Intelligent airport taxiing control method based on graph neural network and multi-agent reinforcement learning

Abstract

The invention provides an airport taxiing intelligent control method based on a graph neural network and multi-agent reinforcement learning. The method comprises the following steps: receiving taxiing state information and a taxiway space structure of each airplane i on an airport scene; converting the obtained state information and taxiway space structures of all aircrafts into a graph form, and substituting the graph form into a graph attention network and a deep neural network; the neural network selects taxiing actions for the corresponding aircrafts according to the obtained taxiway traffic data, the state information of all the aircrafts is trained, and the neural network of each aircraft i on the scene is updated and shared; the learning process is repeated, and all aircrafts can slide to specified destinations on the premise of avoiding sliding conflicts. The optimal and conflict-free taxiing control can be automatically executed under the condition that the airport flight plan and the taxiway traffic condition are given, and the aspects of reducing the taxiing time, reducing the oil consumption, improving the adaptability to different taxiway conditions and the like are obviously optimized.

Images