Research On Arrival Aircraft Control Decision In Terminal Area Based On Machine Learning

As a transitional area between the route and the airport,the terminal area gathers on the approach route and exhibits high complexity and high density.With the rapid increase of air transportation demand,the experience-oriented management mode can not grasp the spatial regularity and dynamic characteristics of traffic flow.In order to achieve targeted and refined decisions,the use of machine learning methods to conduct technical research on terminal area operational data can deeply understand the traffic flow organization model and improve the efficiency of air traffic control.Based on the historical radar trajectory data of the terminal area,this paper conducts in-depth research on the identification of incoming traffic flow,situation analysis and control reference path extraction.Firstly,the trajectory data was preprocessed by resampling method to eliminate the original data redundancy,and the similarity model was established based on the modified Euclidean distance and dynamic factors.Introducing the natural nearest neighbor idea to obtain the density information between the track points.In this way,the adaptive Gaussian kernel function denoising process is realized,and the spectral clustering analysis of the approach flight trajectory is performed to accurately extract the incoming traffic flow in different directions.Secondly,this paper defined the traffic flow properties that could reflect the operational characteristics of the aircraft.Then,it studied the numerical relationship and variation law between different properties to reveal three phase-states of traffic flow: free,steady and congestion.The FCM fuzzy clustering method was introduced to establish a phase-state identification method for a single traffic flow,and the clustering results were used to explore the evolution law of traffic flow phase-states,which could realize objective discrimination and differential measurement of traffic conditions.Research results provide support for terminal area management operation strategy formulation and space-time distribution optimization.Finally,a phase-state prediction model based on SVM was proposed.The input of the model was independent factors which extracted from traffic flow properties by dimensionality reduction.The sample obtained by FCM were used for training model.Then,this paper proposed a method for obtaining optimal parameters by K-fold cross validation which compled the forecast of traffic flow phase-state in all directions.According to the prediction results,the trajectory data of different traffic flows were extracted.Analyzied the performance of the approach flight trajectory data under different situations,a method for obtaining regulatory reference trajectories under different situations was designed,which could provide decision support for controllers from the perspective of data,and provide scientific reference for application research and engineering implementation of decision support methods in terminal areas.The research results provide technical reference for scientific decision and engineering implementation of support methods in terminal areas.