Optimal Traffic Flow Scheduling Using High-Performance Computing

Algorithms for the end-to-end optimized scheduling of aircraft to enhance the efficiency of the National Airspace System are developed. For a given set of flights and desired departing schedules, routes are constructed and unimpeded four-dimensional trajectories are simulated. These trajectories serve as an input to a linear-programming-based approach, and they result in optimized schedules that are deconflicted while assuring adherence to the system capacity constraints. For a large number of flights, the computational effort is formidable and optimization coupled with the Dantzig–Wolfe decomposition technique has been found to be a suitable approach. Techniques for accelerating the decomposition and solver on emerging high-performance computing hardware are discussed. A multithreaded central-processing-unit implementation and a novel implementation on general-purpose graphics processing units show acceleration over a state-of-the-art open-source decomposition-based solver. The acceleration observed can be up to nine times for replicated flights and three times for realistic nationwide traffic flow optimization examples.