A RUL Prediction Method of Lithium-ion Battery Based on Extreme Learning Machine

Abstract

Owing to the complexity of the electrochemical reaction types in lithium-ion batteries and the characteristics of nonlinear degradation, using physical models to describe the degradation process is often not accurate enough. This paper only analyzes from the perspective of data, through a data-driven method, extreme learning machine ( ELM ), to predict the remaining useful life (RUL) of lithium-ion. Firstly, the cycle charge and discharge data of lithium-ion battery are analyzed in depth to find the parameters which can indirectly characterize the degree of degeneracy of lithium-ion battery. The parameters with strong robustness to the actual capacity of lithium-ion battery are selected as indirect health indicators by correlation coefficient analysis. Then, the RUL prediction model based on ELM is constructed, which is trained and tested using the lithium-ion battery data sample set published by NASA. The results show that the RUL prediction method based on ELM can effectively predict the battery RUL, and the root mean square error (RMSE) can reach below 5 %.