A Fair Order Matching and Idle Vehicle Dispatching Algorithm for Ride-Hailing

Abstract

Online ride hailing is increasingly prevalent, and effective order-driver matching is critical for its operations. Existing studies mostly focus on maximizing platform profits or minimizing passenger waiting time, often neglecting interdriver fairness. In fact, ignoring fairness may cause drivers to leave the platform, and thus enhancing fairness is vital to attract more drivers to participate. Furthermore, existing works prioritize short-term fairness while disregarding how current matching and fairness-enhancing operations affect future fairness and total income. Thus, we propose an idle vehicle dispatching and fair matching algorithm guided by the value function and dynamic clustering, aiming to optimize long-term fairness and total income simultaneously. Specifically, we design a fairness enhancement method, which improves fairness for drivers with different incomes and maximizes total income for those with similar incomes. To identify similar-income drivers, we use a dynamic clustering algorithm that generates clusters dynamically based on current fairness. However, due to spatiotemporal fluctuations in supply and demand, maximizing current total income does not guarantee long-term maximization. Therefore, we model the matching process within a Markov decision process, where a value function is constructed to estimate each driver’s expected long-term value across different zones, guiding matching and dispatching to maximize drivers’ long-term total income. Experimental results show our algorithm outperforms benchmarks in idle driver rate, order service rate, drivers’ total income, and fairness. The results also verify that it effectively maintains fairness over both short and long horizons while increasing overall income.