Graph-based multi-agent reinforcement learning with an enriched environment for joint ride-sharing and charging optimization

With the rapid growth of electric taxis, charging demand has become increasingly concentrated in specific urban areas due to the suboptimal pricing strategies. Meanwhile, ride-matching and charging processes are inherently coupled across temporal and spatial dimensions, making their joint coordination critical for boosting driver incomes and balancing energy loads. To tackle these challenges, we propose a graph-based multi-agent reinforcement learning strategy that incorporates an enriched environment for the joint optimization of ride-sharing and charging decisions. In our framework, electric vehicle charging stations, characterized by competitive and cooperative relationships that depend on their charging station operators, are regarded as reinforcement learning agents. The charging market is represented as a dynamic heterogeneous graph, which captures the interactions between charging stations from station-centric and query-centric perspectives. Finally, extensive simulation is performed, demonstrating the effectiveness of the control framework in balancing charging load distribution and boosting service revenue across stations compared to the baseline algorithms. The proposed control method with the ride-sharing process embedded into the environment optimizes the urban taxi distribution, expands the service coverage area, and enhances the overall driver earnings.