Improved Clustering for Route-Based Eulerian Air Traffic Modeling

Abstract

A new approach is developed for identifying and approximating well-traveled routes in a historical dataset of flight trajectories. The approximate routes are intended for use in a route-based Eulerian model of air traffic flow for strategic planning but are useful for other route-based strategic planning. The approach involves coarse clustering, outlier detection, fine clustering, and aggregate route construction. Coarse clustering is based on common origin, destination, and average cruise speed. Fine clustering, based on the Fréchet distance between pairs of trajectories, is applied to each coarse cluster to subdivide it, if appropriate. The coarse-clustering step reduces the number of trajectory pairs for which the Fréchet distance must be computed. The number of fine clusters is automatically determined using a combination of three performance indices. Outliers are identified using previously developed methods. The outliers could be discarded or assessed to identify potential routes for avoiding areas that are flight-constrained. The effectiveness of the approach for determining aggregate well-traveled routes is demonstrated on a historical dataset for a domain composed of six centers with a total of 19 airports.