Deep-learning enabled real-time prediction of wind pressures on building surfaces using a single anemometer

Accurate prediction of wind pressure on building surfaces is essential for structural safety assessment and wind hazard mitigation. However, conventional approaches such as field measurements, wind tunnel testing, and high-fidelity CFD simulations are not capable of achieving for real-time, full-surface evaluation, particularly under sparse sensing condition. To address this challenge, this study develops a WGAN (Wasserstein Generative Adversarial Network) based framework for real-time prediction of full-surface wind pressure using only a single anemometer. The inflow wind speed signal measured at the building-top elevation is adopted as the model input, while the corresponding surface pressure fields obtained from wind tunnel experiments are used as the prediction targets. The adopted model is systematically evaluated under multiple wind directions and across different temporal resolutions, including 1s and 10s mean and extreme wind pressures. Results show that the model achieves high accuracy in predicting pressure distributions from single point inflow measurement and maintains stable prediction performance across multiple wind directions. Although extreme pressure prediction remains more challenging due to its intermittency and nonstationary characteristics, the model still demonstrates reasonable predictive capability, particularly at longer averaging scales. Overall, the proposed approach verifies the feasibility of real-time, full-surface wind pressure prediction from single point inflow measurement, providing a practical pathway for a single anemometer based structural monitoring and wind-resilient design applications.