Research on pressure prediction of complex wind pressure measuring points in typical structures based on artificial intelligence

Abstract

In wind tunnel testing, pressure gauges are typically used to measure wind pressure distribution. However, in areas with complex geometric shapes, arranging pressure measurement points is often challenging. It is worth noting that these regions with significant geometric variations exhibit highly complex wind pressure fluctuations (usually non-Gaussian), which are crucial for structural wind resistance design and make them key observation points. To address this issue, this study aims to propose a method that combines modal decomposition and deep learning to accurately predict wind pressure data for difficult to measure observation points using measurements from surrounding pressure taps. Wind tunnel tests were conducted on typical structures such as large-span roof structures, bridges, and high-rise buildings, and the proposed method was validated using experimental results. Taking skewness prediction as an example, the research results show that the Bi-weighted POD-CNN-LSTM method is superior to other methods, with a mean square error (MSE) range of 0.24–0.26 and a correlation coefficient (R) range of 0.9115–0.924. This technology can be widely applied to various wind tunnel tests, improving its applicability.