Computational Methods for Probabilistic Inference of Sector Congestion in Air Traffic Management

Gaétan Marceau (1, 2, 3) , Pierre Savéant (4) , Marc Schoenauer (1, 2)

1 TAO - Machine Learning and Optimisation
2 LRI - Laboratoire de Recherche en Informatique
3 Thales Air Systems
4 Thales Research and Technology [Palaiseau]

This article addresses the issue of computing the expected cost functions from a probabilistic model of the air traffic flow and capacity management. The Clenshaw-Curtis quadrature is compared to Monte-Carlo algorithms defined specifically for this problem. By tailoring the algorithms to this model, we reduce the computational burden in order to simulate real instances. The study shows that the Monte-Carlo algorithm is more sensible to the amount of uncertainty in the system, but has the advantage to return a result with the associated accuracy on demand. The performances for both approaches are comparable for the computation of the expected cost of delay and the expected cost of congestion. Finally, this study shows some evidences that the simulation of the proposed probabilistic model is tractable for realistic instances.

Domaines: Modélisation et simulation; Intelligence artificielle [cs.AI]

Gaétan Marceau, Pierre Savéant, Marc Schoenauer. Computational Methods for Probabilistic Inference of Sector Congestion in Air Traffic Management. Interdisciplinary Science for Innovative Air Traffic Management, Jul 2013, Toulouse, France. ⟨hal-00862243⟩