Application Of Multivariate Time Series Model In Flight Passenger Load Factor Prediction

The rapid development of the civil aviation industry has led to increasingly fierce competition among air transport companies.Only when airlines carry out effective revenue management can they stand out in the fierce competition.The decision-making basis for airlines to conduct revenue management is to effectively forecast the demand for air passenger transport.Among them,flight load factor is an important measure of flight passenger demand.Accurate forecasting of airline's flight load factor data can help airlines grasp passenger travel needs in advance and further improve corporate management.Since the flight load factor prediction problem can be regarded as a time series prediction problem,this article uses the flight load factor data of MU5614 flights from 2018-09-01 to 2018-12-19 as the original data,and the multivariate time series model is used for modeling analysis and forecasting.First,this article obtains data from the SQL Server database and performs data cleaning.Secondly,this paper uses the passenger load factor collected on the day of departure as the response variable,and the passenger load factor collected from one day to five days before the departure date as a predictor variable to construct a time series vector.Again,this article looks at the missing values in the time series vector,and uses multiple filling methods to fill in the missing values.Finally,this article applies the time series model for modeling analysis.First of all,this article considers the application of ARIMA model and VAR model to fit and predict the time series vector,but the accuracy of the data prediction results of these two models is not high.Therefore,this article considers applying the ARIMAX model to predict flight load factor.The specific steps include: sequence preprocessing of the passenger load factor sequence,variable selection based on comprehensive consideration of complexity and accuracy,pre-whitening noise processing,fitting ARIMAX model and testing the model's significance.Finally,this paper uses the fitted ARIMAX model to predict the passenger load factor data in the next ten days,and the root mean square error of the prediction is 7.4264%.The prediction results are significantly better than the ARIMA model and the VAR model.In order to further improve the prediction accuracy and model stability,this paper also combines the ARIMAX model and the BP neural network model to predict,and uses the reciprocal error weighted average method to calculate the weight of a single model,thereby determining their proportion in the combined model.The prediction outcome of the combined model demonstrates that for 80%of the data points,the resultant prediction effect is better than the ARIMAX model,and the root mean square error of the prediction reaches 6.917%.The prediction effect is more accurate than the single model,and it has certain practice for airline operations.And guiding significance.