Remaining Useful Life Prediction of Lithium-Ion Batteries Using Multi-model Gaussian Process

Abstract

This paper presents a multi-model Gaussian process (MMGP) approach for predicting the RUL of lithium-ion batteries. The proposed MMGP approach incorporates multiple candidate capacity fade models, as the trend functions of a Gaussian process (GP) model, to capture the multi-stage capacity fade trend of the batteries. First, the hypothetical capacities at a predefined number of future cycles are predicted at the current cycle based on the offline data using similarity-based extrapolation. Then, the active fade model is selected by comparing the hypothetical capacities with the GP model-projected capacities using each candidate fade model and the active model is employed as the trend function of the GP model. Finally, the distribution of the RUL is estimated by determining when the projected capacity curves using the GP model down-cross a pre-defined capacity threshold. The MMGP approach was used for the RUL prediction of eight lithiumion battery cells that show multi-stage capacity fade behavior when cycled with a daily current rate (i.e., C/24). The capacities of these cells initially degrade rapidly, followed by a reduced fade rate and then a faster linear fade rate. The RUL prediction results suggest that the proposed MMGP approach can adaptively select proper trend functions for the GP model at different capacity fade stages throughout the lifetime, as well as adapting the models to accommodate variations in the capacity fade performance among different cells.