CoRLNF: Joint Spatio-Temporal Pricing and Fleet Management for Ride-Hailing Platforms

In ride-hailing services, the imbalance between supply and demand is a big challenge for ride-hailing systems. In this paper, we propose the CoRLNF approach, which Combines multi-agent Reinforcement Learning and Network Flow model to implement joint Spatio-Temporal (ST) pricing and fleet management, and can effectively balance the supply and demand for ride-hailing platforms. Different from existing research adopting centralized pricing, our approach narrows the joint action space of agents based on a multi-agent setting and considers the expected future profits of drivers. First, we propose a multi-agent reinforcement learning algorithm to dynamically generate ST pricing strategy. Significantly, each agent utilizes supply-demand distribution and pricing strategies in neighboring areas, so that each agent has more information on urban traffic. Then, considering the future profit of drivers and supply-demand distribution, we design the relocation network flow model to achieve coordinated driver relocation. And the goal is to maximize the total net profit of the drivers. Finally, extensive experimental results demonstrate that our approach outperforms existing approach in terms of drivers' profits, order response rate and passengers' waiting time.