Robust remaining useful life prediction for Li-ion batteries with a naïve bayesian classifier

Abstract

This paper presents an empirical procedure for predicting robust remaining useful life (RUL) using a naïve Bayesian classifier (NBC) with time as the response. The method is illustrated using public data for predicting Li-ion battery RUL to end-of-life (EOL). A battery life prediction is obtained using the capacity values up to the prediction time. The root mean squared error (RMSE) is used for performance evaluation. The predictions over time are compared with the actual time to EOL for each test battery and the RUL is calculated at four time intervals. The prediction performance of the NBC is compared with that of a support vector machine (SVM). The case study shows that the NBC generates competitive prediction performance, even though other factors contributing to Li-ion battery degradation are concealed. This method is also applicable to predicting RUL to end-of-discharge (EOD) and the failure prognostics for other components.