Optimal V2G scheduling method for individual EV considering departure uncertainty and user behavior using reinforcement learning

To develop an optimal V2G (Vehicle-to-Grid) scheduling for electric vehicles (EVs), various factors such as the battery’s state of charge (SOC), driving cycle, and user benefits must be considered. However, due to the uncertainty associated with EV usag e, creating an accurate model remains a significant challenge. There are three key challenges in V2G scheduling: the uncertainty in EV departure times, the anxiety of EV users depending on the vehicle type, and the need for a long-term evaluation model. This paper addresses these challenges by developing a 2-Stage EV Uncertainty Reinforcement Learning (RL) model and proposes an optimal V2G scheduling strategy for individual electric vehicles. In the first stage, a prediction model is emplo yed to estimate the EV's departure time and travel distance for the next steps. In the second stage, the optimal target SOC for each vehicle is calculated by considering charging/discharging costs, battery degradation costs, and user satisfaction. As a result, the proposed model, by considering the individual characteristics of each vehicle, achieved the highest overall benefits, the highest user satisfaction, and a reduction in battery degradation costs in long-term evaluations. The 2-Stage RL model is practical and efficient, as it implements V2G scheduling based on real-world data.