Abstract: Research on lithium-ion battery lifetime prediction is a hot topic, but the complex degradation process involving multiple cells complicates the task. Current studies often have limited sample sizes and rely heavily on future current and voltage data, making true prediction impractical for real-world applications. Additionally, the extensive use of machine learning methods has caused a significant strain on computational resources. Therefore, significantly reducing the computational cost of prediction methods while ensuring prediction accuracy is of great practical significance. To address this, we first designed experiments to study the degradation process between individual batteries and battery packs. Then, we proposed a Lebesgue sampling-based migration-driven lifetime prediction method for lithium batteries, transitioning from individual cells to groups. This method introduces Lebesgue sampling techniques to extract features and construct differential models, optimizing neighborhood differences between the target domain and the source domain, and efficiently computing the probability distribution of the arrival time at the Lebesgue limit. The method trains on degradation data from individual batteries, successfully predicting the failure life of the battery pack. The proposed method achieved a prediction error of around 1.5% in both early and real-time prediction scenarios, while also achieving computational efficiency in the millisecond to second range, significantly surpassing many existing mature lifetime prediction methods in both accuracy and efficiency.

Key words: lebesgue sampling, life time prediction, battery packs, remaining useful life, computing efficient