A Bayesian method for capacity degradation prediction of lithium-ion battery considering both within and cross group heterogeneity

Accurate assessment and prediction of lithium-ion batteries’ (LIBs’) state-of-health (SOH) are crucial for the safety and timely maintenance of LIB-powered systems. During the aging process, due to the variation of initial health status and working conditions, the degradation paths of different LIBs usually manifest a high level of heterogeneity, including within and cross group heterogeneity. However, existing methods usually neglect the heterogeneity among different groups and put LIBs together for model development, leading to inferior prediction performance. To address this gap, a Bayesian estimation method is proposed to model both within and cross group heterogeneity. Specifically, the Gaussian mixture model (GMM) is employed to automatically separate LIBs into different groups, and, within each group, the linear mixed effects (LME) model is further applied to model the degradation path of each LIB. This leads to an LME model with mixture prior (LME-MP), which can effectively capture both within and cross group heterogeneity to provide accurate degradation prediction for the target battery. The accuracy and superiority of the proposed method are validated through the comparison with different benchmark methods using LIBs from the NASA data repository under randomized cycling conditions.