关键词: 城市交通, 剩余使用寿命预测, 数据驱动, 电动汽车, 模型融合

Abstract: Predicting the battery remaining useful life (RUL) of electric vehicle (EV) is a hot topic in the field of battery research. Most of the existing RUL prediction models are based on a single prediction index, with low prediction accuracy and poor generalization. In this paper, a Stacking model of Gray Prediction model and Long-term Memory Neural Network model was developed to predict the RUL of electric vehicle with high accuracy based on real vehicle operating data. First, the movement and environmental parameters of the vehicle were extracted according to the influencing factors of the RUL of the battery, and the optimal features was selected as the model input based on the Random Forest Algorithm. Then, the study used the Auto Regressive Integrated Moving Average model to extend the selected features to overcome the limitation of time dimension. Based on the data characteristics, the Gray Prediction model and Long-term Memory Neural Network model were proposed to predict the battery RUL, and the prediction error was further reduced by the Stacking model fusion. The results show that the average relative error of the fusion model is 1.6%, and the average absolute error is 0.013, which proves a stable and reliable prediction of the RUL with the proposed model.

Key words: urban traffic, prediction of remaining useful life, data driven, electric vehicle, model fusion