Degradation Prediction of Li-ion Batteries based on Linear Mixed Effects Model with Mixtrure Prior

Degradation modelling and prediction of a li-ion battery (LIB) is essential for its health management, which can guide timely maintenance to avoid catastrophic consequences. There are many methods have been developed to achieve this end, among which data-driven methods have received extensive attention and realized promising results. However, exiting methods do not take the clustering behavior of different degradation paths into consideration, while ignoring the difference among various clusters can lead to deteriorated prediction performance. To address this research gap, a linear mixed effects model (LME) with mixture prior, i.e., MP-LME, is proposed in this study. Specifically, a multinomial distribution is introduced to describe the probability that a battery belonging to each cluster; then, within each cluster, an LME model is proposed to simulate the degradation path of LIBs in this cluster; lastly, model parameters are updated with the observed data of a new battery to predict its future degradation path. This model can generate accurate degradation prediction for a LIB by leveraging valuable information of LIBs in clusters with similar degradation behavior and avoiding negative effect due to dissimilar fading characteristics. The performance and superiority of the proposed method are validated with a degradation dataset of LIBs under randomized loading profiles.