Demand Forecast In Revenue Management Of Civil Aviation

With the intensifying competition in the civil aviation market,revenue management has increasingly become an important management and decision-making tool in the civil aviation industry,and the accuracy of demand forecasting directly determines the effectiveness of revenue management,and thus largely determines the revenue of airlines.This article studies the problem of flight booking demand forecasting and predicts the future flight booking demand based on existing sales data.This is a basic task in flight management.At present,it is mainly based on human experience to estimate the demand.Scientific and accurate forecasting methods will play an important role in supporting flight operation management and cabin control.This paper presents Flight Pred,an aviation reservation demand forecasting algorithm based on multi-model integration.The algorithm is designed and implemented in detail,and simulated and verified using real flight operation data.The main work and contributions of this article are as follows.1.An algorithm-Flight Pred algorithm is proposed to solve the problem of flight booking demand forecasting.On the basis of in-depth analysis of flight booking demand forecasting problems and related data,the frame structure,working principle and fusion strategy of Flight Pred algorithm are specifically designed,and the three main modules that make up the algorithm are designed in detail.The three modules are: Light GBM module based on gradient boosting decision tree,DNN module for competitive gated deep neural network and Seq2 Seq module for time series recurrent neural network.The Flight Pred algorithm integrates these three pros and cons of each prediction model,taking advantage of the shortcomings,greatly improving the prediction accuracy.On the issue of flight demand forecasting,the forecasting algorithm designed in this way is the first to be proposed.2.Designed the implementation details of Flight Pred flight demand forecasting algorithm.The construction of the data set is described in detail,the feature extraction process of the data set is described,and the influence of the addition of features on the prediction results is analyzed through numerical experiments.The hyperparameters of each module of Flight Pred prediction algorithm are analyzed in detail and the value of each hyperparameter is determined.3.Multi-level simulation experiment analysis and evaluation of Flight Pred algorithm.First,the single model prediction effect of each module in Flight Pred algorithm was simulated and evaluated,including gradient lifting decision tree model Light GBM,competitive gated deep neural network DNN model,time series recurrent neural network Seq2 Seq model.Secondly,a simulation analysis of the fusion prediction effect of multiple models formed by these models under different initial conditions was carried out.Finally,the overall prediction effect of Flight Pred algorithm is simulated,analyzed and evaluated.Simulation experiment results show that the Flight Pred flight demand forecasting algorithm proposed in this paper has high forecasting accuracy and stability,and is a very practical flight booking demand forecasting algorithm.