Hybrid Machine Learning Framework for Wind Pressure Prediction

This research article presents a hybrid machine learning framework aimed at predicting wind pressure distributions on high-rise building façades using constrained sensor networks. The study addresses the challenges associated with conventional methods that often require extensive sensor deployments, which can be costly and limited by accessibility. The proposed methodology consists of four key stages: first, it reconstructs low-fidelity pressure fields from limited sensor data; second, it reduces dimensionality to extract dominant spatiotemporal features; third, it employs a long short-term memory network for dynamic mapping; and finally, it predicts high-fidelity pressure fields over time. The framework's efficacy is validated through wind tunnel data and case studies on various building façades under different sensor configurations. The results indicate that this approach offers superior accuracy compared to alternative machine learning models, demonstrating its potential for real-time wind pressure estimation in structural health monitoring and digital twin applications. The findings contribute to enhancing the understanding of wind-induced effects on building structures.