Reinforcement Learning Based Airport Runway Risk Decision

The continuous growth of civil airport throughput brings enormous pressure to airport safety management. The existing passive safety management method cannot efficiently control risk in advance and lacks consideration of airport operation efficiency, leading to degradation on airport operation efficiency. To overcome these issues, we propose a reinforcement learning based runway risk decision method which dynamically adjusts the take-off and landing runway for the upcoming flight, based on current pavement crack level and airport operation status. Besides, a risk decision evaluation metrics that can comprehensively reflect the airport operation efficiency and safety is designed to evaluate the pros and cons of risk decision-making. Furthermore, to evaluate the robustness of risk decision, the impact of different parameter settings and network structures on the method performance are compared. Experimental results show that our proposed method improves the comprehensive evaluation of safety and efficiency by 1.50%–37.41%.