Airport And Flight Delay Prediction Research Based On Spatial-Temporal Correlation

With the economic and technological development,air transportation industry has been flourished.More and more passengers travel and work by air due to its convenience.However,the airport infrastructure and airspace are unable to meet the demand of the increasing number of passengers,resulting in increasing flight delays.Recently,the proposed models have focused on the single flight or airport delay,ignoring the spatial-correlated relationship of other airports in the network.Besides,the recent time-series prediction models are mainly single-step and short-term models which don't have instructiveness for air traffic control.In order to improve the shortcomings of the current researches,this paper modifies PageRank algorithm to analyze the spatial-temporal correlation,proposes a multi-step and long-term deep sequence model to predict the change of airport delay and a deep learning model to predict the flight delay.The main work is as follows:(1)Clean the dataset and use complex theory to analyze the air traffic network's structural characteristics.The result verifies the Chinese air traffic network has the small-world and scale-free characteristics.The correlation coefficient and PageRank algorithm can find the spatial-temporal relationship of different airports.Finally,the KMeans algorithm is employed to cluster the networks delay behavior and obtain the similar network delay state distribution.The result shows that the key nodes' delay status has a great impact on the entire network.(2)The multi-step deep sequence learning predict model for airport delay is proposed with using the spatial-temporal correlation information and other auxiliary features.The delay of airports is multiplied by the spatial features of PageRank and then send the weighted sequence input into LSTM and sequence-to-sequence model,which implement the joint mining of the spatio-temporal correlation and predict the trend of airports' delay.Finally,the accuracy of model is improved by introducing temporal attention mechanism and other auxiliary features,and the importance of each time step in the historical data can be obtained.The proposed model has a better performance in long-term prediction among a variety of algorithms.(3)Propose the single flight delay prediction model based on deep learning and introduce weather,network delay status prediction dataset and spatial-temporal correlation data into the model to provide more auxiliary feature.Faced with high-dimensional sparse feature data,the stacked sparse autoencoder and XGBoost regression model is proposed.The multi-layer autoencoder model was used to extract high-dimensional features automatically,and the reduced-dimensional features were input into the XGBoost regression layer.Among a variety of the contrast algorithms,the proposed model has a stronger ability to capture key features and better prediction accuracy.The spatiotemporal delay data significantly improves the model accuracy.Finally,the key features that affect flight delays are extracted.The single flight delay attribution model is proposed and analyze the impact of various features on flight delay.