A novel Long-term degradation trends predicting method for Multi-Formulation Li-ion batteries based on deep reinforcement learning. (August 2022)

• Record Type: Journal Article
• Authors: Wang, Chao; Ding, Yu; Yan, Ning; Ma, Liang; Ma, Jian; Lu, Chen; Yang, Chao; Su, Yuzhuan; Chong, Jin; Jin, Haizu; Lin, Yongshou
• Is Part Of: Advanced engineering informatics. Volume 53(2022)
• Publication Date: 2022-08
• Subjects: Li-ion Battery -- Cycle life test -- Degradation trend prediction -- Deep reinforcement learning -- Deep deterministic policy gradient; Computer-aided engineering -- Periodicals; Engineering -- Data processing -- Periodicals; 620.00285
• DOI: 10.1016/j.aei.2022.101665
• Languages: English
• ISSNs: 1474-0346

Abstract

Abstract: In the design phase of Li-ion batteries for electric vehicles, battery manufacturers need to carry out cycle life tests on a large number of formulations to get the best one that meets customer demands. However, such tests take considerable time and money due to the long cycle life of power Li-ion batteries. Aiming at reducing the cost of cycle life tests, we propose a prediction method that can learn historical degradation data and extrapolate to predict the remaining degradation trend of the current formulation sample taking the initial stage of partial cycle life test results as input. Compared with existing methods, the proposed deep reinforcement learning based method is able to learn degradation trends with different formulations and predict long-term degradation trends. Based on the deep deterministic policy gradient algorithm, the proposed method builds a degradation trend prediction model. Meanwhile, an interactive environment is designed for the model to explore and learn in the training phase. The proposed method is verified with real test data from battery manufacturers under three different temperature conditions in the formulation design stage. The comparisons indicate that the proposed method is superior to traditional degradation trend prediction methods in both accuracy and stability.