Pedestrian-level wind environment prediction and Human-AI collaborative optimization in football stadium-centered high-density urban blocks

High-quality pedestrian-level wind environment (PLWE) underpins urban resilience and public health. In football-stadium-centered high-density blocks, narrow circular wind corridors often emerge, producing co-occurring extreme high winds on windward sides and extreme low winds on leeward sides, inducing thermal discomfort, pedestrian-safety hazards, and pollutant accumulation. Existing studies lack rapid, multi-conditional optimization and largely focus on in-stadium microclimates, overlooking outdoor impacts in such composite urban scenarios. Accordingly, this study aims to enhance PLWE at the block and Street-Canyon scales without compromising land-use efficiency, and proposes a surrogate-model-based Human-AI collaboration workflow integrating performance simulation, surrogate modeling, sensitivity analysis, and platform implementation.The results indicate that stadium plan scale governs overall performance, height regulates vertical air exchange, rotation angle and plan shape redirect flow paths and control extreme wind speeds, whereas the stadium sky view factor shows no consistent effect. Inter-factor interactions display reinforcement, attenuation, or independence, necessitating coordinated control. Optimization indicates that, without reducing land-use efficiency, wind-environment quality can be improved by up to 34.9 % (block scale) and 38 % (Street-Canyon scale); circular plans perform best; for along-wind cases, larger scales with small rotation angles are preferred, whereas for oblique winds, smaller scales with large rotation angles are advantageous. Further optimization across varying conditions reveals convergence of patterns across block morphologies, supporting cross-block generality. The study elucidates wind-environment mechanisms of stadium-high density block complexes, advances generalizable optimization strategies, and demonstrates the feasibility of Human-AI collaboration in complex urban contexts, providing methodological support for health-oriented and resilient urban planning.