Statistical Feature-Driven Multi-Anomaly Diagnosis Method for Real Vehicle Power Battery

Abnormal battery diagnosis is critical for ensuring the safety of the power battery. However, the existing methods are limited in terms of the diagnosis types and the practical application. Therefore, a statistical feature-driven multi-anomaly diagnosis method is proposed. First, a median voltage deviation (MVD) signal processing is offered to reveal the characteristics of battery voltage signals under different anomalies. The MVD signals are used to construct high-dimensional statistical features in the time and frequency domains. Then, a composite model is built by using extreme gradient boost (XGBoost) to capture the association of these features with battery anomalies, as well as to diagnose battery anomalies. Finally, the effectiveness of the proposed method is verified by employing a real power battery data set, demonstrating its potential for practical applications.