Research And Implementation Of Aviation Revenue Management Demand Forecasting Algorithm Based On Machine Learning

The aviation industry has an irreplaceable role and status in the national comprehensive transportation system.The development of the aviation industry is an important manifestation of a country's economic development.In recent years,the civil aviation industry has flourished and competition among airlines has become increasingly fierce.The development of intelligent and efficient revenue management systems has become the key to airlines'success in market competition.As a core part of the revenue management system,demand forecasting is the basis for the system to carry out subsequent cabin optimization configuration,inventory control and dynamic pricing.Accurate demand forecasting is the key for airlines to make right decisions and increase revenueThere are many influence factors in demand forecasting:the airline reservation data itself meets certain time series trends,and is also affected by factors such as weather factors,emergency factors,and competition between airlines.Most existing models only consider the influence factors of a single dimension,and cannot integrate multi-dimensional information to build models and analyze.Moreover,most of the existing research on demand forecasting focuses on the forecast of the total number of reservations at the time of flight departure,ignoring the growth forecast for the number of daily reservations during the flight pre-sale period.The forecast of the cumulative number of reservations per day during the pre-sale period is an important basis for the airlines to make corresponding control measures in the actual production scenario,and has important practical application value.In view of the above problems.combined with the actual needs of airlines,this thesis conducts in-depth research on the daily forecast of the number of reservations during the pre-sale period and the forecast of the total number of reservations at the time of flight departure,and proposes corresponding algorithms.The accuracy of the proposed algorithm is verified by the airline's real reservation data.Aiming at the problem of forecasting the number of daily reservations during the pre-sale period.this thesis proposes a prediction algorithm based on long short term memory(LSTM)neural network.Utilizing LSTM's superiority in processing time series data,fusing multi-dimensional information to extract feature,we build the horizontal timing(subscription date timing)model and vertical timing(flight departure date timing)model,in which the time series characteristics of booking data and related influencing factors are captured at the same time,enable effective forecasting of daily booking during the pre-sale period of a flight.This model makes up for the lack of research on the forecast of flight daily reservations,and plays an important role in guiding airlines to carry out subsequent cabin regulation and dynamic pricing,and has important practical application significance.Aiming at the problem of final booking number forecasting,this thesis proposes a prediction algorithm based on improved pick up model.Based on the traditional pick up model,the method of weighting cabin number is introduced to overcome the interference of the different total number of cabins on different dates.Moreover,the relationship between the reservation data of the same period and the ring period are used to mine the inherent time connection of the historical data.Then a regression model based on the ring period prediction and the same period prediction is finally established.The experimental results show that the accuracy of the moodel is higher than the traditional prediction model at different prediction levels.Finally,based on the above two prediction algorithms,this thesis builds an aviation reservation real-time prediction system,which can retrieve historical booking data from the airline's database,transfer it to the machine learning model for predictive analysis,and return the prediction results to the airline for real-time display.The system has the significance of guiding revenue managers to develop fares and provide a scientific theoretical basis for subsequent revenue management tasks.