Optimization of Rescue Team Scheduling under High Altitude Earthquakes Based on Reinforcement Learning

In the early stages of high-altitude earthquake disasters, efficient rescue team scheduling is critical to minimize casualties and optimize resource utilization. This study proposes HRLPPO, a hierarchical reinforcement learning framework combining Proximal Policy Optimization (PPO) with task stratification, to address the dynamic allocation of rescue teams under complex constraints. The framework divides rescue tasks into a high-level strategy for selecting teams and disaster sites, and a low-level strategy for determining personnel dispatch quantities. Key innovations include integrating high-altitude compatibility constraints, minimizing dispatch costs via distance-aware reward functions, and enabling rapid decisions through pre-trained policies.A custom reinforcement learning environment was designed to simulate real-world scenarios, incorporating rescue team capabilities, site demands, and geographical constraints. Experiments using data from the 2022 Luding earthquake in Sichuan demonstrated HRLPPO’s superiority over traditional methods (e.g., Genetic Algorithm, Ant Colony Optimization). Results showed 18.5% lower dispatch costs, 95% faster decision times (0.43s vs. 487.43s for 10-team scenarios), and 99.79% rescue satisfaction rates under both sufficient and insufficient high-altitude team conditions. The model’s robustness was further validated in large-scale scenarios (40 teams, 20 sites), achieving 81.68% overall satisfaction despite resource shortages.This work provides a novel decision-making tool for emergency management, enhancing rescue efficiency in high-altitude regions. Future efforts will integrate GIS platforms for real-time disaster response.