Accurate Prediction of Lithium-Ion Batteries Lifetime Based on PDE-EHO-XGBoost

Prognostics and health management (PHM) will ensure the stable and reliable operation of lithium-ion battery systems. It is essential to predict battery lifetime accurately and explainably for practical quality assessment and long-term operation planning. This paper proposes a PDE-EHO-XGBoost model based on feature selection and an Extreme Gradient Boosting (XGBoost) with Elephant Herding Optimization Algorithm (EHO). Firstly, 63 features related to capacity decay are extracted from battery charge and discharge cycle data. Secondly, to improve the reliability and stability of model prediction, the Pearson correlation coefficient and differential evolution algorithm are combined for key feature selection and redundancy reduction. Then, to better optimize the hyperparameters and structure of the model, the EHO algorithm is incorporated into XGBoost to avoid convergence to local optima. Finally, the model is trained and verified using real battery cycle life data, and the importance of key features is visualized by SHAP. Experimental results show that the proposed model outperforms other prediction models, with its evaluation indicators (R2, RMSE, MAE) improved by at least 4. 2 2% 3 4. 1 5%, and 1 9. 0 6%, respectively.