A multi-index prediction method for flight delay based on long short-term memory network model

A multi-input multi-output time series prediction model based on the long short-term memory network is proposed for the flight delay multi-index prediction. The model considers multi-dimensional factors such as the flight volume, the planned departure time and arrival time, the actual departure time and arrival time, flight attributes, airport and city weather conditions, etc. Combined with the delay status of the past period of time, the multiple delay indicators such as the flight delay rate and average delay time can be predicted at the same time. The actual flight operation data of one high volume airport are used to verify the model, and the influence of the forecasting sliding window length is further discussed. The model has good prediction accuracy and can track the trends of multiple delay indicators well. The method can find the change of the delay trends in advance and can provide some useful decision support information for flight scheduling management.