Early Prediction of End‐of‐Life and Nonlinear Degradation Knee‐Point for Lithium‐Ion Batteries Using Newton–Raphson‐Based Optimizer‐Optimized Extreme Gradient Boosting Tree‐Random Forest Model

Lithium‐ion batteries demonstrate complex nonlinear degradation patterns during extended operation, with capacity decay rates accelerating markedly after the knee‐point (KP) until reaching end‐of‐life (EOL). This study presents a novel methodology for early prediction of both EOL and KP using initial cycling data. The approach begins with feature selection through Spearman correlation analysis and random forest (RF) importance evaluation to identify key multidimensional indicators. A Newton–Raphson‐based optimizer then optimizes hyperparameters for an integrated model combining extreme gradient boosting tree (XGBoost) and RF, substantially improving predictive accuracy. The optimized ensemble model (XGB‐RF) achieves superior performance in predicting both EOL and KP degradation, with mean absolute percentage errors of 6.5% and 6.7%, and root mean square errors of 77 cycles and 53 cycles, respectively, in test set validation. These results surpass the performance of individual models. Leveraging the strong correlation between KP and EOL, the study incorporates predicted KP values as additional features to enhance the original feature set, further improving prediction accuracy. This method offers a novel perspective for battery degradation prediction and provides a theoretical foundation for optimizing battery performance and formulating maintenance strategies in practical applications.