Remaining useful life prediction of lithium batteries based on jump- connected multi-scale CNN

In order to better utilize the feature information obtained by all convolutional layers in Convolutional Neural Networks (CNN), this paper proposes a lithium-ion battery remaining life prediction model based on jump connected multi-scale CNN. The proposed model takes the health factor of the battery as input, and uses the multi-scale CNN model based on jump connection to extract the local feature information and global feature information of the health factor of lithium-ion battery at different scales. Then, all the local feature information and global feature information are fused through the information fusion module, and finally the predicted remaining useful life is output. The experimental results show that the proposed method can predict the remaining useful life of lithium-ion batteries more accurately. Compared with the classical CNN method, Bi-LSTM method, EMD-LSTM method and VMD-GRU method, the root mean square error (ERMSE) of the proposed method is reduced by 75.7%, 78.3%, 83.8% and 77.8%, respectively. Mean absolute error (EMAE) decreased by 80.7%, 80.9%, 86.8%, 82.3%, and mean absolute percentage error (EMAPE) decreased by 81.0%, 82.2%, 87.0%, 83.1%, respectively. The model determination coefficient (R2) increased by 17.4%, 23.2%, 44.5% and 25.8%, respectively.