Remaining useful life prediction of lithium-ion batteries based on attention mechanism and bidirectional long short-term memory network

Abstract

Accurate prediction of the remaining useful life of lithium-ion batteries is beneficial to prolong the battery's life and increase safety. However, there is insufficient historical data available because of the limitation of charge and discharge cycles of lithium-ion batteries. To enhance the prediction accuracy, a hybrid model based on an attention mechanism and a bidirectional long short-term memory network is proposed, which can acquire good prediction results with early data of the target battery. Multiple temporal features, which strongly correlate with the capacity considered the main feature of battery health, are used as input to train and test the proposed model. The attention mechanism is added in the bidirectional long short-term memory network to precisely decide the attention distribution of multiple features to extract useful information. It helps the bidirectional long short-term memory networks to make a better prognosis and overcome the vanishing and exploding gradient problems. The experimental results show that the proposed framework can decrease the uncertainty for multi-step prediction and outperforms other machine learning models in prediction accuracy.