Research On Prediction Method Of Minimum Ticket Price For Busy Main Lines Of Civil Aviation

In civil aviation industry,whether we can accurately grasp the changing trend of air ticket prices is one of the most concerned issues,which acts a significant role in affecting the economic benefits of a series of groups in this field,such as air passengers and ticket agents.Accurate prediction of low ticket prices is helpful to the flexible docking of civil aviation demand and supply and the full utilization of civil aviation resources.Currently,independent pricing is widely accepted and airline companies are able to hold more choices in pricing their flight products.However,there are considerable fluctuations and randomness in air ticket price,which is also susceptible to lots of factors.Furthermore,complicated and varied pricing stratagems are employed in different airlines.Therefore,it is challenging to predict air ticket price accurately.In this work,we collected real data of ticket prices from a specific website and analysed them with statistic approaches.As a result,there are a series of feathers in ticket price data,such as no obvious periodicity in a single flight sequence,large distinction in the sequences corresponding to the departure date and a regularity of"week" in the consecutive multiple flight sequences with the same flight number.Conventionally used approaches for air ticket prediction usually failed to sufficiently characterise price changes and hence meet the requirement for accurate price forecasting.Convolutional neural network(CNN)is an emerging technology widespread applied with its excellent capacities of information extraction and feature expression,providing great potential to solve this problem.Aiming at the problem of ticket price prediction,this paper proposed a new way of organizing ticket price data and two different kinds of ticket price prediction models based on convolution neural network.This paper designed a two-dimensional ticket price time slice structure,aiming at the fact that there are two dimensions of the ticket price sequence:inquiry date and departure date,and that the price sequence of successive flights of the same number has a strong periodic regularity of "week".According to this structure,the whole structure of the ticket price forecasting model is designed.Firstly,the unknown part of the time slice is filled by simple forecasting method,and then the complete time slice is constructed as the input of the follow-up model.Secondly,the multi-shape convolution neural network is used to extract the features of the time slice,and the multi-dimensional external features are added.Finally,the information is processed in different ways,such as suing the multi-layer fully connected network to fuse the features and using LongShortTerm Memory network to take time series feature extraction.Then the fine prediction of ticket price is realized.This paper carried out relevant experiments on the real ticket price data set of a booking website,and compared the two different ticket price prediction models proposed in this paper with several popular benchmark models.The models we presentedachieved better prediction results than other models in the two evaluation indexes of average absolute error and average absolute percentage error.The experimental results show that the proposed methods can effectively solve the problem of ticket price forecasting.