CONFLICT RESOLUTION WITH TIME CONSTRAINTS IN THE TERMINAL MANEUVERING AREA USING A DISTRIBUTED Q-LEARNING ALGORITHM
Résumé
With the growing number of flights, more and more conflicts have to be solved in Terminal Manoeuvring Areas (TMAs). In order to keep a fluid flow of aircraft arriving on an airport, air traffic controllers use softwares to help them to solve conflicts and sequence aircraft on runways. This paper faces the sequencing and merging problem using a reinforcement learning algorithm (Q-Learning) in order to measure its performance. This algorithm has been run on a scenario representing a regular day at Paris Charles de Gaulle airport (CDG), and gives satisfying results. Then, it has been benchmarked on heavilyloaded scenarios, with more aircraft than the previous ones in order to see the limits of reinforcement learning efficiency. The Q-Learning algorithm can not only solve conflicts on this heavilyloaded scenario but it also has a reasonable computational time. By using a Q-learning algorithm in a distributed way, we aim to find an optimized solution on heavily-loaded scenarios without compromising the computational time.
Domaines
Intelligence artificielle [cs.AI]Origine | Fichiers produits par l'(les) auteur(s) |
---|