A multi-scenario data-driven approach for anomaly detection in electric vehicle battery systems

The increasing adoption of electric vehicle (EV) emphasizes the need for safer battery systems. However, detecting anomalies during charging and discharging processes remains challenging due to the high variability and complexity of EV operational data. This study proposes a multi-scenario data-driven framework to address these challenges. The Pearson Correlation Coefficient is employed for feature selection in charging scenarios, while a Time Series Shape Feature Extraction Algorithm is developed for discharging scenarios to reduce data dimensionality while preserving critical information. An enhanced Transformer model integrated with a Generative Adversarial Network reconstructs voltage data, capturing complex temporal dependencies. Additionally, an improved Cumulative Sum algorithm with a sliding window mechanism enhances sensitivity to localized anomalies. Validation with real-world EV data demonstrates F₁ score of 90.38 % in charging and 86.55 % in discharging, outperforming existing methods. Moreover, the framework can detect anomalies at least two charging and discharging cycles (67 h) before thermal runaway occur. Additionally, a techno-economic analysis reveals that the framework could prevent up to $692.99 million in economic losses for China's EV fleet by reducing fire-related incidents. The presented framework enhance safety, reduce risks, and offer substantial economic benefits, demonstrating its potential for large-scale application in the EV industry.