A Graph-Spatiotemporal Prediction-Aware Optimization Framework for Emergency Resource Scheduling in Rainstorm-Induced Multimodal Transportation Networks

Emergency resource scheduling for rainstorm-disrupted multimodal transportation networks is challenging due to evolving flood conditions, cross-modal cascading disruption, and mobility-constrained deployment. We present a resilience-oriented rolling scheduling framework that integrates rolling horizon optimization with an improved NSGA-II algorithm to jointly optimize cumulative system resilience and effective drainage utility. To improve online efficiency and decision quality, the framework incorporates an affected active set for dynamic search-space contraction, traffic-weight-guided population initialization, an implicit mobility-loss constraint, and a spatiotemporal local cascading approximation strategy for accelerated fitness evaluation. Experiments under multiple rainfall return periods and temporal demand scenarios show clear advantages over representative heuristic baselines. In the 100-year/16:00 scenario, the proposed method achieves a recovery rate of 0.0206, compared with 0.0077 for Greedy-Flow, while maintaining a comparable maximum performance drop of 0.4451 versus 0.4097. Similar gains are observed in the 50-year/16:00 and 20-year/16:00 scenarios, where the recovery rate increases from 0.0083 to 0.0201 and from 0.0114 to 0.0234, respectively. In addition, the average resource utilization efficiency rises rapidly and remains above 0.85 in high-intensity scenarios. These results demonstrate that the proposed rolling multi-objective scheduling framework can effectively suppress cascading disruption and significantly improve resilience recovery in multimodal transportation systems under extreme rainstorm events.