A Bi-directional Long-term Recurrent Network for Remaining Useful Life Prediction of Lithium Batteries

The remaining useful life (RUL) prediction of lithium batteries can effectively prevent battery failure. At present, the deep learning-based RUL prediction of lithium batteries has been widely used, but the influence of long-term dependence information of battery data on RUL prediction is rarely considered. To further improve the prediction accuracy of RUL, this paper uses Bidirectional Long Short-Term Memory (BiLSTM) model to predict the RUL of the lithium battery. BiLSTM can effectively capture the past, current and future information of battery data, and mine the degradation trend of battery from the continuously changing information flow, which can significantly improve the RUL prediction accuracy of the model. Applying the BiLSTM method to the NASA dataset, the experimental results validate its effectiveness, and the comparison shows that it outperforms existing DL-based methods.