Economical vehicle-side strategies for an electric bus charging station in vehicle-to-grid services based on reinforcement learning

The power batteries of idle electric vehicles can be utilized as distributed energy storage units to deliver Vehicle-to-Grid (V2G) services, enabling bidirectional energy flow between the grid and EV batteries. Existing studies on the economic feasibility of V2G focus on conventional passenger vehicles, which have highly random idle times and lack scalability. The distinctiveness of this study is selecting a specific electric bus charging station in the Pearl River Delta region as the research object. This station provides comprehensive data on basic station information, bus batteries, charging, driving, bus departure times, and electricity prices. Compared to existing V2G economic strategy research, it demonstrates operational regularity and scalability, offering strong practical application value. First, this study collected data from bus charging stations covering 25 routes and 377 electric buses. A power battery electrical-health-economic model was developed to simulate electrical behavior, health degradation, and economic performance under various charging and discharging conditions. Then, the constraints of charging stations and electric buses are set according to operational scenarios. The charging station uses reinforcement learning to implement health-aware V2G (V2G-H). Furthermore, strategies including no V2G, uncoordinated V2G, deep-discharge V2G, and V2G-H are compared and analyzed in terms of economic performance. The results show that the reinforcement learning-based V2G-H strategy saves $1,539 in total cost per bus over the entire lifecycle and extends battery life by over 21 months compared to the traditional approach. Finally, this study analyzes the effects of electricity price fluctuations, departure frequency, and battery cost on the V2G strategy, providing a feasible solution for implementing V2G services at charging stations.