NAT: Neural network Aircraft Trajectory prediction


La page de présentation de NAT est aussi disponible en français.



This project deals with the problem of predicting an aircraft trajectory in the vertical plane. A method depending on a small number of starting parameters is introduced and then used on a wide range of cases. The chosen method is based on neural networks. Neural networks are trained using a set of real trajectories and then used to forecast new ones. Two prediction methods have been developed: the first is able to take real points into account as the aircraft flies to improve precision. The second one predicts trajectories even when the aircraft is not flying. After depicting those prediction methods, the results are compared with other forecasting functions. Neural networks give better results because they only rely on precisely known parameters.

The figure below presents the comparison of the forecasted trajectory with a real trajectory, on an example that has not been learned by the network.


Not available.

Useful links

Main publications

Using Neural Networks to predict aircraft trajectories Yann LeFablec, Jean-Marc Alliot IC-AI 99 Las Vegas (1999/7/1)

Prévision de trajectoires d'avions par réseaux de neurones Yann Le Fablec PhD (INPT) (1999/10/18)

Prévision stochastique de trajectoires : procédures paramétriques et non-paramétriques Christophe Bontemps Rapport de DEA IFP (1997/05/25)

Toutes les publications du LOG sont disponibles ici.

Last Update: