A hybrid real-time safety monitoring framework for power battery packs using autoencoders and adaptive extended Kalman filter-based HARA

Power battery packs are a critical component of electric vehicles (EVs), and their safety and stability directly dictate the reliability and commercial viability of modern electrified transportation. Traditional battery monitoring systems often rely on static threshold-based models, which struggle to provide adaptive and accurate safety diagnostics under highly dynamic operating conditions. This paper integrates real-time sensor measurements with a systematic Hazard Analysis and Risk Assessment (HARA)framework to propose a real-time monitoring and analysis model for power battery packs. The system employs an Adaptive Extended Kalman Filter (AEKF) to adjust system parameters using sensor signals derived from battery performance and monitors various fault types based on HARA results. An autoencoder network enhances the performance of state estimation and monitoring. Experiments conducted on OBD, CALCE, and self-constructed datasets validate the model’s performance in terms of fault detection consistency, isolation accuracy, and State of Charge prediction. The experimental results show that the fault detection consistency of the monitoring system based on sensor signals and HARA reaches 94.28% ± 0.76%, the insulation resistance estimation error is 1.492 ± 0.067 Ω/V, and the SOC prediction accuracy is 98.46% ± 0.42%, which is better than the other three models.