Delay Prediction of Flight Operation Network Based on Deep Learning Combination Model

Due to the correlation between flights in the airline's flight operation network, when a flight delay occurs, it usually leads to a wide range of delay propagation. In this paper, a GCN-GRU combined prediction model is proposed to predict the level of departure delay owing to the constraints of the airline flight operation network topological structure and the law of dynamic change with time. Specifically, a graph convolution network (GCN) is used to capture the spatial characteristics of flight delay propagation and the gated recurrent unit (GRU) is used to capture the temporal characteristics of flight delay. Experiments show that the GCN-GRU model can obtain the spatio-temporal features from flight delay data. And on real-world flight delay data sets, the prediction results are better than state-of-art baselines. In addition, the experimental results are analyzed by using complex network theory, and the applicability of the model is obtained.