Parallelization of the Traffic Flow Management Problem

Abstract

This paper describes a parallel approach to solving the national air traffic flow management problem. High fidelity, aircraft-level approaches to this problem can determine which flights should be held on the ground or in an enroute sector. The major drawback to such aircraft-by-aircraft approaches lies in the large amount of data and resulting massive problem instances to be solved for nationwide or long planning horizon scenarios. The presented approach solves twenty individual problems, one for each Air Route Traffic Control Center in the continental United States. Individual client processes coordinate with a central server to solve their respective problems over multiple iterations. Experiments are performed using recent, historic data within a nominal scenario and a weather scenario. These experiments demonstrate the potential for solving this problem using parallel approaches in greatly reduced time. The initial results show solutions that would have taken over 24 hours to obtain optimally in a monolithic system were shown obtainable to within 3% optimality in less than one hour using the parallel architecture. In addition to the runtime and delay cost analysis, delay results from successfully running a high-fidelity, nationwide traffic flow scenario are also detailed.