An Analysis On The Tourist Flow Of Scenic Spot Based On ARIMA And SVR Model

In recent years,China’s tourism industry develops stably and rapidly.However,due to the Covid-19 epidemic in 2020,the tourism industry at home and abroad has been significantly influenced.Thanks to the initial achievements made in China’s epidemic prevention and control work,domestic tourism has begun to recover.Under the requirement of normalizing epidemic prevention and control,many scenic spots set out to regulate the tourist flow.During the periods of both epidemic prevention and normal times,crowds should be prohibited to scenic spots to ensure the personal safety of tourists,which puts forward a high request to the short-term real-time forecasts of tourist flow in scenic spots.Meanwhile,under the background of "Internet plus Initiative",the massive network search data provides more possibilities with the study of tourist flow in scenic spots.On the basis of summarizing and analyzing previous studies,this paper takes the daily tourist flow data of The Mt.Siguniangshan Scenic Area as an example to discuss and establish an appropriate analysis model,so as to achieve more accurate and timely forecasts.Based on the analysis of the features of daily tourist flow data in scenic spots,this paper decomposes the data into linear and nonlinear parts for processing respectively.Besides,we build a time series model for its linear features and a SVR model for its nonlinear features.In this process,time-sensitive variables such as network attention data and workday or not are introduced to help fit models.Then,the respective advantages of the two models are used,based on the analysis of linear features of tourist flow data by the ARIMA model,the prediction error was reconstructed and the SVR model was established to extract nonlinear features,and then the ARIMA-SVR combined model was obtained;based on the analysis of nonlinear features of SVR model,the ARIMA model was established to extract linear features based on its prediction error,and then the SVR-ARIMA combined model was obtained.Finally,the sum of the predicted values of the two single models is the predicted value of the combined model for the tourist flow in the scenic spots.In terms of the comparison of AIC values of the two time series models,it is found that the AIC values of the SARIMAX model including Baidu index sees a significant reduction from430.45 to 317.02 compared with the original model.It shows that the method combining the network attention data(Baidu index)and workday or not with the tourist flow data of scenic spots can effectively improve the prediction accuracy of the model.And it can solve the lag problem of traditional data acquisition better.Compared with the ARIMA model,the MSE value of the ARIMA-SVR model decreases from 0.592 to 0.579,and the MAE value falls from 0.598 to 0.587.And compared with the SVR model,the MSE value of the SVR-ARIMA model declines from 0.328 to 0.314,and the MAE value abates from 0.415 to 0.372.And the combined models perform better in the training set than the single models they were based on.It reveals that the combined model can make full use of the comprehensive information in the original data,and has smaller error and higher prediction accuracy compared with the single model,thus avoiding the limitations of the single model.Meanwhile,results demonstrate that the performance of the single model and the combined model based on ARIMA is better than the single model and the combined model based on SVR in the prediction results of the training set.But in the test set,the prediction is opposite.It indicates that the two models based on SVR have better generalization ability and better extrapolation prediction effect.The establishment of the combined models have positive significance for the management of tourist attractions under the requirements of epidemic prevention and control,and have certain reference value for the research of related issues.