State of health and remaining useful life prediction for lithium-ion batteries based on differential thermal voltammetry and a long and short memory neural network

• Published in: Rare metals 2023-03, Vol.42 (3), p.885-901
• Main Authors: Ma, Bin, Yu, Han-Qing, Wang, Wen-Tao, Yang, Xian-Bin, Zhang, Li-Sheng, Xie, Hai-Cheng, Zhang, Cheng, Chen, Si-Yan, Liu, Xin-Hua
• Format: Article
• Language: English
• Subjects: Biomaterials; Chemistry and Materials Science; Correlation analysis; Data collection; Degradation; Digital twins; Energy; Feature extraction; Life prediction; Lithium-ion batteries; Management systems; Materials Engineering; Materials Science; Metallic Materials; Nanoscale Science and Technology; Network reliability; Neural networks; Original Article; Physical Chemistry; Power management; Rechargeable batteries; Signal analysis; Useful life; Voltammetry
• ISSN: 1001-0521; 1867-7185
• DOI: 10.1007/s12598-022-02156-1

As the lithium-ion battery is widely applied, the reliability of the battery has become a high-profile content in recent years. Accurate estimation and prediction of state of health (SOH) and remaining useful life (RUL) prediction are crucial for battery management systems. In this paper, the core contribution is the construction of a data-driven model with the long short-term memory (LSTM) network applicable to the time-series regression prediction problem with the integration of two methods, data-driven methods and feature signal analysis. The input features of model are extracted from differential thermal voltammetry (DTV) curves, which could characterize the battery degradation characteristics, so that the accurate prediction of battery capacity fade could be accomplished. Firstly, the DTV curve is smoothed by the Savitzky-Golay filter, and six alternate features are selected based on the connection between DTV curves and battery degradation characteristics. Then, a correlation analysis method is used to further filter the input features and three features that are highly associated with capacity fade are selected as input into the data driven model. The LSTM neural network is trained by using the root mean square propagation (RMSprop) technique and the dropout technique. Finally, the data of four batteries with different health levels are deployed for model construction, verification and comparison. The results show that the proposed method has high accuracy in SOH and RUL prediction and the capacity rebound phenomenon can be accurately estimated. This method can greatly reduce the cost and complexity, and increase the practicability, which provides the basis and guidance for battery data collection and the application of cloud technology and digital twin. Graphical abstract