Suitable feature selection for prediction of lithium-ion batteries remaining useful life

In this work, a classification and regression tree (CART) based extreme gradient boosting (XGBoost) model, simply denoted as XCART, is proposed to predict the lithium-ion batteries remaining useful life (RUL). First, the fluctuations in the RUL series are regarded as inherent characteristics, and voltage series which have the similar fluctuations are extracted as important health indicators (HI). Then, the HIs are selected by the feature importance indicator of XCART. After that, XCART is used to learn the selected HIs to acquire the RUL prediction results. Experiments on the NASA Prognostic Center of Excellence dataset show that the proposed XCART method has good prediction performance.