Research On The Recognition Method Of Multi-scale Behavior Patterns In Air Traffic Management

With the rapid development of air transportation industry,both operation burden of ATC system and staff workload are getting heavier,the transportation operation pattern is getting more complicated.Promoting the operation performance of ATC system to meet the demand of development is now becoming the hot spot in the industry.While organization such as ICAO has clearly proposed goals and practices to improve performance,studies on the diversified behavioral properties of the system have never been push forward,and no mutual understanding is achieved in this subject.Therefore,in order to clearly distinguish the ATC system and objectively find out the bottleneck for the operation,it is necessary to analyze in depth the multiple behavior that reflects the essential status of air traffic system first and scientifically mine the distribution pattern and evaluation rules of all categories of behavior so as to accurately make the plan and comprehensively promote the operation performance.After fully studies on the advanced research achievements abroad,this paper is focused on the multi-scale behaviors that appear in the air traffic system.This paper establishes recognition methodologies for different behaviors from aspects of airport traffic behaviors,sector traffic behaviors,ATCOs'(Air traffic controller)behaviors and interrelationship between traffic behaviors and ATCOs' behaviors.The main findings are as followed.(1)The multiple characteristics of airport traffic behaviors were studied from two aspects,including intrins ic properties and correlation features.Firstly,characteristics of traffic behaviors for 161 airports were explored based on chaotic analysis method.Study results indicate that 6 kinds of traffic behaviors including flow,arrival flow,departure flow,the number of delayed flights,the number of delayed arrival flights and the number of delayed departure flights of 48 major busy airports are chaotic.For all airports,chaos occurs in the flow behaviors most frequently.Then the correlation features of airport traffic behaviors(flow and the number of delayed flights)were investigated.The airports with high correlation in traffic behaviors were classified into the same category through spectral clustering algorithm.The analyses based on self-organized criticality theory were performed to examine and compare the distribution patterns of traffic behaviors' correlation for each category of airports from clustering results in spatial and temporal dimensions.It can be concluded that almost all categories of airports are with self-organized criticality.The self-organized criticality is also found in the correlation network composed of all airports' traffic behaviors.(2)The intrinsic properties and distribution patterns of sector traffic behaviors were identified through multi-dimensional metrics system.A metric system was constructed to quantitatively depict traffic behaviors.The metric system included density,dynamic and conflict metrics,which can describe the traffic situations from different points of view.The sector traffic behaviors were examined through chaotic analysis method.Study results indicate that 6 kinds of traffic behaviors including the number of flights,the average flight distance,the average flight time,average velocity,minimal horizontal separation and minimal vertical separation of all sample sectors are chaotic.The conclusion shows that the two types of traffic behaviors are both chaotic.Two kinds of clustering methods were constructed to analyze sector traffic behaviors.Firstly,primary component analysis was used to refine information from the same category of metrics.15 sector samples were clustered based on the primary components of density,dynamic and conflict metrics separately.Secondly,the time series of different metrics were calculated.A k-medoids clustering algorithm based on DTW was developed to cluster high-dimensional time series data effectively.The diverse distribution patterns of 15 sectors were identified under each traffic behavior.The case study shows that the clustering methodology based on main components is more applicable to identify the distribution pattern of multi-sector under one particular category of indicator implication,while clustering methodology based on time sequence is able to identify the distribution pattern of multi-sector under one particular traffic feature.Combined,they can provide effective methodology and basis for the analysis of diversified traffic behavior.(3)The patterns of controllers' behaviors were derived on the basis of communication activity and eye movement activity.In the aspect of communication activity,detrended fluctuation analysis was conducted to exam the long-range correlations in communication activities.The distribution models of inter-communication times were estimated by the means of maximum likelihood estimation.Investigations on historical data found that the inter-communication times of controller are long-rang correlated both at the overall levels and individual levels.The distribution of inter-communication times follows power laws.The type of sector has little impact on the correlations or distribution type.In the aspect of eye movement activity,two commonly investigated oculomotor behaviors,fixation and saccades,were studied.Different levels of controllers were invited to perform simulation exercises.The rules of the number of AOI,fixation duration and saccadic velocity were analyzed with statistical method.The distribution models of eye movement indicators were estimated.We found that controllers were divided into three categories based on fixation behaviors,namely Level-two,from Level-five to Level-three and novices.Controllers were divided into two categories based on saccade behaviors,namely licensed controllers and novices.The distributions of fixation duration and saccadic velocity show power-law features.Experienced controllers can use more efficient information searching strategies and allocate attention more reasonably.(4)The correlations between air traffic behaviors and controllers' behaviors were investigated by using correlation analysis method.Based on the study of traffic behaviors and controller 's behaviors,metrics were selected to construct eye movement-traffic pair of metrics and communication-traffic pair of metrics.Pearson correlation coefficient and Spearman's correlation coefficient were both calculated between every pair of metrics.Then the distributions of linear correlation and monotonic correlation were analyzed.Transfer Entropy offers an approach to detect the pattern of information transfer between eye movement behaviors and traffic behaviors.The results indicate that the eye movement behaviors and traffic behaviors are mostly monotonic correlated(including linear correlated).The eye movement metrics are mostly correlated to density and dynamic metrics.T he communication behaviors and traffic behaviors are mostly linear correlated.The communication metrics are mostly correlated to density metrics.The transfer entropy finds information transport in both directions but indicates more pairs of metrics and a stronger flow of information from eye movement behaviors to traffic behaviors than vice versa.Individual ATCO or control level has no significant effect on the distribution pattern of correlation coefficient or transfer entropy of metric pairs.Finally,the main research results are summarized,and the future research directions are prospected.