Air traffic flow Management aims to control route network flow on regions in which the service is provided. One important challenge is to guarantee that aircraft are able to departure and land safely, with minimum delay. To satisfy this requirement, several restrictive measures are employed, such as: ground delay programs, en-route deceleration, holding patterns at intermediate altitudes (descent and approach phases), and diversions to less congested airports. The problem becomes how to better distribute these measures among the involved entities. In this paper, a new model is proposed to tackle this problem, named RELEASE, which utilizes reinforcement learning techniques to train agents to better negotiate their delays with other aircraft. To do so, the negotiation process was modeled as a Markov Decision Process. The aircraft is considered as agent and then trained to negotiate its position in the environment with other aircraft through time offers with an associated reward. The learning behavior was compared to a reference model. In the case studies, the agents were able to achieve similar results as the reference model, and even getting better results for agents with greater operational costs than their predecessors.
Negotiation Approach by Reinforcement Learning for Takeoff Sequencing Decision in Airports
2019-10-01
350501 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Large Aircraft Takeoff and Approach Path Optimization out of Challenging Airports
British Library Conference Proceedings | 2011
|Engine-Out Takeoff Path Optimization out of Terrain Challenging Airports
British Library Conference Proceedings | 2011
|All Clear for Takeoff : Evidence from Airports on the Effects of Infrastructure Privatization
TIBKAT | 2022
|