Online Car-Hailing Order Matching Method Based on Demand Clustering and Reinforcement Learning

Huifei Gao, Tian Chen, Jingxiang Hao

Abstract: The development of online car-hailing platforms has addressed the limitations of traditional taxis in meeting passengers’ personalized travel needs. However, the online car-hailing model encounters challenges such as the expense of detours for drivers and the optimization of order matching. We propose a car-hailing order-matching approach based on passenger demand clustering and reinforcement learning to tackle these issues. This approach leverages Mean-Shift clustering to categorize passengers with similar pickup and drop-off locations into clusters. Subsequently, employing the reinforcement learning A2C algorithm, we optimize order matching between online car-hailing and passenger clusters to maximize driver utilization while minimizing detour costs. This enhances the quality of ride services for passengers. Experimental findings demonstrate that compared to alternative algorithms, our proposed order-matching method, which integrates passenger demand clustering and reinforcement learning algorithm, diminishes driver detour costs, augments driver revenue by approximately 20.4% and elevates order response rates by approximately 12.5%.