A data-driven framework for lithium-ion batteries safety assessment integrating health degradation and key thermal safety parameters

Accurately estimating the state of health (SoH) and state of safety (SoS) of lithium-ion batteries (LIBs) is essential for ensuring the reliable and safe operation of electric vehicles. This study presents an innovative method for SoS estimation, specifically addressing the state estimation and management issues of batteries under fast charging conditions. Firstly, the aging behaviors of LIB under fast charging are analyzed, and Pearson correlation analysis was used to identify key health factors strongly related to SoH. The Bo-Seq2Seq model is then employed for SoH estimation, with hyperparameters optimized through Bayesian optimization to enhance accuracy and efficiency. Next, thermal safety parameters are identified using experimental thermal safety data combined with the SHAP value method, and the dataset is augmented to improve robustness. To account for aging effects on thermal runaway, SoH and thermal safety parameters are integrated to estimate SoS under thermal abuse conditions, capturing performance differences across various aging stages. Finally, K-means clustering is applied to classify SoS into distinct categories, enabling dynamic thermal safety assessments. The results demonstrate that the proposed model significantly outperforms traditional methods, achieving an RMSE as low as 1 % in SoH estimation and an average RMSE of 3.89 % in SoS estimation. This approach significantly enhances the accuracy of both SoH and SoS predictions, providing valuable insights for the development of advanced battery management system.