Mining Heterogeneous ADS-B Data Sets for Probabilistic Models of Pilot Behavior

The University of North Dakota is developing airspace within the state where Unmanned Aircraft Systems (UASs) can be flown without an onboard sense and avoid system or Temporary Flight Restrictions (TFRs). With funding from the U.S. Air Force, a mobile ground-based radar system capable of detecting aircraft operating in Class E airspace and the software to display such information to UAS operators is being developed. The current system uses an Automatic Dependent Surveillance - Broadcast (ADS-B) transceiver to detect any ADS-B-equipped aircraft within the vicinity, and a Ground Control Station (GCS) to detect and control the UAS. Once one or more ground-based radars are integrated into the system, it will also be capable of detecting non-cooperative aircraft (i.e. aircraft that aren´t equipped with ADS-B transceivers) operating within the vicinity. The current system uses a portable, high-availability architecture. Since the system is intended to detect potential airspace conflicts from the ground, greater computational power is available to it than to onboard sense and avoid systems. The probability of a midair collision is dependent on the proximity of aircraft to each other, the performance characteristics of the aircraft, and the probabilities of pilots performing basic maneuvers with the aircraft. In this paper the authors present the results of data mining an ADS-B data set from 11 days in early 2010. Probabilistic models of pilot behavior were automatically extracted from the data using a genetic algorithm for cluster analysis.