Elsevier

Neurocomputing

Volume 414, 13 November 2020, Pages 245-254
Neurocomputing

Remaining useful life prediction of lithium-ion battery with optimal input sequence selection and error compensation

https://doi.org/10.1016/j.neucom.2020.07.081Get rights and content

Abstract

Accurate prediction of remaining useful life (RUL) for lithium-ion battery (LIB) plays a key role in increasing the reliability and safety of battery related industries and facilities. In this paper, RUL prediction of LIB is investigated by employing a hybrid data-driven method based support vector regression (SVR) and error compensation (EC). Firstly, two health indicators (HIs) are established by using capacity and discharging voltage difference of equal time interval (DVD), respectively. Secondly, the ensemble empirical mode decomposition (EEMD) is adopted to preprocess the obtained HIs, which is used to reduce the influence of capacity regeneration and noise. Especially, phase space reconstruction (PSR) with C–C technique is introduced to achieve optimal input sequence selection pattern, it has an important influence on the accuracy of SVR prediction. As an important innovation of the paper, the idea of EC is implemented by combining the predictions of both forecast error and RUL prediction with PSR-SVR. Last but not least, the genetic algorithm (GA) is utilized to optimize the key parameters of SVR so as to achieve more accurate RUL prediction. To verify the effectiveness of the proposed approach, the real data set of LIBs from National Aeronautics and Space Administration (NASA) is carried out, and the dominant is emphasized by comparison with other important methods.

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)

  • M.A. Patil et al.

    A novel multistage support vector machine based approach for Li-ion battery remaining useful life estimation

    Appl. Energy

    (2015)
  • T. Qin et al.

    Robust prognostics for state of health estimation of lithium-ion batteries based on an improved PSO-SVR model

    Microelectron. Reliab.

    (2015)
  • Y. Shen et al.

    Fusion estimation for multi-rate linear repetitive processes under weighted Try-Once-Discard protocol

    Inf. Fusion

    (2020)
  • H.S. Kim et al.

    Nonlinear dynamics, delay times and embedding windows

    Physica D

    (1999)
  • D. Liu et al.

    Prognostic for state of health estimation of lithium-ion batteries based on combination Gaussian process functional regression

    Microelectron. Reliab.

    (2013)
  • N. Williard et al.

    Lessons learned from the 787 Dreamliner issue on lithium-ion battery reliability

    Energies

    (2013)
  • L.X. Liao et al.

    Review of hybrid prognostics approaches for remaining useful life prediction of engineered systems, and an application to battery life prediction

    IEEE Trans. Reliab.

    (2014)
  • Y. Zheng et al.

    Normalized relative RBC based minimum risk bayesian decision approach for fault diagnosis of industrial process

    IEEE Trans. Ind. Electron.

    (2016)
  • G. Dong et al.

    Data-driven battery health prognosis using adaptive brownian motion model

    IEEE Trans. Ind. Inf.

    (2020)
  • Y. Zhang et al.

    Detection of intermittent faults for nonuniformly sampled multirate systems with dynamic quantization and missing measurements

    Int. J. Control

    (2020)
  • Cited by (82)

    View all citing articles on Scopus

    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.

    View full text