基于KPCA⁃XGBoost机器学习的大跨体育场风荷载预测  

Wind Load Prediction for Large⁃Span Stadium Based on KPCA⁃XGBoost Machine Learning

作　　者：艾辉林[1] 王盛世 陶厚正 AI Huilin;WANG Shengshi;TAO Houzheng(College of Urban Construction and Safety Engineering,Shanghai Institute of Technology,Shanghai 201418,China)

机构地区：[1]上海应用技术学院城市建设与安全工程学院,上海201418

出　　处：《力学季刊》2024年第3期834-841,共8页Chinese Quarterly of Mechanics

基　　金：国家自然科学基金(51778365)。

摘　　要：大跨空间结构风荷载的取值是该类结构抗风设计关注重点,通常借助风洞试验或数值风洞确定,但其费用高周期长等特点限制其广泛应用.机器学习方法近年受到关注,逐渐应用于结构的风荷载预测并取得了不错的效果.利用核主成分分析(Kernel Principal Component Analysis,KPCA)对数据进行降维处理,借助可以集成学习的XGBoost机器学习模型,采用十折交叉验证对超参数进行选择,编写了基于机器学习的大跨空间结构风荷载预测程序.通过对多个已有工程项目风洞试验结果的学习训练和预测结果比对,证明该方法具有处理数据能力较强、预测效率较高及泛化能力较强等特点.随机选取未参与模型训练的风向角下数据进行模型准确性验证,结果表明模型的R2值均达到0.9左右,预测值与试验值较为接近,体型系数在迎风区的预测精度略低于背风区,而极值风压则在背风区的预测精度好于迎风区.The value of the wind load of long-span space structures is the focus of the wind-resistant design of this type of structure.It is usually carried out through wind tunnel tests or numerical wind tunnels.However,its high cost and long period limit its wide application,especially in the construction scheme stage.Machine learning methods have received attention in recent years and have gradually been applied to predict the wind loads on structures,achieving good results.Kernel Principal Component Analysis(KPCA)was used to reduce the dimensionality of the data.With the help of the XGBoost machine learning model that can be integrated with learning,ten-fold cross-validation was used to select hyperparameters,and a machine learning based wind load prediction program for large-span spatial structures was written.Through learning and training of wind tunnel test results of multiple existing engineering projects and comparison of prediction results,it was proved that this method had the characteristics of strong data processing ability,high prediction efficiency and strong generalization ability.Randomly selecting data from three wind direction angles that did not participate in model training for model accuracy verification,the results showed that R2 values of the model reached around 0.9,and the predicted values were close to the actual values,and the prediction accuracy of body shape coefficients in the upwind region was slightly lower than that in the leeward region,while the prediction accuracy of extreme wind pressure in the leeward region was better than that in the upwind region.

关 键 词：XGBoost KPCA 机器学习 体育场 风荷载预测 

分 类 号：TU312.1[建筑科学—结构工程]