2020) - et al.
Degradation data analysis and remaining useful life estimation: a review on Wiener-process-based methods
Eur. J. Oper. Res.
(2018) - et al.
Machinery health prognostics: a systematic review from data acquisition to RUL prediction
Mech. Syst. Signal Process.
(2018)
A novel multistage support vector machine based approach for Li-ion battery remaining useful life estimation
Appl. Energy
Robust prognostics for state of health estimation of lithium-ion batteries based on an improved PSO-SVR model
Microelectron. Reliab.
Fusion estimation for multi-rate linear repetitive processes under weighted Try-Once-Discard protocol
Inf. Fusion
Nonlinear dynamics, delay times and embedding windows
Physica D
Prognostic for state of health estimation of lithium-ion batteries based on combination Gaussian process functional regression
Microelectron. Reliab.
Lessons learned from the 787 Dreamliner issue on lithium-ion battery reliability
Energies
Review of hybrid prognostics approaches for remaining useful life prediction of engineered systems, and an application to battery life prediction
IEEE Trans. Reliab.
Normalized relative RBC based minimum risk bayesian decision approach for fault diagnosis of industrial process
IEEE Trans. Ind. Electron.
Data-driven battery health prognosis using adaptive brownian motion model
IEEE Trans. Ind. Inf.
Detection of intermittent faults for nonuniformly sampled multirate systems with dynamic quantization and missing measurements
Int. J. Control
Cited by (82)
Data-driven degradation trajectory prediction and online knee point identification of battery in electric vehicles
2024, Engineering Failure AnalysisRemaining useful life prediction of lithium-ion battery based on fusion model considering capacity regeneration phenomenon
2024, Journal of Energy StorageRemaining useful life prediction for lithium-ion batteries with an improved grey particle filter model
2024, Journal of Energy StorageTransfer learning based remaining useful life prediction of lithium-ion battery considering capacity regeneration phenomenon
2024, Journal of Energy Storage
Liaogehao Chen received the B.E. in the School of Information Science and Engineering, Wuhan University of Science and Technology, Wuhan, China. Since 2018, he has been studying for master’s degree of control engineering in the School of Information Science and Engineering, Wuhan University of Science and Technology. His current research interests include remaining useful life prediction of key equipment.
Yong Zhang received the Ph.D. degree in control theory and control engineering from Huazhong University of Science and Technology, Wuhan, China, in 2010. From 2014 to 2015, he was a Visiting Scholar with the Department of Information Systems and Computing, Brunel University London, Uxbridge, U.K. He is currently a Professor with the School of Information Science and Engineering, Wuhan University of Science and Technology, Wuhan, China. His current research interests include fault diagnosis and prognosis in industrial systems.
Ying Zheng received the B.S. degree in industrial electric automation engineering, in 1997, and the M.S. and Ph.D. degree in control theory and engineering from the Huazhong University of Science and Technology, Wuhan, China, in 2000 and 2003, respectively. She has been a Post-Doctor Fellow with the Chemical Engineering Department, National Tsing-Hua University, Hsinchu, Taiwan, from 2004 to 2005, and a Faculty Member with the Department of Control Science and Engineering, Huazhong University of Science and Technology (Associate Professor in 2005 and Professor in 2010). Her research interests include process control, data-driven method, fault diagnosis, and networked control system.
Xiangshun Li received the Ph.D. degree in control science and engineering from Huazhong University of Science and Technology, Wuhan, China, in 2009. From 2013 to 2014, he was a Visiting Scholar with the Department of Electrical and Computer, the University of Auckland, Auckland, New Zealand. He is currently a Professor with the School of Automation, Wuhan University of Technology, Wuhan, China. His current research interests include fault detection and diagnosis for industrial processes.
Xiujuan Zheng received the B.S. and M.S. degrees from Wuhan University of Science and Technology, China, in 2009 and 2012, respectively, and the Ph.D degree from the Huazhong University of Science and Technology, Wuhan, China, in 2016. She is currently a lecturer with the School of Information Science and Engineering, Wuhan University of Science and Technology. Her research interests include state estimation, networked control system, and fault diagnosis.