Air holding problem solving with reinforcement learning to reduce airspace congestion

Abstract

SUMMARYThe Air Holding Problem Module is proposed as a decision support system to help air traffic controllers in their daily air traffic flow management. This system is developed using an Artificial Intelligence technique known as multiagent systems to organize and optimize the solutions for controllers to handle traffic flow in Brazilian airspace. In this research, the air holding problem is modeled with reinforcement learning, and a solution is proposed and applied in two case studies of the Brazilian airspace. The system can suggest more precise and realistic actions based upon past situations and knowledge of the professionals and forecast the impact of restrictive measures at the local and/or overall level. The first case study shows performance improvements in traffic flows between 8 and 47% at the local level up to 49% at the overall level. In the second case study, performance improvements were between 15 and 57% at the local level and between 41 and 48% at the overall level.

Review on artificial intelligence techniques for improving representative air traffic management capability

“…Agogino et al [50] proposed a multi-agent algorithm that used RL for reducing congestion; the proposed method significantly improved the traffic flow, and provided adaptive and robust solutions to the flow management problem. Mccrea et al [51] used k-means clustering to conduct an economic benefit analysis and applied it to a large-scale airspace environment management. Cruciol et al [52] proposed a decision support system using multi-agent systems, to organize and optimize the solutions for handling traffic flows in the airspace.…”