基于迎风侧首排数据与深度学习的长跨柔性光伏阵列风压分布预测

Abstract The long-span flexible photovoltaic (PV) structure is a key solution to the challenges in “PV+” development. However, its characteristics—long span, light weight, low stiffness, and high clearance—exacerbate wind-induced vibrations, with wind loads becoming a critical factor in structural design. Given the spatial distribution of wind pressure data and the limitation of pressure measurement points in wind tunnel tests, this study proposes a fully convolutional network (FCN) model, integrating multi-scale and skip connections within the convolutional neural network (CNN) framework, which predicts the wind pressure distribution on PV arrays using the wind pressure field of the first row of the flexible PV arrays. The results show that the relative error of the predicted wind pressure is about 9 %, with the correlation coefficient between the predicted and actual wind pressure exceeds 0.95. This demonstrates that the FCN model effectively captures the wind pressure characteristics of long-span flexible PV structures, offering significant scientific and engineering value.