Abstract: In response to the difficulties of online obtaining enough direct performance parameters for lithium battery remaining useful life (RUL) prediction, a novel prediction method that combines the long short-term memory (LSTM) neural network and an improved GM(1, N) model is proposed in this paper. Firstly, two correlation analysis methods are used to obtain the health indicators that characterize the health status of the lithium battery. Then, the existing lithium battery dataset is used to train the LSTM model and predict the initial capacity of the same type lithium battery in operation. Finally, the initial capacity and corresponding indirect health indicators are substituted into an optimized damping accumulation discrete multi-variable convolution model, and accurate prediction of the entire life cycle of lithium battery is achieved through rolling technique. Experiments are conducted by using for lithium battery datasets from the Center for Advanced Life Cycle Engineering at the University of Maryland. The results show that the proposed model performs much better than the LSTM model and the conventional GM(1, N) model. In particular, the mean absolute percentage errors of the proposed method for predicting the RUL of four lithium batteries are less than 2.5%, the root mean square errors are about 0.02, and the mean absolute errors are only about 0.01. 

Key words: Lithium battery , RUL prediction , LSTM , Improved GM (1,N) model