Attention-Enhanced Temporal Convolutional Network with Gating Mechanism for Tourism Forecasting

Accurate prediction of tourist numbers is very important for resource management, service quality improvement, and sustainable tourism development. Understanding tourist flow in advance can optimize economic benefits, reduce environmental pressure, and improve tourist satisfaction. However, the high complexity of tourism data, with significant time dependence and nonlinear dynamic characteristics, poses a great challenge to traditional forecasting methods. To this end, this paper proposes a forecasting framework that combines multiple mechanisms. The framework captures short-term and long-term dependence through time convolution filter, accurately models highly volatile patterns, and adopts Gated Recurrent Unit (GRU) to maintain the state of key information and improve the sensitivity of the model to dynamic changes. In addition, the self-attention mechanism learns global dependence and allocates attention in long time series, thus improving prediction accuracy and generalization ability. This hierarchical modeling strategy enhances the ability of the model to deal with non-stationary data and better captures complex interactions. Ultimately, the framework provides reliable support for strategic decision-making and lays a solid foundation for promoting the sustainable development of regional tourism.