Ride-hailing origin-destination demand prediction with spatiotemporal information fusion

Abstract

Accurate demand forecasting for online ride-hailing contributes to balancing traffic supply and demand, and improving the service level of ride-hailing platforms. In contrast to previous studies, which have primarily focused on the inflow or outflow demands of each zone, this study proposes a Conditional Generative Adversarial Network with a Wasserstein divergence objective (CWGAN-div) to predict ride-hailing origin-destination (OD) demand matrices. Residual blocks and refined loss functions help to enhance the stability of model training. Interpretable conditional information is employed to capture external spatiotemporal dependencies and guide the model towards generating more precise results. Empirical analysis using ride-hailing data from Manhattan, New York City, demonstrates that our proposed CWGAN-div model can effectively predict the network-wide OD matrix and exhibits strong convergence performance. Comparative experiments also show that the CWGAN-div outperforms other benchmarking methods. Consequently, the proposed model displays potential for network-wide ride-hailing OD demand prediction.