Formulating infrastructure restoration as a geospatial multi-project scheduling problem using agent-based simulation

Urban flooding increasingly disrupts transportation networks, requiring efficient coordination of repair operations. This paper addresses how to optimize post-flood road restoration scheduling as a scattered repetitive project in a sophisticated, real-time decision-making environment. The proposed framework integrates agent-based modeling with deep learning, where autonomous repair crews dynamically prioritize tasks based on real-time accessibility predictions from a neural network proxy model. The Beijing case study demonstrated that the accessibility-driven strategy significantly improved recovery of network functionality compared to nearest-first and random approaches, particularly during critical early restoration phases. This improvement matters for emergency managers and infrastructure operators who must rapidly restore community access to vital facilities such as hospitals after disasters. Future research can extend this framework to other hazards and infrastructure systems, incorporating advanced uncertainty quantification, climate-informed risk assessments, and adaptive decision-making mechanisms for enhanced disaster planning. • Post-flood restoration is formulated as geospatial multi-project scheduling prioritizing accessibility. • Agent-based modeling with deep learning enables rapid, scalable restoration of large-scale urban networks. • Interpretable Gantt charts and crew trajectories provide practical decision support tools for emergency managers. • Comparative analysis of restoration strategies offer actionable insights for emergency response.