Multi-feature weighted battery pack consistency evaluation based on massive real-world data

Abstract

The widespread application of electric vehicles and energy storage systems has led to an increasing use of battery packs, and the problem of inconsistency among battery cells has become prominent. This issue stems from differences in manufacturing processes and usage conditions, and it severely affects the performance, safety, and service life of battery packs. Most existing studies are based on limited laboratory data and are unable to comprehensively analyze battery consistency, often neglecting the correlation of characteristics. This study proposes a consistency evaluation scheme based on information fusion, which comprehensively and accurately evaluates the consistency of battery packs in actual operation by integrating multiple factors, providing an effective guide for management optimization. Firstly, multi-dimensional consistency characteristics such as voltage, internal resistance, capacity, and temperature are comprehensively extracted, and a consistency score weighted by multiple characteristics is obtained through principal component analysis. Then, the score samples are optimized based on the Box-Cox transformation, and the consistency level is divided according to the normal distribution law. Finally, a mask-conformer deep learning model is constructed based on the characteristics of battery data to predict the consistency state. Experiments show that the proposed evaluation method can accurately distinguish the consistency state of batteries, and the mask-conformer model has excellent performance. It can directly predict from charging data without complex feature calculations, reducing the dependence on a large amount of operating data.