Comparison of deep learning approaches for electric vehicle load curves forecasting

Electric Vehicles (EVs) are becoming increasingly prevalent in modern transportation systems. Their fast penetration is followed by a need for accurate forecasting of EV load curves, a process particularly important for market entities like electricity suppliers and/or electric vehicle aggregators that participate in wholesale electricity markets. In this paper, two deep learning forecasting models, which have been proven effective for time-series forecasting, the Temporal Convolutional Networks and the Temporal Fusion Transformer are developed and evaluated for EV load curves forecasting. Their forecasting performance is compared with the Persistence baseline model using common forecasting error metrics. Furthermore, an ensemble methodology is proposed, which combines the deep learning models and the baseline model to improve the overall performance of the predictions. The performance of the examined forecasting models is evaluated on a publicly available EV dataset and results are presented and discussed.