Ride-hailing dispatch systems face significant challenges under fluctuating demand and dynamic traffic conditions, where efficient coordination is essential for both platform performance and driver income among large-scale ride-hailing vehicles. This paper constructs a grid-based ride-hailing vehicle dispatch decision model (GRV-DDM), which provides a structured and quantifiable representation of vehicles and orders, effectively capturing spatio-temporal heterogeneity in dynamic traffic environments. Based on this model, a Bi-Level Optimization Multi-Directional Dispatch Decision Algorithm (BO-MDDA) is proposed. At the macro level, evolutionary game theory is employed to adaptively guide collective vehicle strategies toward supply–demand equilibrium, while at the micro level, deep reinforcement learning optimizes individual drivers’ real-time dispatch decisions to maximize long-term profits. A bidirectional feedback mechanism is further designed to integrate macro-level collective intelligence with micro-level individual decision-making. Experimental results across diverse traffic scenarios demonstrate that the proposed approach outperforms classical dispatch algorithms in terms of efficiency and robustness.
Yan et al. (Sun,) studied this question.