Second-Life Lithium-Ion Battery Lifetime Early Prediction Based on Hierarchical Bayesian Model

The rising adoption of electric vehicles has led to a significant increase in retired lithium-ion batteries, creating substantial economic value in their echelon utilization. Accurate lifetime early prediction of retired batteries is important to ensure the safety during their second-life applications. Lifetime early prediction of second-life batteries (SLBs) is challenging due to the large diversity among SLBs, even under the same working conditions. In this paper, we propose a novel hierarchical Bayesian learning framework for SLBs lifetime early prediction, using only condition monitoring data from the first 50 cycles. In this framework, 4 features highly related to degradation are extracted from the early aging data. The selected features are then fed into a hierarchical Bayesian model for SLBs lifetime early prediction. Additionally, batteries under similar working conditions are grouped using the k-means clustering algorithm. Each group has its own hierarchical Bayesian model (HBM), allowing the predictions to account for the specific characteristics of the batteries under various working conditions. The effectiveness of the proposed framework is demonstrated using a self-tested dataset that included 84 SLBs under 21 different working conditions, all with the capacities around 85% of their nominal capacity. The proposed method achieves a mean absolute percentage error of 13.34%, which performs better than traditional regression model.