Online remaining useful life prediction of lithium-ion batteries using bidirectional long short-term memory with attention mechanism

As battery management systems are widely used in industrial applications, it is important to accurately predict the online remaining useful life (RUL) of batteries. Due to side reactions, the battery will continue to decline in capacity and internal resistance throughout its life cycle. Additionally, battery systems require reliable and accurate battery health diagnostics and timely maintenance and replacement. To obtain accurate RUL prediction, we propose a bidirectional long short-term memory with attention mechanism (Bi-LSTM-AM) model to predict online RUL by continuously updating the model parameters. In this study, normalized capacity was used as state of health (SOH). Multi-step ahead prediction using a sliding window method was used to obtain the SOH estimates. Six cylindrical and prismatic lithium-ion (Li-ion) batteries were used to evaluate the performance of the proposed model. Using our online RUL prediction model, the relative errors for the six Li-ion batteries are 0.57%, 0.54%, 0.56%, 0%, 1.27% and 1.41%, respectively. To evaluate the reliability of the proposed model, the prediction interval for the RUL prediction is also provided using the Monte Carlo dropout approach.