Simulating Urban Futures: A Digital Twin Framework for Proactive Mobility Management Based on Hybrid Spatio-Temporal Graph Neural Network

The integration of transportation and urban planning is a key challenge for contemporary megacities.  A significant problem is the absence of tools capable of quantitatively assessing the future impact of urban development on mobility patterns.  For proactive urban mobility management, a framework of a comprehensive Digital Twin (DT) structure is proposed, based on the methodology of a multilayered, dynamic digital representation of the city's public transport system, relying exclusively on open data sources such as GTFS, OpenStreetMap, and OpenWeather.  To this end, a novel Hybrid Spatio-Temporal Graph Neural Network (HST-GNN) has been developed, trained to forecast passenger demand at the granular level of route segments.  The developed HST-GNN integrates Graph Attention Network (GATv2) to capture complex spatial dependencies and Long Short-Term Memory (LSTM) networks to model temporal dynamics.  A key innovation of the developed DT is its ability to forecast passenger demand at a granular, edge-centric level (journeys between stops), rather than at the traditional node level.  The high accuracy of the model was empirically validated on a test dataset ($\mathrm{MAE}=0.0018$ and $\mathrm{RMSE}=0.0061$), confirming its applicability as a simulation platform.  The proposed data-driven DT serves not merely as a monitoring tool but as a powerful modeling platform, transforming it into an active instrument of strategic planning for proactive urban mobility management.