Abstract: Due to the randomness caused by time variability, path dependence and inconsistency of power battery degradation, it is a trend to judge battery abnormalities and identify battery safety risks through background historical operation data. At present, the fault diagnosis for historical data has the problems of ignoring the time-varying battery aging and excessive dependence on the equivalent model and cycle data. Abnormal batteries are screened based on statistical iteration and healthy battery reference voltage is extracted through inconsistency analysis in series-battery pack. By establishing the deviation index representation model, the variation law of voltage outliers during a single charging process is analyzed, and the degree and type of battery are diagnosed abnormality from a short time scale. Based on the coefficient of variation method, a comprehensive deviation index matrix for the entire historical process is constructed, and the abnormal change speed of the battery is judged from the abrupt change of the long-term dimension. The fault examples and batch analysis results show that the abnormality diagnosis method proposed can not only accurately diagnose abnormal batteries in the battery pack, but also effectively quantify the deterioration trend, providing a basis for fault early warning and rationalized intervention measures.

Key words: lithium-ion battery, historical data, abnormal diagnosis, capacity, state of charge