Optimizing the spatial coupling mechanism of mountain tourism architecture is essential for improving spatial adaptability, functional coordination, and sustainable development under complex environmental constraints. This study proposes an artificial intelligence-driven optimization model based on environmental behavior theory for intelligent spatial analysis and decision-making. The model integrates a robustly optimized BERT pretraining approach and K-medoids clustering to extract and classify tourist behavioral characteristics, while Gated Graph Neural Networks and Temporal Convolutional Networks perform multidimensional spatial coupling analysis by considering environmental interactions and functional relationships. A Genetic Algorithm combined with Simulated Annealing is further employed to optimize the spatial coupling mechanism and improve adaptive configuration. Experimental results show that the proposed model achieves a pain-point identification accuracy of 95.32% and a coupling coordination level of 0.827, outperforming comparison models in spatial optimization effectiveness. The model contains 30.85M parameters and requires only 7.51 s for execution, demonstrating high computational efficiency. The above results indicate that the proposed framework can provide an effective data-driven method for intelligent analysis and layout optimization of mountain tourism architectural spaces, demonstrating good computational accuracy and coupling optimization effects in the case scenarios covered by the experiment, and providing methodological references and technical support for architectural spatial planning and facility configuration in similar mountain tourism areas.
