Empirical analysis of air traffic controller dynamics

This paper addresses an empirical analysis of air traffic controller’s activities using a human dynamics and complex systems approach. Recent studies on human dynamics show several empirical evidences that, different from common belief respecting random-based Poisson distributions, patterns of human activities fit into power law distribution with heavy tail patterns. Our hypothesis lies upon the question whether or not controller’s dynamics obeys the same power law pattern. The analysis based on a 2-weeks simulation dataset is first performed to examine the interaction between traffic activities and controller’s communication activities. Two widely studied complexity metrics, the Dynamic Density (DD) and the complexity based on dynamical system modeling (C-DSM) approach, have been constructed from the aircraft trajectory data. It is, however, found that neither the DD nor the C-DSM has significant influence on the controller’s communication temporal behavior, except that few approach sectors show close relationships between the DD and communication. Beside this simulation dataset, three other datasets which include another simulation dataset and two operational datasets are also investigated to study the temporal characteristics of controller activities. The use of detrended fluctuation analysis (DFA) found that the inter-communication times of controller are long-rang correlated, showing a heavy tailed pattern. We show that the Inverse Gaussian distribution is better than the Power-law distribution to describe the temporal data. This indicates that the mechanism underlying controller’s activities is different from the general one proposed by Barabasi (2005). The Lévy process with positive drift may be capable of explaining the adaptive behavior of the controller.

► We show that air traffic complexity has little impact on controller’s communication. ► We use detrended fluctuation analysis to study controller’s communication pattern. ► It found that the controller inter-activities times are long-rang correlated. ► The Inverse Gaussian distribution is identified to better describe the temporal data. ► It indicates that the mechanism underlying controller’s activities is different.