Prediction Of Travel Spatio-temporal Trend By Combining Linear And Nonlinear Regression Methods

The increase in the number of tourists not only promotes the booming development of tourism,but also requires the relevant tourism management departments to improve their own management level and business capabilities,so that tourists' tour experience and personal safety can be guaranteed.Based on the prediction results of the tourism trend prediction model,the number of service personnel can be increased in advance to improve the tourists' experience during the peak period,while the infrastructure construction of the scenic spot in the valley period can be carried out based on the long-term prediction results,which is conducive to the continuous improvement of the tourism quality of the scenic spot so as to win the sustainable development.Therefore,the development of tourism trend prediction research can promote the scientific management level of tourism related administrative departments.In this paper,a tourism trend prediction framework based on the temporal and spatial characteristics of variables is proposed.Through the prediction of Jiuzhaigou 's tourism trend,the prediction framework is determined to be feasible and excellent in prediction."Jiuzhaigou" Baidu index is chosen as the empirical research to explain the source of the variable,Jiuzhaigou is daily travel as explained variable source,through the analysis of the characteristics of time and space of Baidu index and the number of tourist,on the basis of the characteristics of space and time build a space data structure of the explanatory variables and time characteristics is more distinct explained variable,makes the explanatory power of the variable capacity is improved at the same time reduce explained variables prediction difficult.The research on the tourism trend prediction of Jiuzhaigou based on the spatial and temporal characteristics of variables is composed of the following parts :(1)the standard deviation ellipse is used to analyze the spatial differentiation features of Baidu index collected from different regions,and then the spatial clustering method combined with economic distance is used for clustering;(2)the principal component analysis method is used for spatial clustering results,and the Baidu index collected from different regions is directly used as the explanatory variable,which will cause serious multicollinearity problem.The use of the principal component as the explanatory variable can weaken the number and correlation of explanatory variables,so as to realize the purpose of weakening multicollinearity;(3)to Jiuzhaigou travel daily data time series decomposition method is used for processing,time changes resulting from the decomposition characteristics is more bright and concise than the trend component and cycle component and component,is more than using the trend component and component as explained variable,according to the weight of time variation characteristics corresponding with different prediction methods to help reduce degree of difficulty of forecasting is interpreted variable;(4)the combination model is constructed with linear and nonlinear prediction methods to predict the trend of tourism,so that the combination model has an excellent ability to predict the trend of the predicted objects with complex change characteristics.The combination model uses ridge regression method to predict the trend component,and XGBoost method to predict the residual component,to realize the combination of linearity and nonlinearity.The sum of predicted value and periodic component is the predicted result of Jiuzhaigou tourism trend.At the same time,it is determined that there is a 2-day lag period between the Baidu index of "Jiuzhaigou " and the tourism trend of Jiuzhaigou through calculation,which enables the prediction results of the tourism trend prediction model of Jiuzhaigou to be obtained 2 days in advance and increases the usability of the tourism trend prediction model.The partial prediction scores of the linear and nonlinear combination models for Jiuzhaigou 's tourism trend are as follows: the R2 score of the training set is 0.88,and the RMSE score is 3202.The R2 score of the validation set was 0.76,and the RMSE score was 4762.The scores of the training set and test set of the combined model show that the combined model has excellent ability of fitting and prediction outside the sample period.The combination model can accurately predict the occurrence time and duration of the peaks and troughs of the tourism trend,which reflects that the combination model can identify the turning point of the change of the tourism trend,so that the combination model can play a decision support role in the adjustment of the reception capacity of the scenic spot and the diversion of tourists.By using ridge regression,XGBoost and ARIMA models to predict the tourism trend of Jiuzhaigou,the combination model was determined to have the best prediction ability by comparing the prediction results of different models.Compared with previous studies,this paper is innovative in the following aspects:(1)collect search engine data in provincial administrative regions,study the spatial characteristics of search engine data,and enhance the interpretation ability of search engine data to the changes in tourism trends according to the spatial characteristics;(2)travel data time series decomposed component data,use the time characteristic is more distinct component data as explained variable,decrease the difficulty of the prediction and forecasting method of linear and nonlinear combination,combined model to predict the change of the tourism trend,results show that the combination forecast model with complex changes trend of tourism have good fitting ability and the ability to predict.