| The precision prediction of flight delays in flight delays is critical to further enhance the quality of airline flight operation.Therefore,according to the historical flight operation data of an airline in 2018,from the perspectives of a single flight and multiple flights in the next 1 hour,with high accuracy and efficiency as the goal,based on data mining methods,the quantitative prediction,short-term prediction and early warning optimization methods of flight delay time are studied,and theoretical and technical support is provided for the normal operation and control of flights.From the perspective of a single flight,a dimension table of flight delay impact parameters for airlines,airports,weather,and other aspects was established,and GRA and GA algorithms were used to construct a GRA-GA-BP flight delay time quantitative prediction model based on association mining.The results show that:compared with the unoptimized model,the MAE result of this model is the smallest,only 12.027,and R~2 is 0.938,its computing power is significantly improved,and the prediction effect is relatively best;After that,based on the K-means clustering mining method,a single flight delay early warning optimization model under five police levels and police areas was constructed,and meteorological environmental factors were introduced to improve the input parameters of the GRA-GA-BP model.The results show that:the R~2 of the optimized model has increased by at least 0.225,and the MAE has decreased by at least 1.618.Compared with the Elman neural network and ELM model,the result has the smallest residual distribution and the highest accuracy of delay warning,which meets the performance requirements of single flight delay warning.From the perspective of multiple flights in the next 1 hour,a time series under the average delay time per hour was constructed,and wavelet decomposition and improvement of Shapley value were used for the delay time sequence,and the ARMA-RBF flight delay time short-term prediction model for the next 24h,72h and 168h was constructed based on the time sequence mining method.The results show that:the performance of the improved model has been improved overall.Compared with a single ARMA and RBF model,the 24h model has the highest prediction accuracy,and its RMSE and MAE have decreased by at least 7.4829 and7.7629;After that,based on the K-means algorithm,the delay early warning optimization model of multiple flights in the next 1 hour for five police levels and districts were constructed.The Hilbert transform was introduced to explore the time-frequency characteristics of the components under wavelet decomposition,and the WD-ARMA-RBF modeling prediction was realized based on this.The results show that:compared with the 72h and 168h models,the 24h model has the smallest RMSE and MAE,and compared with the four models of WD-ARMA,ARMA,RBF,and GM(1,N),the errors are kept below 1.000,which satisfies ractical needs for the delay early warning of multiple flights in the next 1 hour.In this paper,a systematic and comprehensive research on comprehensive forecasting methods is carried out on the problem of flight delay time prediction.The constructed prediction models are effective and feasible,which further improves the theory and method of flight delay time prediction. |
