Abstract:

Smart electric vehicle charging stations (EVCSs) with distributed energy resources (DERs), such as photovoltaic systems and energy storage systems (ESSs), are crucial for increasing profits and maintaining stable distribution grid operations. However, prediction errors of PV generation outputs and EV loads can decrease profits and destabilize the distribution grid. To address this issue, a two-stage framework for smart EVCS scheduling and inverter-based Volt-VAR control (VVC) using prediction error-integrated deep reinforcement learning (DRL) is proposed. In the first stage, EVCS agents train their neural network models to maximize profit through day-ahead charging/discharging scheduling of ESSs while responding to prediction errors. In the second stage, the VVC agent trains its neural network model to minimize real power loss and voltage violations through real-time reactive power scheduling of ESSs in EVCSs via their inverters. The proposed approach outperforms DRL methods that do not consider prediction errors in terms of profitability and reduction of real power loss.