Fault Identification Method of Electric Vehicle Power Battery Single Cell Voltage Inconsistency Based on DWT and PCA

Voltage inconsistency is one of the important factors that cause safety problems in power batteries. Therefore, based on electric vehicle operation data from the automotive enterprise monitoring platform, this paper builds upon the Principal Component Analysis (PCA) method and incorporates Discrete Wavelet Transform (DWT) characterized by possesses multi‐resolution analysis and time‐frequency localization characteristics to achieve data denoising and feature extraction, thereby proposing a fault identification method for power battery cell voltage inconsistency based on DWT and PCA. First, according to the change of false alarm rate (FAR), the voltage data of 20 rounds of charge‐discharge cycles is selected as the dataset for model threshold training, which solves the problem of FAR fluctuation caused by early voltage oscillation. Second, the influence of different wavelet functions on the diagnostic results of this method is investigated, and it is found that db4 wavelet can eliminate the FAR caused by noise. Finally, the method is used to analyze the three vehicles randomly selected from the monitoring platform of automobile enterprises which have voltage inconsistency alarms. The results show that for voltage inconsistency faults of the same vehicle model, the method can detect the faults 2 days earlier than the platform; for voltage inconsistency faults of gradual and transient type occurring in different vehicles, the method can detect the faults 9 days and 22 min earlier than the platform; for voltage inconsistency faults occurring in multiple units in the same vehicle, the method can also combine the anomaly rate to rank the degree of faults of the single cell. Based on the primary performance metrics reported in the literature and algorithmic principles, we find that, compared with current mainstream methods, the proposed approach outperforms them in accuracy, precision, recall, and computational time. The research findings significantly enhance the efficiency of power battery fault identification.