A hybrid remaining useful life prediction method for lithium-ion batteries based on transfer learning with CDRSN-BiGRU-AM

Abstract

The prediction of the remaining useful life (RUL) of lithium-ion batteries (LIBs), which are increasingly used in a wide range of applications, has been an important study. Existing techniques are difficult to strike a balance between prediction accuracy and execution time. To realize highly accurate RUL prediction in a short time, a hybrid RUL prediction method for LIBs was developed. The method first adopts a channel-wise deep residual shrinkage network (CDRSN) for adaptive extraction of input data feature, enhances important information features, and suppresses ineffective features according to the importance of feature information. Then, a bidirectional gated recurrent unit (BiGRU) was utilized to extract bidirectional temporal features of the processed data, and an attention mechanism (AM) was introduced to maximize the extraction of the important temporal mutual information in the data. Finally, a fully connected layer transfer strategy was applied to deploy the model from offline training to online prediction, which could prevent the problem of unstable prediction due to random initialization of the model and greatly improve the computational efficiency of the model. The simulation results indicated that the RMSE of the proposed method did not exceed 1.77% and the MAE did not exceed 1.44%. Both were smaller than the results in the other literature compared. Therefore, the proposed method could obtain accurate RUL prediction accuracy for LIBs online.