A multidimensional anomaly detection framework for battery capacity degradation in electric vehicles using real-world data

Abnormal capacity degradation in electric vehicle (EV) battery systems poses a major threat to system reliability and safety. This study proposes a multidimensional anomaly detection framework for identifying abnormal battery capacity degradation using real-world EV operational data. The framework consists of three key components. First, a capacity estimation model is developed using a stacking technique to estimate the capacity in real-world EV scenarios, achieving an average prediction error of 2.34% on the test set. Second, Rate of Capacity Degradation (RoCD) is defined as a metric that incorporates both mileage and time to quantify capacity degradation. Third, a dynamic threshold-setting method based on the statistical distribution of RoCD is designed to detect anomalies across multiple influencing factors, including temperature, charging current, and depth of discharge. The proposed framework is validated with real-world data and effectively identifies EVs with abnormal capacity degradation, contributing to enhanced battery monitoring and safety management.