Flight Conflict And Deployment In Terminal Area Based On Deep Reinforcement Learning

The terminal area is one of the central parts of air traffic control.Its task is to manage aircraft to take off,touch down,and arranged enter and depart.As the amount of aircraft in the airspace raises,the flight volume that lacks to be processed in the terminal area will grow accordingly.There’s no doubt that it would lead to more extended flight conflicts.In detail,it will generate air traffic congestion and flight delays,and thereby affect flight security.Accordingly,it is of great significance to investigate flight conflicts and deployment in the terminal area.Take a more extensive dive into the flight conflict and deployment in the terminal area would make great significance.Terminal flight Conflict and Deployment refer to the prediction of conflict between aircraft,and the process of taking measures to evade potential collisions based on the current aircraft parameters.Through the parameters,whether the bilateral distance between aircraft in the future would cause a crash can be apprehended.Rely on the existing air control technology,this paper divides the terminal area structure,composes a flight approach route and set up the moving time slot,and stipulates aircraft to land according to the approach plan when entering the terminal control area.In terms of the flight conflict and deployment obstacle while the approach process,the paper explained it as an interactive and collaborative model between multi-agents according to the theory of multi-agent reinforcement learning,which is the Markov Model.In this model,respectively aircraft is identified as a separate agent.The agent would adjust its dispatch to switches pace to avoid collisions between aircraft by discerning the conditions and movements of other agents in this airspace and join the time slot.The model also presents a reward function that operates stand on the corresponding distance and speed between aircraft and form the whole multi-agent system reaches the maximum bonus.In the last part,the paper integrates the deep learning Pytorch framework and the reinforcement learning Gym environment to build an experimental environment in the air traffic management simulation software Bluesky.The paper employs the multi-agent MADDPG algorithm to resolve the flight conflict difficulty in the airspace of the terminal area and compares it with other algorithms.