Demand forecasting of smart tourism integrating spatial metrology and deep learning.

To enhance the dynamic perception and accuracy of tourism demand forecasting in smart tourism scenarios, this paper proposes a forecasting framework integrating a spatial econometric model and deep learning. This framework aims to address the limitations of traditional methods, namely insufficient spatial correlation modeling and weak interpretability of deep learning models. The model leverages spatial lag factors, spatial agglomeration indicators, and regional interaction behavior features. It constructs a geographical dependency structure based on spatial econometric methods, which is then embedded into a long short-term memory (LSTM) network for joint forecasting. This design achieves a balance between time series modeling and spatial structure identification. In this study, three types of datasets are selected: tourist flow data of scenic spots in Beijing, online tourism behavior data from Ctrip, and GeoLife Global Positioning System Trajectory (GeoLife) data. A multi-dimensional experimental system covering 12 performance indicators is established. The results show that the optimized model achieves the following performance on the CityBrain Beijing Tourism Flow (CB-BJTF) dataset: mean absolute error (MAE) of 9.653, root mean square error (RMSE) of 12.118, mean absolute percentage error (MAPE) of 14.538%, and R