基于广义回归神经网络的大跨度球面屋盖风荷载预测及其应用

Wind load prediction of large-span domes based on generalized regression neural network and its application

大跨度屋盖结构脉动风荷载特性复杂,一般需要通过风洞试验确定。典型屋盖的风洞试验数据的积累为相似体型屋盖结构风荷载取值提供了依据。为拓展典型屋盖风洞试验数据的应用范围,基于广义回归神经网络,结合典型球面屋盖系列风洞试验建立了大跨度球面屋盖的风荷载预测模型。其中,风荷载由平均风压系数、脉动风压系数、偏度、峰度、3个自功率谱密度参数(包括高频段衰减斜率、无量纲谱峰值和无量纲峰值频率)以及互功率谱密度相干指数8个特征参数描述。通过交叉验证和试算确定了广义回归神经网络模型中的平滑因子取值。以安庆电厂球面网壳结构为例进行了风荷载预测,通过对比预测风荷载与风洞试验得到的风振分析结果,验证了预测方法的可行性。

The distribution and fluctuation of wind load on large-span roofs are complicated. Wind load on typical roofs can be sometimes determined based on the wind tunnel tests carried out on roofs of similar shape. To expand the application scope of the test data, generalized regression neural network （GRNN） was introduced. The prediction models on large-span domes were given, where the wind load was expressed by eight parameters： mean, RMS, skewness, kurtosis of wind pressure coefficients, three auto-spectral parameters （including descendent slope in high frequency range, peak reduced spectrum and reduced peak frequency） and coherence exponent for cross-spectra. Cross validation and trials were carried out to determine the smooth factor in the GRNN model. The wind load prediction was applied on Anqing power plant dome. The wind-induced responses were calculated and compared with the results of wind tunnel tests. The results are very close. Therefore, it can be concluded that GRNN is feasible in predicting wind load on roof structures.