Hybrid approach of iterative updating for lithium-ion battery remaining useful life estimation

Abstract

Remaining Useful Life (RUL) prediction plays a critical part in many battery-powered applications. Statistical filter, i.e., particle filter (PF) is widely used to predict RUL with various models as well as its uncertainty representation. However, PF commonly used suffers from the lack of poor adaption of long-term prediction and iterative prediction. This disadvantage may further reduce the RUL estimation performance. To overcome this difficulty, this paper proposes a hybrid approach with dynamic updating for lithium-ion battery RUL estimation. The estimation results based on data-driven model of long-term degradation trend estimation are used as the observation value for regularized PF (RPF) to obtain the optimal estimation. Moreover, this optimized estimation value is utilized as the update online input to dynamically train the data-driven model, to improve the iterative predicting capability. The proposed approach comprises two ideas: (i) a dynamic updating strategy to predict the capacity of Li-ion battery and (ii) a modified combination of regularized particle filter and ND-AR (Nonlinear Degradation-AutoRegressive) model for accurate and stable RUL estimation. Experiment results suggest that the proposed approach, as a dynamic updating method combined with data-driven and empirical models, achieves better performance on both estimation accuracy and uncertainty representation.