Abstract:
Aiming at the noise interference and the improvement of prediction accuracy in the prediction of remaining useful life (RUL) of lithium-ion batteries by data-driven methods, we propose a lithium-ion battery RUL prediction method combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and tightly-coupled convolutional Transformer (TCCT) model based on multi-head information-entropy sparse attention (MESA). Firstly, we select the battery capacity decline trend as the health factor, and decompose the battery capacity data by CEEMDAN to reduce the impact of capacity regeneration and noise. Secondly, we use the MESA mechanism to reduce the computational complexity and optimize the attention weight distribution. Finally, we use the improved TCCT model to deeply mine the decomposed residuals and components to construct a more accurate lithium-ion battery RUL prediction model. In order to verify the effectiveness of the method, we use the public data sets of NASA and CALCE for verification. We compare and analyze the CNN-Transformer model, CNN-LSTM model and Informer model. The error index of the proposed method is reduced. Experiments show that the proposed method can more effectively predict the RUL of lithium-ion batteries.