Optimizing electric vehicle battery health monitoring: A resilient ensemble learning approach for state-of-health prediction

Abstract

State of Health (SoH) prediction is critical for optimizing electric vehicle (EV) battery performance and longevity. This study proposes an Ensemble of Ensemble Models (EEMs) framework to enhance SoH prediction accuracy by combining ensemble learning methods—Random Forests, Gradient Boosting, and AdaBoost—using a stacking-based meta-learning approach. The method captures complex patterns in key input features such as voltage, temperature, and charge-discharge cycles. The approach was tested using a Li-ion battery dataset, with evaluation metrics including MSE, RMSE and R-squared. Results demonstrate that EEMs with 99.9 accuracy and nearly error-free predictions (RMSE of 0.00000025), validate the importance of advanced ensemble techniques in optimizing SoH prediction and outperform individual and conventional ensemble models, providing accurate and reliable SoH estimates. This framework offers practical implications for improving battery management, extending battery lifespan, and promoting energy sustainability in EV systems.