Abstract:

In order to solve the problem of airport departure delay prediction in air traffic management, the sequence to sequence (S2S) framework was used to combine the Gate Recurrent Unit (GRU) and Convolutional Neural Network (CNN). A flight delay prediction model based on S2S-CNN-GRU was proposed using the sequence-to-sequence structure. The CNN was used to capture the structural features of airport flight delay status as the input of the encoder, and the GRU was used to capture the time features of the delay status and output the prediction results as the decoder. At the same time, the attention mechanism framework was added to improve the accuracy of prediction. Finally, the effectiveness of this model was tested with the actual data of the United States and compared with other models. The results show that mean absolute error(MAE) of the prediction of this model is 3.03, and root mean square error(RMSE) is 5.82, which is significantly better than the prediction results of other models.

Key words: sequence to sequence(S2S)-convolutional neural network(CNN)-gate recurrent unit(GRU) (S2S-CNN-GRU) model, departure flight, delay prediction, neural network, feature extraction