Aviation Analysis Using a Distributed Agent-Based Infrastructure

Distributed agent-based infrastructures are a natural computational representation for complex "system-of-systems" such as exhibited by the National Airspace System. In this pa- per we describe the basic architecture behind agent-based systems and demonstrate why they represent suitable computational abstractions for complex systems using an example derived rom National Airspace System analysis. i transporting over seven hundred million passengers per year an average of eleven hundred miles each. ii The NAS accounts for about sixty percent of worldwide air travel, and does so with an unparalleled degree of safety. The infrastructure supporting this system includes over nineteen thousand airports, about five thousand of which are open to the general public, iii as well as over twenty thousand air traffic controllers staffing airport towers, airport radar control centers, enroute traffic control centers, and traffic flow management facilities. A system as large, geographically dispersed, and diverse as the NAS can best be characterized as a system of systems. Some of the systems consist of procedures followed by pilots, rules governing flight operations, require- ments for air carriers, operating procedures for controllers, and even laws governing the proper behavior of passen- gers and the transport of cargo. Other systems are hardware-based: a network of radars and multilateration equip- ment supports surveillance, strategically placed antennas enable communication, and a network of ground-based antennas and space-based GPS satellites provide navigation. Despite its large size, enviable safety record, and large volume of operations, upgrades to the NAS are required to handle ever-increasing traffic volumes along with in- creased safety levels and decreased operating cost. The only feasible way to evaluate the effectiveness of proposed upgrades is through extensive analytical and human-in-the-loop computer simulation. Simulating such a system-of-systems is itself a computational challenge. There are a diversity of different com- putations, including continuous-time physics-based trajectory computations, discrete-event protocol-stack commu- nication models, queueing models for runways and airports, strategic planning models that must "look-ahead" in simulation time and select a feasible tactical alternative, scheduling software that must compute required times-of- arrivals of flights at waypoints and airports, and much more. The diversity of the NAS combined with its computa- tional complexity makes it an ideal candidate for distributed agent-based modeling. The remainder of this paper describes a distributed agent-based modeling system called CybelePro®, iv and details its use as the main platform on which a distributed model of the NAS, called the Airspace Concept Evaluation System (ACES), is built.