Coordinated Ride-hailing Order Scheduling and Vehicle to Grid for Autonomous Electric Vehicles Based on Independent Proximal Policy Optimization

Autonomous electric vehicles (AEVs) fleet will bring about tremendous change in the future ride-hailing system, with huge potential to increase operational efficiency in transportation networks and provide vehicle-to-grid (V2G) services in power networks. However, the uncertain spatio-temporal distribution of orders increases the dispatch difficulty. The time-varying charging and discharging prices among charging stations (CSs) exert further challenges in coordinating the order demand, V2G supply, and cost-saving purposes. This paper introduces a novel scheme for coordinated ride-hailing order scheduling and V2G for AEVs through the independent proximal policy optimization (IPPO) method. The model divides the AEV fleet into multiple parts and matches them to CSs through the optimal matching scheme. Then each CS is regarded as an agent, and through multi-agent reinforcement learning, it decides to charge and discharge corresponding parts of the vehicle. The proposed model strategically enables AEVs to charge at lower prices and discharge at higher ones while considering the balance of orders and V2G services at the same time, thus lowering operational costs and obtaining higher revenue. Numerical experiments are conducted to verify the superiority of the proposed method through a New York City zone-based power-transportation coupled network. The results validate the effectiveness of the proposed framework in better financial performance of the AEV fleet with higher computational efficiency.