Flight Delay Prediction Based On Deep SE-DenseNet

Nowadays,the rapid development of big data and artificial intelligence technology has benefited all trades and professions.In the civil aviation field,the concept of intelligent airport and air traffic control have been put forward for a long time.A large number of intelligent civil aviation applications have gradually landed.As a difficult problem in the civil aviation field,flight delay is urgently needed to achieve high-precision delay prediction.Therefore,taking this as a goal,the method based on big data and deep learning is proposed to predict flight delay.It aims to mine the potential value of data by efficient and intelligent algorithms,and provide reference for decision-making of air traffic control,airports,airlines and other relevant departments.The main research works of this thesis are as follows:Firstly,under the premise of combining flight data and considering various factors of flight delay,flight data,meteorological data and associated airport delay information are preprocessed with big data analysis technology respectively in this thesis,meanwhile the influence level of three characteristics is studied.The characteristic properties of the three data sets and the data preprocessing process are introduced in detail.At the same time,in order to further improve the accuracy of flight delay prediction,a data fusion method is proposed,which uses the associated primary key merging method to input three kinds of feature into the network simultaneously.Secondly,in order to solve the problem that the traditional Convolutional Neural Networks(CNN)is difficult to be trained due to the loss of feature information and vanishing gradients,thus a flight delay prediction model based on the DenseNet is constructed.The model uses the densely connected design to extract features automatically,and the Softmax classifier is used to predict the delay level of flight.Among them,Bottleneck-Compression layer and Padding are used to optimize the model.On America’s airspace flight data,the prediction accuracy reaches 96.82%,compared with the traditional CNN,the final accuracy of DenseNet is greatly improved.Finally,although the dense connection design of the DenseNet network can solve the the problem of vanishing gradients,but it also could lead to a large amount of redundant information is reused,and the weight correlation between feature channels is not learned from the feature dimension level,which reduces the efficiency of the feature extraction process.Thus,a Squeeze Excitation-Densely Connected Convolutional Network(SE-DenseNet)is proposed.SE-DenseNet adds an "SE module" for feature recalibration in each dense block based on the DenseNet.The results indicate that the prediction accuracy reaches 97.51%,which is superior to the DenseNet in terms of all aspects of algorithm performance.At the same time,the influencing factors of flight delay are analyzed.the prediction accuracies of SE-DenseNet can reaches 91.36%,97.03% and 94.96% respectively.All three characteristics can affect the delay,but the meteorological data has the greatest impact on flight delay.