An Efficient Hybrid Deep Learning Approach for Accurate Remaining EV Range Prediction

The accurate remaining Electric Vehicle (EV) range estimation is necessary to overcome EV users' range anxiety and infrastructure limitations. However, the traditional methods of EV Remaining Driving Range (RDR) estimation assumes the vehicle speed and energy consumption are consistent with the profiles in the recent history. But in the real world, the driving mode changes rather dynamically according to the user's speed profile, which significantly impacts RDR. Thus, the key question to be addressed in this work is how to accurately predict RDR considering the variation of the user speed profile during the driving trip. So, this work proposed a hybrid deep learning approach for accurate RDR estimation, where the future speed is then updated according to the average speed predicted in a 15-min prediction window. The deep learning approach combines a convolutional neural network (CNN) with Long Short-Term Memory (LSTM) to predict the remaining range of EVs based on historical EV speed data. The proposed CNN-LSTM-hybrid model is trained by exploiting the historical driving data of about 50 users in a two-week test-drive period. The test performance of the proposed EV range estimator is validated using real-world driving data that shows the high accuracy of RDR prediction with an average error of 3.762 km in a testing time window of 7.5 hours. The test results demonstrate the effectiveness of the proposed approach in the EV speed profile prediction, and thus RDR estimation with a high accuracy.