Data-Driven Prediction of Wind Pressures on Long-Span Cantilevered Roofs: A Hybrid Framework for Temporal and Spatial Modeling

Wind pressure prediction for long-span cantilevered roofs remains a difficult challenge in wind engineering community, given that the pressure field is governed by complex aerodynamic effects that can produce highly nonlinear and non-stationary pressure fluctuations. This study introduces a machine-learning framework for wind pressure prediction on a representative long-span cantilevered roof. The model combines signal preprocessing and optimized least squares support vector machine regression to improve the representation of noisy, multi-scale pressure signals. The dataset was obtained from systematic wind tunnel tests, and the predictive performance of the proposed framework was examined for both single-point wind pressure prediction and cross-point wind pressure prediction. The results demonstrate that the hybrid framework is effective not only for local time-series prediction at individual pressure taps, but also for cross-point prediction that reflects the spatio-temporal correlation of roof pressure responses. Within the range of tested conditions, the proposed method provides an effective data-driven approach for wind pressure prediction and supports more reliable assessment of wind effects on long-span roof structures.