MVSTD: Multi-view spatio-temporal graphs with external disturbance consideration for ride-hailing demand prediction

Abstract

Accurate forecasting of ride-hailing demand is crucial for intelligent transportation systems, optimizing fleet management, reducing idle vehicle times, and alleviating urban traffic congestion. This task is particularly challenging due to intricate spatio-temporal dependencies, dynamic demand fluctuations, and the influence of external factors such as weather and public holidays. Existing methods often fail to adequately capture these non-linear dynamics or integrate the full spectrum of external disturbances. To address these challenges, we propose the Multi-View Spatio-Temporal Graphs with External Disturbance Consideration (MVSTD) framework. MVSTD innovatively combines multi-view spatio-temporal graph modeling with explicit incorporation of external disturbances, enabling more accurate and robust demand forecasting. The framework captures critical interactions between historical demand patterns and external variables, such as weather and holidays, to identify key drivers of demand variability. Extensive experiments on two large-scale, real-world datasets demonstrate that MVSTD consistently outperforms state-of-the-art methods across multiple evaluation metrics. Notably, MVSTD demonstrates superior performance in high-variability scenarios, including public holidays and adverse weather conditions, showcasing its practical relevance for real-world ride-hailing demand prediction.