Evaluation of battery inconsistency in electric vehicles based on fusion of image features and temporal features

Inconsistencies in electric vehicle monobloc batteries can lead to reduced capacity and shortened service life, making accurate evaluation of battery inconsistency crucial for driving safety. This study addresses the challenges of early and potential inconsistency detection and the limitations of single-feature analysis. A novel method combining image and temporal features is proposed to assess battery inconsistency. The Gramian Angular Field technique is used to convert temporal data into 2D images, from which inconsistency information is extracted. Temporal features are further utilized to quantify the battery’s behavioral fluctuations over time. Additionally, a cumulative inconsistency metric is introduced to capture performance degradation and potential issues by analyzing inconsistency trends through a sliding window. In particular, an inconsistency-based cell and battery pack classification method is also proposed to accurately categorize inconsistency levels. Data validation across vehicles with varying driving ranges demonstrates that the method effectively identifies and quantifies inconsistencies, offers standardized evaluations for different vehicles, and exhibits high reliability, showcasing significant potential for application.