Flight Delay Prediction In China Based On Delay Characteristics Analysis And Deep Learning

With the rapid development of the air transport industry,flight delay is becoming an increasingly common issue in air travel.Flight delays have caused great inconvenience to airports,airlines,and passengers,which not only greatly increase the costs of airports and airlines,but also affect the travel of passengers and the reputation of airlines.If the rules of flight delays can be learned from historical data,and a flight delay prediction model can be established,airports can be assisted to make corresponding decisions in advance.So,it can reduce airport operating costs as much as possible and improve the reputation of airlines.Flight delay prediction has important practical significance.To solve the existing research problems of the current situation in domestic and foreign research on flight delay prediction,this article focuses on two aspects of research: characteristics analysis of domestic flight delays and prediction models of flight delay building.By analyzing the characteristics of domestic flight delays,a prediction model of flight delays is built.This article mainly includes the following five aspects:The first section is the acquisition and basic characteristics analysis of domestic flight delay data.In this article,6,724,664 data from 2017 and 2018 is used.The basic characteristics and distribution characteristics of domestic flight delays are explored,by analyzing flight delay characteristics in airports and airlines,and flight delay characteristics of overall flights and external variables such as climate.The analysis of airport flight delay characteristics includes the basic characteristics and distribution characteristics of flight delays in airports.The analysis of airline flight delay characteristics includes the basic characteristics and distribution characteristics of airline flight delays.The delay characteristics analysis of the overall flight includes the delay characteristics of departure time,arrival time and flight time.The analysis of flight delay characteristics of external factors such as climatic factors includes the flight delay characteristics of weather factors and the seasonal characteristics of flight delay distribution.The second section is the analysis of factors affecting domestic flight delays based on congestion internalization theory,which is used to analyze the influencing factors of flight delays.Firstly,congestion internalization theory for airports is introduced,and related influencing factors are analyzed based on the theory.The definition of hub airports,hub airlines,and airport concentration for Chinese flight delays are defined.After that,congestion is measured by two dependent variables,excess travel time and departure delay based on the dual perspectives of airlines and passengers.Three independent variables of hub airport,hub airline,airport concentration,and control variables such as departure and arrival areas,weather,holidays,departure and arrival time,and seasons are used to build two models based on OLS: the basic model and the model of adding the hub airline effect.Through the analysis of the model results,important conclusions such as the internalization of congestion,the size of airport hubs,and the size of airline hubs are obtained.The third aspect is comparison and interaction analysis of the factors affecting domestic flight delays based on decision trees.This section analyzes the differences between different influencing factors,and which interactions are more likely to cause delays.After preprocessing the data,a univariate decision tree model and a cross-variable decision tree models of flight delay are built respectively.Through the analysis of the univariate decision tree models of flight delay,the comparison of the ability to cause delay of airport variables,flight variables,and external variables such as seasonality are discussed.Through the analysis of cross-variable decision tree models of flight delay,the interaction between the external climatic variables and airport variables on delays are discussed,the interaction between the external climatic variables and flight variables on delays are discussed,and the interaction between airport variables and flight variables on delays are discussed.The fourth aspect,a flight delay prediction model based on hub size and deep learning is established to predict delays of a single flight.Firstly,distribution of flight delays in airports is estimated and fitted by polynomial regression method.Then an improved DBN-SVR model is built.After that,the combination framework of flight delay prediction model based on polynomial regression and improved deep neural network is built,and parameters of the model are set.The powerful deep learning model can not only predict the delay of a single flight,but also ensure the prediction effect.Finally,the top ten domestic hub airports are empirical analyzed,prediction results between different classifiers are discussed.Finally,a route delay prediction model based on smoothing spline and LSTM neural network is established to predict delays of individual flights.Firstly,distribution of flight delays in routes is estimated and fitted by smoothing spline.Then a LSTM model is built.After that,the combination framework of delay prediction model based on smoothing spline and LSTM neural network is built.The data is preprocessed before building the model.The continuous variables are encoded by Min-Max normalization method,and the discrete variables are coded by One-Hot.Then,the time series function is constructed,and the parameters of the model are set.Finally,the top ten domestic routes are empirical analyzed,prediction results between different classifiers are discussed.