Remaining Useful Life Prediction of Lithium-Ion Batteries Based on Health Indicator and Gaussian Process Regression Model

Abstract

Achieving accurate and reliable remaining useful life (RUL) prediction of lithium-ion batteries is very vital for the normal operation of the battery system. The direct RUL prediction based on capacity largely depends on the laboratory condition. A novel method that combines indirect health indicator (HI) and multiple Gaussian process regression (GPR) model is presented for the RUL forecast to solve the capacity unmeasurable problem of operating battery in this paper. First, three measurable HIs are extracted in the constant-current and constant-voltage charge process. Both the Pearson and Spearman rank correlation analytical approaches show that the correlations between HIs and the capacity are good. Then, the GPR model is optimized with combined kernel functions to improve the ability to predict capacity regeneration. Next, based on the measurable HI versus cycle number data, three GPR models are built, and HIs prognosis results are achieved at a single point. The HIs prediction results are added in the multidimensional GPR model, which is accomplished by using HIs and capacity as input and output, respectively. The predicted capacity is used to compare with the threshold to acquire the RUL prediction result. The approach is validated by the two different life-cycle test datasets. The results indicate that an accurate and reliable RUL forecast of lithium-ion batteries can be realized by using the proposed approach.