Trade-Off Selection of Data-Driven Methods for EV Demand Forecasting in a Real Office Environment

As the EV market grows, smart charging is needed to maximize users' benefits and comfort, and mitigate the impact on the distribution network. According to current communication protocols, the state of charge and charging demand information of EVs are not known, however, these are of great importance to smart charging algorithms. To solve this problem, we apply data-driven-based methods for energy demand forecasting of EVs in a real-world office building case in Belgium. In this work, we evaluate, compare, and assess both statistics modeling and machine learning methods to predict the EV energy demand. We analyze the impact of various input features, including historical energy consumption, time information, car type and weather information. The demand forecasting results demonstrate the performance of different data-driven technologies and guide the selection of the best forecasting method in various situations such as limited / extensive information. The finding is of particular use to support the charging schedule planning in office environments for future smart charging services.