Comparative analysis of machine learning models for characterizing spatial variability of wind pressure coefficients

Abstract

This study explores the global proliferation of tall buildings, highlighting the critical examination of wind pressure during their design. It systematically addresses the spatial variability of mean pressure coefficients on building surfaces, an aspect often overlooked in prevailing design standards. Utilizing Computational Fluid Dynamics (CFD) simulations with RANS turbulence model, this investigation focuses on rectangular buildings with varying plan dimensions. CFD model validation against wind tunnel experiments yields consistently reliable results. Mean pressure coefficient characterization on a horizontal plane employs three machine learning models: Non-Linear Regression (NLR), Artificial Neural Network (ANN), and Support Vector Regression (SVR). The model's performance is evaluated based on various error metrics and the number of parameters required to describe the model. Results show all models performing well with accuracy exceeding R2 > 0.98. Notably, the NLR emerges as the optimal model, requiring the fewest parameters and incorporating observed physical constraints within the variations.