Public Sports Facility Location Planning Based on LLM-Enhanced Deep Transfer Optimization

With the improvement of urban public service systems, selecting sites for public sports facilities is essential for enhancing well-being and promoting spatial equity. However, existing spatial statistical and optimization methods struggle to interpret policy semantics, address regional data gaps, and balance multiple objectives, leading to poor policy consistency and limited generalizability. This study proposes a large language-model-enhanced deep transfer optimization (LDTO) framework that integrates LLM-based semantic parsing with deep spatial representation learning through a Semantic-Spatial Collaborative Alignment mechanism. A cross-regional adaptive transfer module strengthens robustness in data-scarce areas via adversarial alignment and meta-parameter updating. A dynamic multi-objective reinforcement learning strategy further enables adaptive balancing of coverage, cost, and equity. Experiments show that LDTO significantly improves policy consistency, transferability, and multi-objective coordination, offering a new paradigm that fuses semantic understanding with spatial optimization for equitable urban governance.