Remaining Useful Life Prediction of Lithium-Ion Batteries Based on Data Preprocessing and Improved ELM

Abstract

Accurate prediction of the remaining useful life (RUL) of lithium-ion batteries (LIBs) can provide an important reference for the safe operation of LIBs. At present, there are some problems in the prediction of RUL of LIBs, such as redundancy or deficiency of characteristic parameters, local regeneration in degradation characteristics, and poor stability of single model prediction. In this paper, an RUL prediction method combining kernel principal component analysis (KPCA), improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), whale optimization algorithm (WOA), and extreme learning machine (ELM) is proposed. Firstly, in the data preprocessing stage, the KPCA method is used to reduce the dimension of the extracted indirect health indicators (HI), and then the ICEEMDAN algorithm is introduced to eliminate the local regeneration phenomenon in the fusion HI sequence. Secondly, in the model construction stage, a capacity prediction model based on WOA-ELM and a fusion HI prediction model are built to predict the intrinsic mode functions (IMF) components of the fusion HI sequence. The prediction results are superimposed and then input into the capacity prediction model to realize RUL indirect prediction. Finally, the proposed method is verified by using NASA battery degradation data sets at different temperatures and experimental environments. The results show that the RUL prediction accuracy of the proposed method is greatly improved, and it has good stability and robustness.