Embedded Statistical Analysis and Selection Procedures in Air Traffic Simulations

Abstract

Statistical analysis and adaptive sampling techniques can be embedded within existing air traffic simulations to provide several benefits. First, embedded statistical calculations can eliminate the need for extensive storage and post hoc processing of simulated outputs. Second, adaptive sampling can identify the number of observations required for rigorous statistical comparison, often dramatically reducing the number and duration of simulation runs from those predicted a priori by standard statistical techniques, which tend to be quite conservative. Third, these methods facilitate efficient distribution of simulator runs over a network of workstations without requiring parallelization of the simulation software.This paper describes three central components required for these benefits: (1) embedded statistical calculations; (2) adaptive statistical selection techniques; and (3) the server-client structure for a network of workstations in which individual workstations gather and report interim statistics and then are commanded to perform those further simulation runs required for a given statistical comparison. Their application to air traffic simulation is demonstrated using the Georgia Tech Reconfigurable Flight Simulator in a simulation of a standard terminal arrival route into Atlanta Hartsfield-Jackson airport in which the appropriate simulation runs were automatically carried out until the ‘best’ (or ‘worst’) systems were identified.