A Data-Driven Approach for Battery System Safety Risk Evaluation Based on Real-world Electric Vehicle Operating Data

The safety evaluation of battery systems is crucial to prevent thermal runaway in electric vehicles (EVs) and ensure their safe and efficient operation. This paper proposed a data-driven approach that utilizes real-world operational data to evaluate the safety risk of EV battery systems. Five key parameters related to voltage and temperature were selected from the lifecycle data of normal and thermally runaway (TR) EVs, and features were extracted based on the differences in parameter distributions. A dynamic safety risk evaluation model (DSREM) was constructed in three steps. Firstly, Fuzzy Logic was employed to discretize the features using Membership Functions (MF). Then, a Bayesian network (BN) was constructed to assess safety risks. Finally, a dynamic safety risk evaluation framework was established to achieve effective real-time evaluation of safety risks. The accuracy of the proposed method was validated using both small and large sample datasets, demonstrating the accuracy of 96.67% while maintaining excellent computational efficiency. Furthermore, based on Receiver Operating Characteristic (ROC) curve and dynamic evaluation results, a safety warning strategy was proposed to provide timely alerts and maintenance, effectively reducing the risk of TR accidents.