Research On Tourist Flow Prediction In Scenic Areas Based On Web Search Data And TCN-LSTM Combined Model

The number of visitors to scenic regions increases as people’s disposable income and time increases,but the rapid growth of tourist flow offers numerous issues for government regulation and scenic area management,therefore precise forecasting of tourist flow in scenic areas is especially important.The flow of visitors to the landscape itself is also influenced by a number of factors.As a result,the development of accurate forecasting models is critical not only for guiding government and scenic spot management,but also for encouraging the healthy and sustainable development of China’s tourism industry and improving tourist satisfaction.This paper establishes a TCN-LSTM combination model based on network search data to predict the daily tourist flow of the scenic spots,and takes the MT.Siguniang scenic spot as an example.First,six keywords were identified through Baidu search,and then the tourist flow and Baidu search data were aggregated daily,weekly and monthly,for data analysis of tourist flow and Baidu search data.In order to further determine whether baidu search index can be used to predict the tourist flow,the daily tourist flow of MT.Siguniang and six keywords baidu search data correlation analysis and granger causal test,the results show that network search data can be used to predict MT.Siguniang tourist flow.Next,a TCN-LSTM passenger flow prediction model is established and the optimal hyperparameter settings are found using the grid search method,and the influence of the optimization algorithm on the model is considered.For the effect of the timeliness of the web search data,the TCN-LSTM model and the web search data from 0-3 days in advance were used to predict the daily passenger flow of MT.Siguniang scenic,and it was found that the search data from 1 day in advance had the highest prediction accuracy.In addition,we also compared the differences of the combined TCN-LSTM model with the long short-term memory network model LSTM and the temporal convolutional network model TCN for passenger flow prediction,and found that the prediction performance of the TCN-LSTM model has stronger generalization ability compared with the single model,and the TCN model can effectively improve the model training speed and speed up the convergence process.Although the TCN-LSTM model has higher prediction ability than the LSTM and TCN models and is able to portray peak and trough times,the accuracy of peak prediction is not high.This paper finally used the combined 1-3 days in advance data and holidays as features for model training to predict the visitor flow of the MT.Siguniang scenic,and the results showed that the combined 1-3 days of Baidu index and holiday features could predict the trend and peak of the visitor flow of the MT.Siguniang scenic more accurately.