Joint Prediction of SOH and RUL for Lithium Batteries Using CPO-BP-AdaBoost and GPR Models

To accurately assess the State of Health (SOH) of lithium batteries (LBs), this study proposes a prediction methodology for SOH and RUL that leverages both the CPO-BP-AdaBoost and Gaussian Process Regression (GPR) models. Initially, multiple feature parameters are extracted from the charge-discharge curves, followed by selection and dimensionality reduction to derive the Indirect Health Factor (IHF). Subsequently, the Crested Porcupine Optimizer (CPO) is utilized to optimize a Back Propagation (BP) neural network, which is then integrated into an AdaBoost ensemble as a weak learner to establish the CPO-BP-AdaBoost model for SOH prediction. To mitigate the impact of capacity regeneration phenomena on RUL prediction, IHF is employed as an input for the GPR model. Finally, the IHF predicted by the GPR model is used as an input to the trained CPO-BP-AdaBoost model to predict RUL. The proposed methods are validated through experiments using NASA’s battery dataset. Results indicate that the optimization of feature extraction, while preserving feature information, significantly enhances the model’s operational speed. In three groups of battery SOH prediction experiments, the model’s MAE is below 0.5%, and RMSE is below 0.8%, indicating a substantial improvement in SOH prediction accuracy. Similarly, for RUL prediction, the average error in remaining cycle predictions remains within three cycles.