Abstract:

The prediction of remaining useful life （RUL） of lithium-ion power battery is of great significance for understanding the safety and reliability of electric vehicles in the whole life cycle and improving the design of battery management system. Generally， the time series prediction method based on deep learning is a recursive process. The error of each prediction will accumulate with the increase of prediction times， which is difficult to ensure the prediction accuracy and efficiency. Based on the theory of deep learning sequence prediction and error analysis， an ARIMA-EDLSTM fusion model is established for lithium battery remaining useful life prediction. The encoder decoder （ED） framework is used to improve the long short-term memory neural network model （LSTM）， establish the EDLSTM model of sequence to sequence prediction， and fuse the ARIMA model to predict the error trend and modify the prediction results. Theoretical analysis and real vehicle data verification show that this method can still better fit the real vehicle SOH decline curve when the prediction proportion exceeds 35% of the total history data， and effectively improve the prediction accuracy of the remaining useful life of lithium battery.

Key words: lithium-ion power battery, remaining useful life prediction, data driven, deep learning