Segmented Prediction Model of Lithium-ion Battery Remaining Useful Life Based on LSTM and SVR

As an important part of battery health management, accurate prediction on remaining useful life of lithium-ion battery can maintain and replace the batteries in advance to guarantee the safety and stability of the energy storage system. Most extant models use a single model throughout the entire stage for prediction, which may lead to the decrease of the accuracy for subsequent prediction due to insufficient learning, as well as disability to predict battery remaining useful life with data from other batteries. To solve these issues, a segmented prediction model of lithium-ion battery remaining useful life is proposed. In this method, the knee point of the capacity decline curve is detected and then long short-term memory and support vector regression models are used to predict the remaining useful life before and after the knee point, respectively. The experimental prediction results represent that the proposed method achieves higher prediction accuracy compared with long short-term memory method.