Intelligent Decision Making for Tanker Air Control Conflict Deployment

Abstract

Flight conflict, as the highest level of safety in air traffic control operation, has always been the focus of air control work. The research of air traffic control conflict deployment intelligence technology is the current hot direction. In this paper, we propose a control conflict deployment strategy solving method based on deep Q network (DQN). The value of action value function Q in Q learning algorithm is used as a criterion to evaluate the goodness of the strategy, and the multi-layer perceptron is used as a neural network to approximate the Q value; stochastic gradient descent algorithm is used to update the parameters of the neural network; the contradiction arising from the combination of neural network and Q learning is solved by the way of experience playback and establishment of dual network structure. A large amount of sample data of conflict scenes is generated through simulation data for the training solution of the model to obtain the optimal strategy. The experimental results show that the tanker control conflict deployment strategy obtained by the deep Q-network algorithm training in this paper can play a good effect in the designed multiple conflict scenarios, and also can better take into account the control rules and the overall airspace operation situation; it lays the foundation for the future control operation of the conflict deployment auxiliary decision-making technology.