An Improved RL Approach for Dynamic Pricing and Scheduling Control Considering EV-to-Aggregator Interaction

The increasing penetration and density of electric vehicles (EVs), especially the significant time distribution similarity between the EV charging/discharging behaviors and the base load curves, causes a great increase in electricity demand and challenges the peak-shaving policy. In light of the EVs characteristics of distributed energy storage and being regulated by electricity price, this paper proposes an improved reinforcement-learning (RL) approach to express the demand response (DR) interaction between EVs and EV aggregators (EVAs) in the vehicle-to-grid (V2G) market. The entire interaction system follows a hierarchical optimization framework. To that end, we developed a dynamic pricing strategy for EVA and a scheduling control method to describe EVs' response mechanism. Through simulations with real-world datasets, we verified the effectiveness of the strategies learned by this improved RL approach. The results demonstrated the earned benefits for both EVA/EVs (including additional discharging income) and the grid (including peak-shaving command).