Daily Tourism Flow Forecast Based On Machine Learning

In recent years,the development of China's society promoted the upgrading of traditional industries.The tourism industry has also entered a prime time for development.The vigorous development of tourism can greatly promote the improvement of national economy,but at the same time,the growing tourist arrivals has also brought some hidden dangers to domestic tourism industry.The tourist flow and resource allocation of tourist attractions gradually appear unbalanced in time and space,which has adverse impact on the management of tourist attractions and tourist experience.Hence,the use of scientific and effective visitor volume prediction models to predict the visitor flow in a accurate manner,improving the ability of prediction,prejudge,and preparation,is of great significance for enhancing the travel experience of tourists and optimizing the development of the tourism industry.Firstly,by analyzing the temporal and spatial distribution characteristics of daily tourism flow in tourist attractions,it is found that tourist flow has the characteristics of non-linearity,periodicity,uneven distribution in low and high seasons,and uneven distribution of holidays and non-holidays.And by analyzing the related influencing factors of daily tourism flow,we extract 6 types of factors that have a strong impact on daily tourism flow from the perspective of relevance and manipuility,and select 15 relevant indicators to provide theoretical foundation for the forecasting task,including weekends,holidays,tourism flow in the same period last year,air quality level,AQI index,concentration of 6 kinds of pollutants,4 kinds of weather type,high temperature,low temperature,average temperature.Secondly,after data pre-processing,6 indexes are selected from the 15 influencing indexes as features according to the feature selection.Then a SPCA-LSTM based model for daily tourism flow forecast is established.SPCA is adopted to reduce the dimensions of features,and long-short term memory neural network is adopted to learn the inner law of data and make prediction.Through comparative experiments based on data set collected from Siguniang mountain,it is proved that the SPCA-LSTM framework has a greater improvement in prediction performance than traditional machine learning models such as random forest and support vector regression,indicating that the SPCA-LSTM model has good performance.Due to the strong nonlinear characteristics of the specific part of data,the model often fails to fit the holiday data enough,further improvement is required.Finally,in order to address the problem that the SPCA-LSTM model fails to predict the tourism flow value in holiday accurately,the convolutional neural network is added to the original model.A tourism flow forecast model based on SPCA-CNNLSTM is established.The purpose of adding convolutional neural network is to extract local features of data from the time steps,strengthen the nonlinear fitting ability of the model,and improve the prediction accuracy.The experimental data was collected from Mount Siguniang.According to the comparative experiments,SPCA-CNNLSTM model is better than the SPCA-LSTM model and other comparative methods,it can also achieve a higher accuracy for holiday's tourism flow.