A novel remaining useful life prediction method for lithium-ion battery based on long short-term memory network optimized by improved sparrow search algorithm

The remaining useful life (RUL) estimation is one of the key functions of lithium-ion battery management systems (BMS). After the battery reaches its end-of-life (EOL), its capacity decreases rapidly and it is prone to failure, which affecting the operation of equipment and even causing safety accidents. In addition, part of the user may prematurely replace the battery for the safety of battery use, resulting in a waste of battery resources. Therefore, the accurate RUL prediction can avoid both many safety accidents and the waste of resources, which is a key and challenging problem. Accordingly, a novel RUL prediction method based on long short-term memory (LSTM) network optimized by improved sparrow search algorithm (ISSA) for lithium-ion battery is proposed in this paper. Firstly, the hyperparameters of LSTM which need to be optimized are selected since they directly affect the prediction accuracy. Then, according to the battery capacity data of different datasets, the hyperparameters of LSTM are optimized by ISSA to achieve RUL prediction. Finally, the proposed RUL prediction method is respectively compared with the support vector regression (SVR), convolutional neural networks (CNN), recurrent neural network (RNN) and LSTM. The experiment results show that the proposed RUL prediction method is more accurate and robust which contributes to the rational use of lithium-ion battery to a higher degree.