Wind load assessment of free-form open roofs is a safety-critical task constrained by high-fidelity computational fluid dynamics (CFD) computational latency. Standard data-driven models often struggle to resolve significant spatial heterogeneity, leading to an efficiency-fidelity dilemma. To address this aerodynamic bottleneck, this paper proposes a physics-guided hybrid framework centered on adaptive aerodynamic zoning and particle swarm optimization-back propagation neural networks. A novel zoning strategy is introduced by synergizing unsupervised K-means clustering with a rigorous multi-criteria evaluation system effectively embedding the domain-specific logic of flow regimes into the data-driven architecture. Leveraging high-fidelity large eddy simulation datasets, the framework learns to map spatial coordinates to complex pressure coefficient fields with high precision. Ultimately, this work transcends the traditional trade-off between fidelity and efficiency establishing an interpretable physics-guided zoning paradigm that transforms structural wind engineering from empirical estimation to intelligent mechanism-aware precision prediction.
Peng et al. (Thu,) studied this question.