Research On The ATC Behavior Modeling Based On Agent And Machine Learning

In recent years,the development of civil aviation and the continuing growth of air traffic flow test the ability of air traffic management operation.As the important part of air traffic management operation,controllers need to focus on the air traffic in a work situation,and in a short period of time to make scientific decision to solve flight conflicts.Controllers in busy area face enormous pressure,it becomes one of the important bottleneck that restricts air traffic management ability.Building air traffic simulation platform and simulating controllers operating behavior research based on the Agent and the theory of machine learning can provide advanced exploration to auxiliary decision-making system of intelligent controllers for the future research and development,it has important theoretical significance.Machine learning techniques can simulate the operation of the human behavior.Considering the controllers operating behavior of state recognition and optimization of decision making,the basic characteristics of the deep learning and reinforcement learning are keys of behavior modeling.In air traffic simulation platform,we design the controller Agent framework and four controllers basic operation behaviors.Based on the FIPA-ACL communication protocol,we realize the communication behavior of controllers,and study the behavior of other key in the system.Firstly we establish the track prediction model based on BADA and implement conflict detecting module,which can forecast aircraft future fly data.At the same time we define three categories of conflict situation and according to the categories to complete extraction and classification of conflict,implement the controller detection behavior simulation.We preprocess the historical radar track data and generate the learning sample set.secondly we establish the control conflict release module which is based on deep belief networks and introduce the module into the Agent conflicts release behavior module,implement the behavior simulation of find release plan to solve conflict from the historical experience.Thirdly we set based knowledge library and the controller Agent learning model based on reinforcement learning which implement the controller Agent learning behavior module,through the analysis of the effect evaluation,we develop the controllers Agent's conflict release behavior.The conclusion through the simulation shows that the controller Agent behavior module can simulate the different behaviors of controllers,and the communication between different behaviors which proves the effectiveness of the ATC behavior model based on Agent and machine learning.