A multi-scale learning approach for remaining useful life prediction of lithium-ion batteries based on variational mode decomposition and Monte Carlo sampling

Abstract

Reliable and accurate prediction of remaining useful life for lithium-ion batteries has tremendous significance, since they can alleviate users' anxiety about mileage and safety. However, accuracy and reliability of remaining useful life prediction are affected by capacity regeneration and uncertainty quantification. In this study, we propose an approach to predict the remaining useful life of lithium-ion batteries, where multi-scale learning is developed to catch the uncertainty and capacity regeneration. Specifically, the multi-scale learning approach based on Gaussian process regression and dropout-Monte Carlo gated recurrent unit is applied to establish accurate prediction model with uncertainty quantification. Meanwhile, the optimal charging voltage interval is extracted with a high correlation coefficient. The variational mode decomposition is selected to multi-scale decompose the proposed health indicator as intrinsic mode functions and residual term. Finally, the observed data has been selected to verify the accuracy and robustness of the proposed method. Compared to the existing single data-driven methods, the proposed method can obtain high accuracy and strong robustness for RUL prediction with root mean square error limited below 3%.