基于集成学习方法的钢管混凝土柱轴压承载力预测  

Prediction of axial compressive capacity of concrete-filled steel tube columns based on ensemble learning methods

作　　者：童丰华 马骁健 石旷 史先振 史言稳 孙亭亭 TONG Fenghua;MA Xiaojian;SHI Kuang;SHI Xianzhen;SHI Yanwen;SUN Tingting(CCCC Second Highway Engineering Co.,Ltd.,Xi\an 710065,China;School of Transportation and Logistics Engineering,Wuhan University of Technology,Wuhan 430063,China;Scientific Research Institute,Hefei University of Technology,Hefei 230009,China;Jiangsu Hengyi Geotechnical Engineering Technology Co.,Ltd.,Nantong 226003,China;Jiangsu Jinpu Engineering Consulting Co.,Ltd.,Xuzhou 221700,China;School of Road Bridge&Harbor Engineering,Nanjing Communication Institute of Technology,Nanjing 211188,China)

机构地区：[1]中交第二公路工程局有限公司,陕西西安710065 [2]武汉理工大学交通与物流工程学院,湖北武汉430063 [3]合肥工业大学科研院,安徽合肥230009 [4]江苏恒一岩土工程技术有限公司,江苏南通226003 [5]江苏金普工程咨询有限公司,江苏徐州221700 [6]南京交通职业技术学院路桥与港航工程学院,江苏南京211188

出　　处：《四川建筑科学研究》2025年第1期29-40,共12页Sichuan Building Science

基　　金：江苏省产学研合作项目(BY20230330、BY20230087);江苏高校“青蓝工程”优秀青年骨干教师项目;南京交通技术学院院级重大科研项目(JZ2301)。

摘　　要：为提高钢管混凝土(concrete-filled steel tube,CFST)柱轴压承载力预测的泛用性、稳定性和准确性,建立了包含矩形截面和椭圆形截面CFST柱的数据集,采用回转半径统一输入参数,利用树结构parzen估计器(tree-structured parzen estimator,TPE)和交叉验证法进行了超参数优化,建立了4种机器学习单模型和4种集成学习模型进行训练并对测试集预测。结果表明:4种集成学习模型的表现均优于4种单模型,其中,XGboost在测试集上表现出最佳性能,决定系数R^(2)达到0.987;采用沙普利加性解释(shapley additive explanations,SHAP)法对极限梯度提升(exctreme gradient boosting,XGboost)模型进行了可解释性分析,长轴回转半径、混凝土抗压强度、钢外壳厚度对CSFT柱轴压承载力影响相对较大,SHAP法还可对单个样本进行局部解释,量化各参数对结果的影响。To enhance the generality,stability,and accuracy of predicting the axial compressive capacity of concrete-filled steel tube(CFST)columns,a dataset including rectangular and elliptical CFST columns was established.The dataset used a unified input parameter of the gyration radius and employed the tree-structured parzen estimator(TPE)and cross-validation for hyperparameter optimization.Four machine learning single models and four ensemble learning models were trained and tested on a validation set.The results indicate that the performance of the four ensemble learning models are superior to that of the four single models.Among them,XGBoost exhibits the best performance on the test set with a coefficient of determination(R^(2))reaching 0.987.The shapley additive explanations(SHAP)method was used to conduct interpretability analysis on the XGBoost model,revealing that the gyration radius along the major axis,compressive strength of concrete,and thickness of the steel shell have a relatively significant impact on the axial compressive capacity of CFST columns.Additionally,the SHAP method can provide local explanations for individual samples,quantifying the influence of each parameter on the results.

关 键 词：钢管混凝土柱 轴压承载力 机器学习 集成学习 

分 类 号：TU398.1[建筑科学—结构工程] TV335[水利工程—水工结构工程]