Battery impedance spectrum prediction from partial charging voltage curve by machine learning  

作　　者：Jia Guo Yunhong Che Kjeld Pedersen Daniel-Ioan Stroe 

机构地区：[1]AAU Energy,Aalborg University,Aalborg 9220,Denmark [2]Department of Materials and Production,Aalborg University,Aalborg 9220,Denmark

出　　处：《Journal of Energy Chemistry》2023年第4期211-221,共11页能源化学（英文版）

基　　金：supported by a grant from the China Scholarship Council (202006370035);a fund from Otto Monsteds Fund (4057941073)。

摘　　要：Electrochemical impedance spectroscopy(EIS) is an effective technique for Lithium-ion battery state of health diagnosis, and the impedance spectrum prediction by battery charging curve is expected to enable battery impedance testing during vehicle operation. However, the mechanistic relationship between charging curves and impedance spectrum remains unclear, which hinders the development as well as optimization of EIS-based prediction techniques. In this paper, we predicted the impedance spectrum by the battery charging voltage curve and optimized the input based on electrochemical mechanistic analysis and machine learning. The internal electrochemical relationships between the charging curve,incremental capacity curve, and the impedance spectrum are explored, which improves the physical interpretability for this prediction and helps define the proper partial voltage range for the input for machine learning models. Different machine learning algorithms have been adopted for the verification of the proposed framework based on the sequence-to-sequence predictions. In addition, the predictions with different partial voltage ranges, at different state of charge, and with different training data ratio are evaluated to prove the proposed method have high generalization and robustness. The experimental results show that the proper partial voltage range has high accuracy and converges to the findings of the electrochemical analysis. The predicted errors for impedance spectrum are less than 1.9 mΩ with the proper partial voltage range selected by the corelative analysis of the electrochemical reactions inside the batteries. Even with the voltage range reduced to 3.65–3.75 V, the predictions are still reliable with most RMSEs less than 4 mO.

关 键 词：Impedance spectrum prediction Lithium-ion battery Machine learning EIS Graphite anode 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TM912[自动化与计算机技术—控制科学与工程]