Agent Evaluation Function Considering Airspace Dynamic Density

This paper has the objective of presenting inteligent-agent methodology to evaluate the Air Traffic Flow Management (ATFM) scenario which aggregates the Dynamic Density criterion. The new evaluation function is proposed to determine the rewards in a reinforcement learning ATFM agent. The paper includes a specific analysis focused on the air scenario observed in Brazilian airspace. The comparative results of the application of the scenario evaluation function for each criteria: aircraft counting and Dynamic Density (or Airspace Complexity) are promising. It is possible to verify that the scenarios with the same number of aircraft presented a state value with significant variations between these two functions (with or without Dynamic Density). These variations have a remarkable influence on the reinforcement learning performance and, consequently, on decision making supporting system for the traffic flow manager.