Priority-aware independent path-driven multi-agent reinforcement learning for optimal post-earthquake road network recovery

Ensuring the functionality of post-earthquake road networks is critical to effective emergency response and recovery. However, devising efficient restoration strategies is challenging because real-world networks are complex and emergency facilities have heterogeneous traffic demands. This study proposes a strategy optimisation framework based on multi-agent reinforcement learning that prioritises and sequences repairs to enhance the emergency response capability of damaged road networks. The framework integrates functional indicators such as traffic demand and accessibility indicator to quantify emergency service needs and the network’s ability to connect key facilities with affected areas, thereby providing a quantitative basis for recovery decisions. To improve scalability on complex networks, the Priority-Aware Independent Path (PAIP) method is introduced to accelerate the computation of independent paths. The simulation experiments conducted in Guangzhou Higher Education Mega Centre and Tangshan City demonstrated that the proposed model outperformed other methods in terms of enhancing recovery efficiency and emergency response capabilities.