| With the rapid increase of China’s economic size and the improvement of national income,China’s civil aviation industry has ushered in a rapid development.At present,China has a large number of airlines,large airports and small and medium-sized airports,with increasing capacity year by year,with huge potential.According to the Civil Aviation Administration of China(CAAC),the number of passenger trips in 2018 reached 612 million,with approximately 11.5352 million flight hours and 4.2595 million trips,up 10.9%,8.9%and 7.6%respectively over the same period in 2017.These data show that China has made great progress in civil aviation,but also in the continuous rapid development.Although China’s civil aviation fleet is the world’s first class,but the current management level of domestic air transport industry is still in the original stage of manual workshop,and the gap with the world’s advanced airlines for at least 20-30 years.In an effort to align with advanced aviation revenue management technologies,China has spent billions of dollars in recent years trying to introduce the most widely available aviation revenue management systems,such as:PROS,airRM,etc.,with no success.The reason why it cannot be introduced is that there are obvious differences between China and foreign mature aviation markets.These differences are mainly reflected in the rapid growth of domestic civil aviation capacity,the fierce price war between airlines,the great influence of high-speed rail,and the concentration of passengers before take-off.Aviation industry is a low-profit industry.In order to improve the advantages of airlines in the fierce market competition,it is a strong demand in the civil aviation market to find a revenue management system suitable for the development of China’s civil aviation industry.Revenue management is inseparable from the control of passenger flow,and the forecast of airline passenger flow is the decision-making basis of air-line revenue management.Accurate forecasting model can help airlines develop more accurate sales strategies to reduce costs,maximize profits and stand out in the fierce competition.However,at present,domestic air passenger flow is affected by many factors and has great randomness and volatility.In the face of huge daily flight tasks,passenger flow prediction and seat control are extremely tedious tasks.Usually,the methods used refer to city size,year-on-year and sequential data,etc.,but this method is difficult to fully reflect the market rules and grasp the market demand.According to the characteristics of the domestic aviation industry,this paper applies machine learning algorithm to the passenger flow prediction of flight routes,and on the basis of the prediction results,puts forward the man-machine coordination seat control strategy to improve the airline revenue level.The main contents of this paper are as follows:(1)In order to predict flight passenger flow,this paper proposes two different neural network models and makes a comparative analysis of their differences.Firstly,this paper combines the time and space characteristics of aviation network,uses graph convolution and long and short term memory network to extract the time and space features of aviation simultaneously,so as to realize the unification of time and space,and make the prediction results of neural network achieve better accuracy.In order to simplify the process of data processing,this paper tries to predict flight passenger flow based on attention mechanism gated cyclic unit network.In order to obtain the spatial structure information of aviation,complex triplet information needs to be established first,so this paper tries to do feature engineering on the original data directly.After feature engineering,the data dimension will still be very high.If the direct training,not only the model training speed is slow,but also will affect the accuracy of prediction.Therefore,it is necessary to reduce dimension.Therefore,deep belief network is selected for dimensionality reduction in this section,and then logical gating unit based on attention mechanism is used to process the dimensionality reduction data set,so that the key features in time series can be preserved.The model simplifies the process of feature extraction,increases the expansibility and efficiency of the model,and achieves high prediction accuracy.(2)In order to provide global airline passenger flow information for cabin controllers,this paper analyzes the airline passenger flow data according to prior knowledge,removes the redundant information,and extracts key information from different angles by combining the Transfomer and TextCNN models,which not only improves the parallelism of the algorithm,but also improves the performance of the algorithm.Meanwhile,the high-level semantic features of the model are further explored.Moreover,the nature of residual network solves the problem of network degradation existing in deep network.The combination of local flight information and global route information provides a theoretical premise for maximizing aviation revenue.(3)On the basis of accurate prediction of flight and airline passenger flow,this paper puts forward the man-machine collaborative seat control strategy.In this paper,the depth time clustering algorithm is used to cluster the passenger flow forecast data in time series,discover the similar demand situation in history,and cluster-the samples with similar demand situation into the same cluster.In the process of accommodation adjustment,on the one hand,accommodation adjustment can be carried out according to the passenger flow predicted by the prediction model;on the other hand,accommodation adjustment schemes adopted in the case of similar demand in history can be provided to the accommodation adjustment personnel for reference,which provides quantitative reference standards for the accommodation adjustment personnel in the process.In order to realize a more automatic seat control system,this paper proposes a way to adjust the parameters of the simulation model by comparing the simulated passenger flow with the actual ticketing situation,and makes the passenger flow infinitely close to the real situation by considering various factors affecting the flight passenger flow.Further,with the support of the main content,the innovation of this paper focuses on the following three aspects:(1)In terms of flight passenger flow prediction,this paper puts forward two different models for prediction.In the fir-st model,aiming at the problem that flight passenger flow has complex spatio-temporal two-dimensional characteristics,this paper first predicts flight passenger flow through a relatively intuitive combination model of graph convolution long and short-term memory network.In this model,LMTN can extract time series features and spatial dimension information by Four-ier transform,so as to realize the unification of space and time for prediction.In the second model,this paper tries to use the original data to do feature engineering directly to simplify the process of feature extraction,and proposes a combined model based on attention mechanism and gated recurrent neural network to achieve flight passenger flow prediction.Firstly,deep belief network was used to reduce dimension of data,and then gated recurrent neural network and attention mechanism were used to extract important time series features.Compared with the graph convolutional LONG and short term memory network,the latter improves the accuracy of model prediction.(2)In terms of airline passenger flow prediction,an air-line passenger flow prediction algorithm based on t-TCNN twin-tower structure multi-semantic feature extraction is proposed for airline passenger flow with more complex influencing factors.The passenger flow prediction is realized by the parallelism of the algorithm.This is not only conducive to the efficiency of algorithm execution,but also for the complex air-line data,the high-level semantic features of the model can be further mined.(3)In the aspect of flight space control,an auxiliary space control strategy based on clustering of historical similarity is proposed.With in this paper,based on the prediction of creative time clustering algorithm,the use of depth of history on the temporal clustering of the passenger flow data and historical demand those samples for the same cluster,the class member will forecast of passenger flow compared with similarclusters in a similar cluster of shipping space for the current space control reference.This is of great value to the cabin adjuster when he is faced with sudden and periodic events.This paper analyzes the rapid development of China’s aviation market,fierce competition,different user habits and other aspects,and concludes that China needs to find the air passenger flow forecast scheme in line with the development status of domestic civil aviation.Based on the analysis of the booking data of civil aviation passengers,the paper puts forward the prediction of flight and airline passenger flow by machine learning algorithm.On the basis of the accurate forecast,this paper also puts forward the man-machine coordination cabin control strategy based on passenger flow forecast.From the research conclusion,the solutions to these problems have good application value. |
PDF Full Text Request