A naive Bayes model for robust remaining useful life prediction of lithium-ion battery

• Robustness of RUL predictions for lithium-ion batteries is analyzed quantitatively.
• RUL predictions of the same battery over cycle life are evaluated.
• RUL predictions of batteries over different operating conditions are evaluated.
• Naive Bayes (NB) is proposed for predictions under constant discharge environments.
• Its robustness and accuracy are compared with that of support vector machine (SVM).

Online state-of-health (SoH) estimation and remaining useful life (RUL) prediction is a critical problem in battery health management. This paper studies the modeling of battery degradation under different usage conditions and ambient temperatures, which is seldom considered in the literature. Li-ion battery RUL prediction under constant operating conditions at different values of ambient temperature and discharge current are considered. A naive Bayes (NB) model is proposed for RUL prediction of batteries under different operating conditions. It is shown in this analysis that under constant discharge environments, the RUL of Li-ion batteries can be predicted with the NB method, irrespective of the exact values of the operating conditions. The case study shows that the NB generates stable and competitive prediction performance over that of the support vector machine (SVM). This also suggests that, while it is well known that the environmental conditions have big impact on the degradation trend, it is the changes in operating conditions of a Li-ion battery over cycle life that makes the Li-ion battery degradation and RUL prediction even more difficult.