A Deep Reinforcement Learning-Based Strategy for Optimizing V2G Energy Scheduling on Supply and Demand Sides

With the rapid development of new energy vehicle technology, there is a need for vehicle-to-grid (V2G) as energy management and storage. In this study, a novel V2G scheduling method using the Deep Q Network (DQN) algorithm is introduced, aiming to optimize the charging and discharging usage needs of both V2G supply and demand, while jointly reducing the operating costs of the grid and electric vehicle users. The suggested approach tackles the issues of high-dimensional state-action spaces in V2G systems by the integration of a dual learning mechanism and a priority replay method. A V2G experimental platform was established utilizing real city grid data, incorporating 100 electric vehicles, and the method was assessed under diverse load circumstances and levels of renewable energy integration. The results indicate a $\mathbf{1 5 \%}$ enhancement in peak shaving on the demand side and a $10 \%$ decrease in user expenses relative to traditional rule-based approaches, concurrently improving system stability. This study offers a significant theoretical and technical foundation for the implementation of smart energy management in vehicle-to-grid applications.