Research On Optimization Of Flight Delay Prediction Based On Operation Data

In recent years,with the improvement and development of the air transportation network,air travel has gradually entered people’s lives and become the first choice for travel.However,flight delays have come along and become one of the major challenges limiting the development of the civil aviation industry.Therefore,this thesis takes certain domestic departing flights as the main subject of research,using flight delay data,and conducts targeted research from large-area flight delays and short-period flight delays respectively.The details of the study are as follows.Firstly,an introduction is made to the background of civil aviation development,the basic theoretical concept of flight delay is clarified,and the main influencing factors and degree of influence of delay are analysed.Through the analysis of current domestic and foreign scholars’ research and practical application production needs,it is decided to carry out delay prediction analysis based on historical operation data through machine learning and combination models,in order to achieve the expectation of improving prediction accuracy.For large area flight delay prediction.Firstly,the research time range and factors to be considered are determined,and the delay time of the period is taken as the index.After pre-processing the data,the prediction analysis of the flight delay time series is carried out based on the logistic regression prediction model and the BP neural network algorithm modelling respectively,and the delay series is taken as the initial value to predict its future development trend in the immediately following period,and finally combined with examples for testing.For short-period flight delay forecasting.Firstly,the length of the forecast period is determined,the delay time and number of flights are selected as indicators,and the data is pre-processed to establish an ARIMA model and LSTM network algorithm modelling as well as a combined forecast model.Using the characteristics of the model in forecasting,a combined forecasting model based on both weight assignment and combination optimisation is established to achieve the capture of the overall timing characteristics.Finally,the model evaluation metrics are compared to the single prediction model in conjunction with the algorithm example to determine the prediction model with better accuracy.Finally,the validity of the prediction model is again verified by testing the model on a larger scale and with more data sets based on a larger number of runs.