Research On Flight Delay Prediction Based On Petri Net And Fusion Deep Learning

With the rapid development of national economy and civil aviation industry,the delay of transit flights has brought great challenges to the normal operation of civil aviation.According to statistics,the number of flights in China is increasing year by year,and the flight normal rate is decreasing year by year amidst fluctuations.Traditional flight delay analyses mostly rely on manual estimation or using the flight’s own information attributes for delay prediction,which make it difficult to achieve accurate delay warning.The work of overstay flights on the ground is centered on ground protection services.Quantitative analyses of flexible and complex ground protection services,as well as accurate prediction of flight delay levels,are keys to improving the efficiency of airport flight operations.Based on the above considerations,this thesis describes the flight assurance process based on Petri Net to analyze the process rationality and key operations.Two fused deep learning models are constructed to predict the flight delay level.The main contents are as follows:(1)Based on Petri Net modeling theory,a ground support workflow model for a large domestic airport is established.The logical relationship in the model is described.The state space analysis method of Petri Net is used to analyze the rationality of the model,and the critical path in the model is extracted.The longest operation item in the critical path is used as a class of features,combined with other regular flight features as the input of the prediction model.(2)Considering the problems of gradient disappearance in convolutional neural networks and the unstable prediction accuracy of a single model,a Dense Net-LSTM fusion network model is constructed for flight delay prediction.The data features consisting of key working hours,flight attributes,airport delay attributes and weather information are input to the densely connected convolutional network to obtain feature information.A one-dimensional code is generated and input to the long short-term memory network to achieve delay level classification.(3)The flight delay prediction method based on Att Condense Net-LSTM is proposed,which uses Condense Net network to reduce the computation while guaranteeing the accuracy of delay classification.The model is optimized by embedding the attention mechanism and evaluated in terms of algorithm complexity and accuracy.Experiments show that the model can obtain better classification results with less computation.