Accurate forecasting of the hotel accommodation demands is extremely critical to the sustainable development of tourism-related industries. In view of the ever-increasing tourism data, this paper constructs a deep learning framework to handle the prediction problem in the hotel accommodation demands. Taking China's Hainan province as an empirical example, the internet search index is used from August 2008 to May 2019 to forecast the overnight passenger flows for hotels accommodation in Hainan Province, China. Forecasting results indicate that compared to benchmark models, the constructed forecasting method can effectively simulate dynamic characteristics of the overnight passenger flows for the hotel accommodation and significantly improve the forecasting performance of the model. Forecasting results can provide necessary references for decision-making in tourism-related industries, and this forecasting framework can also be extended to other similar complex time series forecasting problems.

The search of consumers under internet environment reflects the potential tourism demands of tourists. This paper takes Sanya as an example and attempts to use the traffic flow data related to Sanya tourism from August 2009 to March 2016 and the network search data to construct the input sets of SVR model. And this paper also forecasts the domestic tourists received in Sanya by applying the ABA-SVR model, thereinto the Adaptive Bat Algorithm, ABA is used to optimize the free parameters of Support Vector Regression model, SVR. The 12-month forecasting results and the significance testing show that this method can effectively improve the forecast accuracy of the model. The forecast results can provide necessary reference for the macro-administration of policymaking departments related to tourism.

In order to evaluate the predictive ability of network search data of daily sampling frequency for monthly tourist flow, this paper predicts the monthly tourist flow of Chongqing, China. In consideration of the inconsistency of sampling frequency of network search data and tourist flow data, an autoregression mixed data sampling model (AR-MIDAS) is constructed for prediction to avoid the loss of information. This paper adopts factor analysis technology to extract the characteristic information contained in the consumer search data related to Chongqing tourism, and then puts the obtained comprehensive factor into the model for a prediction experiment. The research results show that AR-MIDAS model can improve the precision of monthly tourist flow prediction better than ARIMA and MIDAS prediction techniques. The research results can provide necessary reference for scientific decision-making of tourism related departments.