Enhancing State of Health Estimation of Lithium-Ion Batteries Through Strategic Feature Analysis

Accurate battery state of health (SOH) estimation is crucial for the safe and reliable operation of electric vehicles (EVs). Conventional estimation methods often face disturbances in complex and dynamic environments. The key challenge is to disregard such interferences and distill valuable information from battery aging data to establish an accurate SOH mapping. This study proposes a novel data-driven framework for precise and robust SOH estimation, comprising three modules: health indicator (HI) extraction, multiview HI selection, and SOH estimation using a convolutional neural network (CNN) with integrated feature interaction and channel attention (CA) blocks. The estimation module flexibly applies two interaction types, linear and cross, to achieve three modes of interaction. The impact of linear and cross interactions on SOH estimation is explored by two check methods. The check results on three lithium-ion battery (LIB) aging datasets highlight the positive effect of linear interaction and the negative effect of cross interaction. The estimation results demonstrate that the proposed approach outperforms existing methods, achieving up to a 30.516% improvement in mean absolute error (MAE). Moreover, the model’s scalability is verified across different types of LIBs under data-limited scenarios, with transfer learning (TL) yielding a remarkable 61.612% MAE improvement over models without it.