Research On Delay Prediction And Delay Resolution Characteristics Of The Aviation Network

With the progress of society and the continuous improvement of the economic level,the demand for air transportation continues to increase.Facing the increasing demand for air transportation,how to improve efficiency,reduce flight delays,and thereby reduce social and economic losses has become a concern.In fact,severe weather,increased traffic,air traffic control,equipment failures,and delay propagation will cause varying degrees of disturbance to the normal operation of the aviation system,which will cause flight delays.Therefore,accurately predicting flight delays and grasping the delay resolution process under different disturbance conditions have important theoretical and practical significance for reducing or avoiding flight delays.This paper collects the data of flight operation record of the United States in 2016and 2017,as well as the corresponding weather data.From the perspective of complex networks,machine learning algorithms are used to predict flight delays,establish mechanism models to simulate the real operation state of the aviation system and disturbance events leading to flight delays,and the resolution process of flight delay under different conditions is analyzed.The paper includes the following three aspects:（1）Research on aviation network topology and departure delay characteristics.This paper takes the airport as the node and the OD pair with the flight as the edge to establish a directed aviation network₀.By the analysis of node degree distribution,network assortativity coefficient,average clustering coefficient and average shortest path length,a total of four indicators,the topological structure characteristics of American aviation network in 2016-2017 are studied.In addition,on the basis of₀,taking the average daily flight number not less than 10 as the standard,we obtain the American backbone aviation network₁,and study the departure delay characteristics and laws of American aviation network（including the possible range of delay propagation and the departure delay states of the aviation network）.（2）Prediction of departure delay of the aviation network.This paper takes the time,the weather of the origin,the departure and arrival delays of other nodes and the departure delay states of the aviation network as the input variables,and flight delay prediction models are established based on LSTM neural network and random forest algorithm respectively,to predict the departure delay of edges in aviation network at the next moment.The prediction performances of the two models are compared.（3）Study on the characteristics of delay resolution under different disturbances.In this paper,a mechanism model is established to simulate the real operation of the aviation system.On this basis,under the condition of given disturbance scale,different intensities of disturbances are set to observe the resolution process of flight delay;Under the condition of given disturbance intensity,different scales of disturbances are set to observe the resolution process of flight delay under various conditions.The research results show that the aviation network is a scale-free network,and the nodes with large degree tend to be connected with the nodes with small degree.In addition,the average clustering coefficient is high and the average shortest path length is small,which shows that the American aviation network has the characteristics of small-world network,and flight delays can be propagated repeatedly between nodes.With the increase of the time window size,the relative sizeof the giant strongly connected component in the edges that have experienced departure delay increases rapidly at first,and when the time window size reaches 3 hours or more,the change range ofis very small,indicating that the relative size of the possible propagation range of flight delay will increase rapidly,and then tend to be stable,and the propagation speed of flight delay on the aviation network is very fast.In the aspect of flight delay prediction,the flight delay prediction model based on double-layer LSTM with 15 neurons in each layer has the best comprehensive performance,which is superior to the flight delay prediction model based on random forest algorithm,and the average median absolute error is reduced by 44%.In terms of delay resolution,when the disturbance scale is given and the intensity increases,the flight delay will become more serious and the delay resolution will take longer;When the intensity is given and the scale is expanded,the departure delay of aviation system will increase first,and the delay resolution will take a long time.When the scale reaches a certain level,the departure delay and the delay resolution time will decrease.