大跨平屋盖风荷载特性及风压预测研究

Here, rigid model tests of a long-span flat roof structure were conducted in an atmospheric boundary layer wind tunnel. The wind pressure time histories of measured points on the roof surface were obtained, and characteristics of mean wind pressure and fluctuating wind pressure on the roof surface in typical wind direction were analyzed. Combining the proper orthogonal decomposition (POD) technique and BP neural network method, a machine learning method-POD-BPNN method for spatial interpolation of long-span structures was proposed to realize the high-efficiency prediction of wind pressure. The predicted mean wind pressure coefficient, fluctuating wind pressure coefficient and fluctuating wind pressure characteristics in time domain and frequency domain were consistent with wind tunnel test values. The results showed that using the proposed POD-BPNN method to predict wind pressure on surface of long-span structure is feasible.