Research On Flight Delay Propagation Prediction Based On CBAM-CondenseNet

In recent years,with the rapid development of the air transport industry,the problem of insufficient airspace resources and ground support capabilities has become increasingly prominent,resulting in frequent flight delays.The occurrence of flight delays and spread has brought great challenges to the normal operation of civil aviation.This paper takes flight delay propagation prediction as the goal,proposes the use of deep convolutional neural network to solve the problem of flight delay propagation prediction,and evaluates the degree of flight delay propagation.The main work of the thesis is as follows:Firstly,in view of the problem of flight delay spread,a series of large-scale flight delays are caused by the delay of the previous flight in the continuous flight,so a chain model of flight delay spread is proposed.According to the definition of the model,the flight chain data set with the characteristics of flight delay spread is constructed by integrating the characteristics of flight data.At the same time,in order to input the information features into the network model conveniently,the paper uses big data analysis and preprocessing technology to carry out data cleaning,data fusion and feature coding in turn.Finally,the construction of flight chain model and the preliminary data processing are completed.Secondly,a flight delay spread prediction model based on Condense Net algorithm is proposed.This model utilizes the design method of dense connection between each layer of network for feature extraction to classify and predict flight delay spread.Condense Net,as a new type of lightweight convolutional neural network,effectively solves the problem of the disappearance of gradients generated during the deep training of traditional neural networks.At the same time,the number of training parameters is less and the calculation cost is less.It is expected to be deployed on equipment with limited computing resources,and it will be more convenient and effective to provide decision-making suggestions to relevant departments.Through experimental analysis,Condense Net has good network performance and strong data processing capabilities,and has achieved high prediction accuracy in actual data experimental analysis.Finally,in order to improve the efficiency of Condense Net feature extraction and prediction accuracy,a prediction model of flight delay spread based on the CBAM-Condense Net algorithm is proposed.The model is based on the original Condense Net structure,and CBAM is integrated after each unit structure block.The improved model structure starts from channel attention and spatial attention,which makes the neural network have the ability to focus on its input feature subset.In the actual data experiment analysis,the performance of the improved CBAM-Condense Net algorithm is better than that of Condense Net algorithm in all aspects,and the prediction accuracy is higher.In addition,two evaluation indexes of flight delay spread are proposed from different perspectives.According to the prediction results of flight delay spread,the spread degree of flight delay can be evaluated and analyzed.According to the evaluation and analysis results of delay spread,it can provide corresponding suggestions for air traffic control departments and airlines to control delay spread and optimize flight plans.