A parallel computing framework for air traffic flow management

Abstract

The nationwide air traffic flow control for the National Airspace System is a complicated large-scale optimization problem which is significant to the future development of Air Traffic Management. Based on a Link Transmission Model and dual decomposition method, the large-scale air traffic flow optimization is decomposed into smaller independent optimization subproblems and solved using parallel computing. As the model is solved with a Mixed Integer Linear Programming, searching for an optimal integral solution usually entails longer runtime than Linear Programming. To improve the applicability of this model, a parallel computing framework is developed, which explores the parallelism of the model in order to increase the computational efficiency. Heterogeneous computers are clustered in a Client/Server topology to carry out parallel computing. By further exploring the multithreading capability of multi-core computer, the optimization task is distributed to multiple processors to process in a parallel fashion. Simulation shows that the parallel computing framework decreases the runtime of nationwide air traffic optimization from hours timescale to minutes timescale. Moreover, compared to conventional single thread optimization method, the framework can achieve expected runtime reduction by deploying more computer resources without worrying about the increased complexity of the traffic network, therefore making the near real-time air traffic flow optimization possible.