Machine learning and applications to air traffic management

Artificial learning is a rapidly expanding field of science. With the emergence of massive data (Big Data), it has become an indispensable tool for extracting useful information or predictive models from these data, in many fields of application.

The field of learning can sometimes appear to the neophyte as a veritable jungle, obscure and inaccessible. This article in no way claims to explore it exhaustively, or even superficially. The aim is to provide the reader with an entry point and a few tools (a machete, to continue our analogy), so that he or she can then explore the field on their own, through specialized reading. Reference works include books by T. Hastie et al. , C. Bishop , and in French, the work by A. Cornuejols and L. Miclet .

This introduction to artificial learning is illustrated by two applications to air traffic management problems.

In the first, the workload prediction model for air traffic controllers ("aiguilleurs du ciel") presented is a neural network, learned from aircraft trajectory recordings and control sector opening archives. Once learned, this model is combined with a tree search method to predict optimal air traffic control sector opening configurations.

In the second application, machine learning can be used either to directly predict the altitude of a climbing aircraft, or to estimate certain parameters of the aircraft's physical model, not available in ground-based predictors, in order to improve this altitude prediction. The elements presented in this article are further detailed in