Multi-scale safety risk assessment method for electric vehicle power batteries

Abstract Currently, Automaker’s Monitoring Platforms (AMPs) use threshold-based alert criteria for power battery fault monitoring, prone to false and incorrect alarms. Existing data-driven models only focus on pack-level safety risks. To address these issues, this paper proposes a multi-scale safety risk assessment method for electric vehicle (EV) power batteries. Feature parameters from pack and cell scales are extracted from AMPs’ real-vehicle historical data, with key features selected via Spearman correlation analysis. The EWM+TOPSIS model and EWM model are adopted for pack and cell-level risk assessment, respectively, with alarm grade intervals defined by boxplots. The Random Forest (RF) algorithm predicts pack safety risks. Validated by real-vehicle failure data, the method achieves earlier and more accurate pack risk alerts than AMPs, precisely calculates cell-level risk scores with clear temporal trends, and accurately locates multiple faulty cells.