To delve deeper into the complexities of flight delay propagation,we constructed an airport flight network using real flight operational data.This paper presents a flight delay propagation model based on the Susceptible-Infected-Recovered(SIR)model,and we utilized the gradient descent method to solve the model parameters.In the airport flight network,we utilized the maximum connected subgraph ratio(S)as the network's invulnerability index,and we analyzed its invulnerability through selective attack and random attack strategies.Furthermore,we compared the flight delay propagation model with the Long Short Term Memory(LSTM)prediction model and the Markov model for delay prediction,citing historical data for instance analysis.The results reveal that the average path length of the airport flight network is 2.387,with a clustering coefficient of 0.58.The degree distribution of the airport flight network follows a double power law distribution.The airport flight network demonstrates the characteristics of small-world and scale-free networks.In the invulnerability analysis of the airport flight network,a random attack that deletes a total of 300 nodes results in S＞0.5,indicating strong invulnerability of the airport flight network.When the proportion of selective attack deletion nodes ranges from 15％to 20％,the value of S is 0,indicating poor invulnerability of the airport flight network.Additionally,the Mean Absolute Percentage Error(MAPE)for the flight delay propagation model is 16.7％,while the MAPE values for the LSTM and Markov models are 21.11％and 67.68％,respectively.The MAPE value of the flight delay propagation model differs by 4.41 percentage point from the LSTM prediction model and by 50.98 percentage point from the Markov model,indicating its higher accuracy.As such,the flight delay propagation model serves as a reliable prediction tool for flight delays,offering valuable insights for airlines and airports to formulate effective delay mitigation measures.
safety engineeringcomplex networkinvulnerabilitySusceptible-Infected-Recovered(SIR)modelMarkov model
万方数据
维普
