A Hybrid Data-Driven Method Based on Data Preprocessing to Predict the Remaining Useful Life of Lithium-Ion Batteries

Abstract

Abstract The estimation of remaining useful life (RUL) for lithium-ion batteries is an essential part for a battery management system. A hybrid method is presented which is combining principal component analysis (PCA), improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), sparrow search algorithm (SSA), Elman neural network (Elman NN), and gaussian process regression (GPR) to forecast battery RUL. First, in the data preprocessing stage, the PCA + ICEEMDAN algorithm is creatively proposed to extract features of capacity decay and fluctuation. The PCA method is used to reduce the dimensionality of the extracted indirect health indicators (HIs), and then the ICEEMDAN algorithm is introduced to decompose the fused HI sequence and actual capacity data into residuals and multiple intrinsic mode functions (IMFs). Second, in the prediction stage, feature data are corresponded one-to-one with the mixed model. The prediction models of SSA–Elman algorithm and GPR algorithm are established, with the SSA–Elman algorithm predicting the capacity decay trend and the GPR algorithm quantifying the uncertainty caused by the capacity regeneration phenomenon. The final prediction results are obtained by superimposing the two sets of prediction data, and the prediction error and RUL are calculated. The effectiveness of the proposed hybrid approach is validated by RUL prediction experiments on three kinds of batteries. The comparative experimental results indicate that the mean absolute error (MAE) and root mean square error (RMSE) of the presented prediction model for lithium-ion battery capacity are less than 0.7% and 1.0%.