Research On Nonlinear Prediction Of Traffic Flow Based On Network Search Data

In recent years,with the improvement of people's material living standards,more and more people choose to travel on holidays and the tourism industry is booming.Predicting tourism demand is a key issue for tourism.By accurately predicting trends and patterns,governments and the private sector can develop well-organized tourism strategies,and provide services to visitors,benefiting from the growing tourism industry.Therefore,accurately predicting the number of tourists in the scenic spot has a strong practical significance.This paper mainly uses the network search data to make a non-linear prediction of the daily visitors of Jiuzhaigou.Eight network search keywords are identified by the range selection method and the direct word selection method,namely ?Jiuzhaigou?,?Sichuan Zhaizhaigou?,?Jiuzhaigou Pictures?,?Jiuzhaigou Tourism?,?Jiuzhaigou Tourism Raiders?,?Jiuzhaigou Map?,?Jiuzhaigou Scenic Area? and ?Jiuzhaigou Weather?.The Granger causality test was conducted on the daily tourist volume of Jiuzhaigou and the Baidu search data of eight keywords,and the conclusion that the network search data can be used to predict the amount of tourists in Jiuzhaigou is obtained.Using the traditional time series analysis method to analyze the time series of individual Jiuzhaigou tourists,the AR(1)model was established,R^2 reached 86%,RMSE was 0.372359,and the goodness of fit was higher,but it could not reflect the network search data.The impact of the amount of tourists.In order to study the population forecast of Jiuzhaigou based on network search data,the SVR algorithm model and GBRT algorithm model were combined with the network search data to fit and predict the tourist volume of Jiuzhaigou.In the SVR study,linear kernel function,polynomial kernel function and Gaussian kernel function are selected to fit the sample.It can be found that the performance of support vector regression is related to the choice of kernel function,and the R^2 of the Gaussian kernel function reaches 74%.The RMSE is 0.506183,and the goodness of fit is the best.At the same time,the prediction effect is also the best among the three kernel functions.The overall prediction trend is close to the true value,but even if it is the best fitting Gaussian kernel function among the three kernel functions,It is also inferior to the AR model in terms of fitness.In the GBRT study,the gradient prediction algorithm is used to fit the sample.The goodness of fit R^2 is 95% and the RMSE is 0.211645.The overall trend is similar to the real value,and it has excellent performance at each wave turning point.accurate predictions are made at both high peaks.Comparing the three prediction methods,the GBRT model with network search data is the best,and the prediction performance is the best.Finally,using the best-performing GBRT algorithm to study the timeliness of network search data,that is,whether the network search data has advancement in predicting the amount of tourists.Using Baidu search data one day to seven days in advance to predict the number of tourists in Jiuzhaigou,it is found that Baidu search data two days in advance,three days in advance and four days in advance can effectively improve the accuracy of the model,among which Baidu search data in advance two days is forecasting.The model has the highest accuracy when the Jiuzhaigou tourist volume data for the day.With RMSE as the criterion,the accuracy of the model is improved by 6.36%.This reflects the decision-making process of tourists from the side.Most people will conduct corresponding online searches for the destinations before they travel,mainly in the two or three days before the trip,including more searches two days in advance.Since GBRT has a good forecast for holidays and peak travel seasons(summer holidays,National Day holidays),Baidu search data two days in advance is used to fit and predict the summer vacation tourist data of Jiuzhaigou.The results show that Baidu search data two days ahead of time can not only predict the daily tourist volume trend of Jiuzhaigou scenic spot during the summer vacation,but also predict the peak value accurately.The goodness of fit R^2 reaches 99.53%.Based on the RMSE criterion,the accuracy of the model is improved by 4.5% compared with the prediction results of the current Baidu search data.