Lifetime Early Prediction for Second-Life Lithium-Ion Battery: A Hierarchical Bayesian Learning Method

With the rapid growth of electric vehicle (EV) adoption, the widespread use of lithium-ion batteries has brought increasing attention to the echelon utilization of second-life batteries (SLBs). These SLBs, typically retired from EVs upon reaching 80% of their rated capacity, possess considerable potential for deployment in stationary energy storage systems. To fully exploit their remaining capacity, accurate lifetime prediction during second-life applications is essential. This study addresses the early lifetime prediction of SLBs by proposing a hierarchical Bayesian model (HBM). Considering the heterogeneity in degradation behavior under varying working conditions, the method initially performs clustering based on operating conditions. Subsequently, a two-level hierarchical Bayesian framework is constructed to model the relationships between battery groups and individual units, enabling more accurate lifetime estimation. Bayesian inference is employed to optimize model parameters using extracted features, thereby reducing predictive uncertainty and enhancing overall accuracy. Validation on a self-collected SLB aging dataset demonstrates that the proposed method outperforms conventional machine learning approaches, especially under data-limited conditions. Quantitatively, the model achieves an average root mean square error (RMSE) of 113.444 cycles (±31.423), mean absolute error (MAE) of 76.616 cycles (±15.516), and mean absolute percentage error (MAPE) of 17.188% (±4.073%) on the test set.