A Novel Wavelet-based Deep CNN-LSTM Model for Electric Vehicle Travel Behavior Modelling

The large-scale integration of Electric Vehicles (EVs) into the transportation sector represents a significant stride towards environmental sustainability. However, this transformation also presents a new challenge for power distribution systems. EVs, with their inherently variable travel patterns, significantly impact grid demands, leading to unpredictable peak loads and heightened overall strain. To address the inherent difficulty in forecasting EV travel patterns due to their variability, this study proposes a novel approach by an EV aggregator. We proposed a deep learning model utilizing a wavelet transform (WT) coupled with a convolutional neural network-long short-term memory (CNN-LSTM) architecture. This method tackles the challenge by first clustering historical trip end times and then employing separate WT-based CNN-LSTM models within each cluster. One model forecasts arrival times, while the other predicts trip distances. This unique approach goes beyond simple accuracy by explicitly considering the interrelated nature of travel parameters, a feature often neglected by conventional models. The model’s effectiveness is validated by high training R-squared values approaching 1, signifying exceptional prediction accuracy. Moreover, comparative analysis with established deep learning techniques confirms the superiority of our WT-based CNN-LSTM model, suggesting its potential for broader applications in power distribution.