Graph Neural Network Reinforcement Learning Based Coordinated Charging Optimization of Multiple Electric Vehicle Charging Stations

With the increasing number of electric vehicles (EVs) and the widespread establishment of EV charging stations (EVCSs), the grid faces significant challenges, drawing attention to the system’s economy and safety. To reduce charging costs while ensuring safety and meeting user demands, we have established a coordinated optimization charging scheduling system for multiple EVCSs based on graph neural network reinforcement learning. The scheduling problem is modeled as a markov decision process, and EVCS decisions are solved using multi-agent proximal policy optimization (MAPPO), with the final charging and discharging plan generated based on the urgency of user demand. A graph attention network (GAT) is introduced to extract coupling information between EVCSs, enhancing the feature representation of reinforcement learning and improving the accuracy of system scheduling. Experimental results show that this method achieves greater cost reductions while ensuring the safe operation of the power system.