Two-Stage Remaining Useful Life Prediction for Lithium-Ion Batteries Using Adaptive Change-Point Detection

Abstract

Accurate prediction of the remaining useful life (RUL) of lithium-ion batteries is crucial for ensuring safe operation and efficient recycling. However, most existing RUL prediction methods rely on a single model, which makes it difficult to fully and accurately characterize the complex battery degradation process. Although some studies have adopted two-stage prediction approaches to improve overall performance, accurate change-point identification remains a key challenge. To address this issue, this article proposes a battery RUL prediction framework that integrates change-point detection based on the jump information criterion (JIC) with adaptive model selection. First, a robust median regression estimator (RMRE) is constructed, and a degradation mutation identification indicator (DMII) is designed to extract structural change features from the capacity degradation trajectory. Next, an adaptive JIC-based change-point detection method is developed to accurately identify key change points, enabling stage-wise segmentation of the battery degradation process over its entire lifecycle. Based on this segmentation, an integrated depth–error model selection strategy (IDEMS) is proposed to adaptively select the optimal prediction model for different degradation stages, thereby improving the accuracy and stability of RUL prediction. Finally, experiments conducted on the MIT and CALCE public datasets demonstrate that the proposed method outperforms existing approaches across multiple evaluation metrics, achieving high prediction accuracy and strong generalization capability.