Dual-driven battery life prediction based on data augmentation and capacity recovery identification

Accurate prediction of the remaining useful life (RUL) of lithium-ion batteries is a critical prerequisite for predictive maintenance, optimizing charge-discharge strategies, and ensuring their long-term safe operation. However, during the cell resting, side reactions among cell components, especially on electrolyte/electrode interface, lead to an unwanted temporary capacity recovery (CR). This phenomenon poses significant challenges to the accurate prediction of RUL. This study proposes a RUL prediction framework that incorporates dynamic identification of CR and data augmentation to concurrently address capacity fluctuation and few-shot scenario. In this study, input data purification is achieved through an innovative quasi-online identification and CR quantitative correction strategy, and training sample expansion is accomplished by conducting strategic data augmentation based on early-cycle data. The proposed framework can significantly improve the prediction accuracy of RUL for application scenarios with CR interference and limited early-cycle data.