Airport Delay Prediction Based on Spatiotemporal Analysis and Bi-LSTM Sequence Learning

The air traffic network is the small-world system whose delay propagates among the entire network quickly. These frequent delays bring huge economic and resources loss. Therefore, accurately predicting the airport delay is essential for the airlines or air traffic controllers to adjust flight schedules. This paper proposes a multi-step deep sequence learning model (Bi-LSTM+Seq2Seq) to predict airport delay which considers the spatial-temporal correlation of other airports in the network. Firstly, the dataset is processed to analyze the temporal delay correlations of airports based on the complex network theory. The PageRank and K-means algorithm are used to cluster the behavior of the networks and to know the state of the entire network. Secondly, based on time series data about the current state of the network and delay relationship between airports, the Bi-LSTM+Seq2Seq model has been proposed and trained. Through the experiments, the proposed model has better accuracy and stability compared with other prediction algorithms.