Multi-agent deep reinforcement learning approach for EV charging scheduling in a smart grid

As the competitive advantages of electric vehicles, both in terms of operating costs and eco-friendly characteristics have gained attention, the demand for electric vehicles has increased, and studies for efficiently charging electric vehicles are being actively conducted. Previous studies have mainly focused on scheduling one electric vehicle visiting a charging station or scheduling multiple electric vehicles in a centralized execution method. However, a decentralized execution method that can schedule multiple vehicles according to their status is more suitable in a realistic smart grid charging environment that requires quick decisions. Therefore, we propose a multi-agent deep reinforcement learning approach with a centralized training and decentralized execution method that can derive charging scheduling for each electric vehicle. Computational experiments show that the proposed approach shows desirable performance in minimizing the operating cost of electric vehicles.