Probabilistic Electric Vehicle Charging Demand Forecast Based on Deep Learning and Machine Theory of Mind

Abstract

Electric Vehicles (EVs) and corresponding charging stations have been widely popularized, increasing the power grid's operational risk and pressure, especially for the distribution network. Accurate EV charging demand forecast can potentially benefit the market through real-time robust scheduling. This paper proposes a deep-learning-based method for short-term probabilistic EV charging demand prognostics, which forecasts the quantiles of future charging demand of a charging station 5 minutes ahead. Plug-in EVs' charging behavior mainly depends on two crucial factors: (1) the user's living habits, which usually take a week as a cycle and can be extracted from the historical charging behaviors; (2) The user's stochastic behavior at the current timestamp, reflecting the short-term trend of charging demand variation, which is the difficulty of the short-term charging demand forecast. The proposed model has taken both the above historical charging habits (regularities) and the current trend of charging demand variation into consideration based on the paradigm of the Machine Theory of Mind (MToM), and two case studies on real EV charging demand datasets have verified its superiority over state-of-the-arts.