Comparing deep learning methods to predict the remaining useful life of lithium-ion batteries

Abstract

The prediction lifetime of Lithium-ion (Li-ion) batteries can be used as an early warning system to prevent their failure, which makes them very significant to ensure their safety and reliability. In this paper, we suggest a comparative study of four neural networks, i.e. Deep Neural Networks (DNN), Recurrent Neural Networks (RNN), Gated Recurrent Unit (GRU), and Long short-term memory (LSTM), to predict and improve the accuracy of remaining useful life (RUL) of Li-ion batteries. The results of performance prediction were assessed using two statistical indicators, i.e. MAE and RMSE, to demonstrate the superiority of the proposed prediction method among themselves and compared with other papers' methods. Experimental validation is performed using the Li-ion battery datasets extracted the NASA and the CALCE. The LSTM method proves its effectiveness in reducing the prediction error and achieving good performance results of RUL prediction compared to other methods.