Early Anomaly Detection of Power Battery Based on Time-series Features

Early anomaly detection in power batteries is crucial to ensure safe and reliable operation of electric vehicles. Although a lot of research has been conducted on battery anomaly detection, little attention has been paid to the time-series features of the charging curves of single batteries. This paper proposes a power battery early anomaly detection method based on time-series features. By dynamically matching the charging segments with the historical charging data, seven different multi-timescale timing features are extracted, and the local outlier factor (LOF) algorithm is used to achieve the anomaly detection of a single unit. The experimental results show that the method can accurately detect abnormal cells in the battery pack before the warning of the battery management system by using only two of the features, and the false detection rate is less than 3%.