Li, Mingyue (Tongji University), Li, Xinghua (Tongji University), Guo, Yuntao (Tongji University)

A Hybrid Attention-Enhanced Spatio-Temporal Graph Network for Continuous-Time Multi-Task Urban Human Mobility Forecasting

Scheduled for presentation during the Regular Session "S34b-Data-Driven Optimization and Governance in Intelligent Urban Mobility" (FR-EA-T34), Friday, November 21, 2025, 14:30−14:50, Surfers Paradise 1

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Understanding urban human mobility is essential for intelligent transportation systems (ITS) and applications such as traffic management and public transit planning. Although high-resolution spatiotemporal data from GPS devices, smart cards, and mobile services have enhanced mobility modeling, effectively capturing the continuous, multi-scale, and context-dependent nature of human movement remains challenging. Existing Spatio-Temporal Graph Neural Networks often rely on discrete-time assumptions, treat key tasks independently, and neglect external factors like weather and public events. To address these limitations, we propose HA-SCTNet (Hybrid Attention-Enhanced Spatio-Temporal Graph Network), a unified encoder-decoder framework for multi-task urban mobility forecasting. HA-SCTNet integrates continuous-time representation learning, hierarchical spatial reasoning, and residual-based task-aware optimization to jointly predict origin-destination demand, inflow, and outflow. It incorporates Spatio-Temporal Cross Attention Blocks to model multi-scale spatial-temporal dependencies and employs a memory-augmented message-passing mechanism for long-term information retention and asynchronous temporal state updates. Additionally, external contextual factors such as weather, holidays, and special events are embedded into the model to improve robustness. Experimental results demonstrate that HA-SCTNet outperforms state-of-the-art baselines in capturing the complex dynamics of urban mobility. This work contributes to the development of more accurate and context-aware forecasting models for ITS.