Enhancing graphical simulation output using cluster analysis

Abstract

Recent work performed for the Federal Aviation Administration to support the development of future concepts of air traffic management has involved simulation modeling of patterns of airspace usage by commercial and business air traffic. The objective of these efforts has been to investigate the impacts of a pattern of airspace usage known as "free flight", whereby pilots and flight dispatchers have much more freedom to choose, say, direct or wind-optimal routing through airspace. One of the figures of merit investigated is a count of "convergence pairs" as a measure of the complexity of various traffic patterns. These are cases when aircraft in the simulation model fly close to each other. Interestingly, geographic plots of convergence pairs accumulated over time bring out certain features or patterns of congested air traffic flows or flight alignments. However, these plots are also thick with "noise" or extraneous convergence pairs, whose presence detracts from the ability to perceive congested air traffic flows. Cluster analysis has been found to be an effective method of filtering these displays so that the congested flow features are discernible. The process developed for this purpose is based on a two-pass clustering approach. The process has worked well for the simulation modeling performed to date. Classification of the locations of convergence pairs into congested flow corridors is visually appealing, and has helped distinguish differences in contrasting scenarios of airspace usage. The paper presents graphical results and describes the clustering algorithms employed.