基于小样本机器学习的矩形截面高层建筑风荷载体型系数预测

Prediction of wind pressure shape coefficient of high-rise buildings with rectangular section based on small sample machine learning

为研究小样本训练集下机器学习矩形截面高层建筑整体风压预测效果,应用多种经典机器学习方法对建筑基底整体风荷载体型系数进行了预测模型的训练与测试。比较了不同机器学习方法训练得到的模型预测精度差异,并提出基于二阶集成的广义回归网络。经测试表明,基于该方法训练得到的预测模型具有良好的精度及鲁棒性。利用该预测模型预测了矩形截面高层建筑基底的顺风向和横风向最大体型系数随着长宽比的变化规律。预测结果表明,顺风向体型系数随长宽比增大先增大后减小,最大值出现在长宽比为2左右,而当长宽比大于4后基本稳定在约1.35;而横风向体型系数最不利值分布在0.35~1.10之间,且随长宽比增大呈增大趋势。

To study the efficiency of machine learning with small sample training sets for predicting the overall wind pressure of rectangular cross-section high-rise buildings,several typical machine learning methods were used to determine the overall wind pressure shape coefficient of the building.The accuracy of prediction models obtained from different machine learning methods was compared.A generalized regression network based on second-order integration was proposed,which can provide a prediction model with higher accuracy and better robustness.Using this model,the variation of the maximum shape coefficient along and across wind directions with aspect ratio was predicted for the high-rise building with rectangular section.The prediction results show that the maximum shape coefficient along wind increases to the maximum value when aspect ratio is 2 and then decreases with increasing aspect ratio.When the aspect ratio is greater than 4,the maximum shape coefficient is stabilized at a constant value of 1.35.The maximum shape coefficient in cross wind direction ranges between 0.35 and 1.10,which increases with the increase of aspect ratio.