A Cluster Analysis to Classify Days in the National Airspace System

Abstract

Simulations spanning the entire National Airspace System (NAS) are of growing interest. For NAS simulations, researchers must select an appropriate set of days to validate their simulation models. However, within vast quantities of historical NAS data, certain days have had abnormal events that create unusual or anomalous traffic flow patterns – September 11, 2001 being the most notable. For such atypical days, researchers may wish to avoid them or reserve them for special analysis or modeling. Furthermore, the researcher is confronted with the problem of how to consolidate vast quantities of descriptive NAS data into a more easily interpretable description of daily operational behavior that can help in selecting appropriate days for simulation validation. In this paper, we design a NAS feature vector to characterize the NAS behavior for comparing across days. Cluster analysis is used to condense an initial collection of 65 aggregate, daily NAS variables into a more manageable set of variables. Each feature vector then represents NAS performance on its respective day. In a second but different cluster analysis, we rank NAS feature vectors (and therefore days) by levels of normality. Finally, we provide recommendations on how to use these “typical” days to validate NAS simulations.