Virtual graph-based holiday-aware prediction of intercity origin-destination mobility flows

Accurate prediction of intercity origin-destination (OD) flows is critical for resilient transportation systems during holiday periods. Existing models primarily focus on spatial adjacency and overlook city association changes from temporal OD flow dynamics. Moreover, predictive models for intercity OD flows that consider holiday effects are lacking. Therefore, we propose the Spatio-Temporal Correlation Virtual Graph Network (STCV-Net), which constructs two types of virtual graphs based on the urban socio-economic attributes and dynamic similarity of the inflow-outflow patterns. Additionally, STCV-Net incorporates a spatiotemporal feature update method based on similarity graph convolution while integrating refined holiday features with global spatiotemporal features, enabling the model to simultaneously capture long-term spatiotemporal dependencies and short-term holiday-induced mobility variations. Experimental results show that STCV-Net reduces MAE, RMSE, and MAPE by 15.71%, 14.87%, and 19.45% compared to the second-best model. Furthermore, STCV-Net provides valuable insights into AI-based mobility modelling and the dynamic spatiotemporal interactions within smart transportation systems.