Learning Higher-Order Migration Patterns: A Hypergraph Approach to Urban Mobility Prediction

Accurate intercity population mobility prediction is crucial for urban planning and transportation management, yet existing graph-based approaches struggle to capture complex higher-order spatial dependencies. This paper presents a novel Spatiotemporal-Aware Hypergraph Neural Network (STAHG) framework that integrates three key innovations: (1) A Wasserstein-distance-based hyperedge construction method for multi-node correlation modeling, (2) An adaptive hypergraph convolution operator preserving migration heterogeneity, and (3) A Hybrid Hypergraph GRU (HGGRU) architecture unifying spatial and temporal modeling. Through extensive experiments on real-world Amap migration data, our framework demonstrates superior performance, achieving 12.77% reduction in MAE for next-day predictions and 11.62 % for emigration indices compared to state-of-the-art baselines. These results validate STAHG's effectiveness for large-scale urban mobility prediction tasks across multiple timescales.