Dual-driven prediction of wind-driven rain obstruction in street canyons

Accurate assessment of wind-driven rain (WDR) exposure on building facades in high-density cities is essential for moisture-proof design and building durability. However, existing methods struggle to balance computational efficiency and high-fidelity accuracy when handling aerodynamic shielding and non-linear wake coupling in complex street canyons. This study systematically quantifies the impact of multi-parameter coupling on the WDR obstruction factor by constructing a comprehensive database of over 26,000 high-fidelity CFD simulations in 2D continuous street canyons. Based on this dataset, a dual-driven surrogate model architecture is proposed: symbolic regression (SR) is utilized to extract a physically transparent explicit equation for the mean obstruction factor, while an artificial neural network (ANN) is employed for the high-resolution mapping of vertical spatial distributions. Through this framework, the height ratio is identified as the pivotal parameter governing the flow transition between acceleration enhancement and shielding suppression. Furthermore, building spacing dictates the horizontal reach of wake interference, while wind speed non-linearly elongates the aerodynamic rain shadow. Validation shows the SR and ANN models achieving R² values of 0.901 and 0.997, respectively. Crucially, a real-world application demonstrates that this multiplicative coupling framework improves isolated baseline predictions from R² = 0.93 (RMSE = 0.36) to R² = 0.97 (RMSE = 0.17). Notably, this 2-second correction effectively replaces a 1256-second CFD simulation. This framework enables high-fidelity, sub-second predictions of upstream building effects, providing an efficient assessment tool for performance-based building design.