Classification and Multi-Stage Prediction of Electric Vehicle Travel Behavior Using Deep Learning

The increase in the Electric Vehicle (EV) adoption to reduce air pollution due to transportation is also increasing the EV power demand in the power distribution system. It also imposes other burdens like peak load increment and uncertainty in the load demand, primarily due to the uncertain behavior of the EV travel parameters like arrival time, departure time, and travel miles. In this paper, an EV aggregator is assigned, which will accurately classify and predict these travel parameters on a day-ahead basis using the proposed Deep learning-based technique. After clustering the input historical departure time, a deep Long Short-Term Memory (LSTM) model is proposed to classify the clusters. Then, two deep LSTM models are proposed for predicting each cluster's arrival time and travel miles. The proposed technique not only predicts the travel parameters accurately but also maintains the correlation between the parameters. The training results show the effectiveness of the method. Also, the R-squared value and Root Mean Square Error metrics are used to evaluate the models, and the R-squared value close to unity is achieved for each cluster.