基于集成学习的圆钢管再生混凝土短柱轴压承载力研究-《建筑科学与工程学报》

文章信息/Info

Title:: Study on axial compressive bearing capacity of recycled aggregate concrete-filled circular steel tubular stub columns based on ensemble learning

文章编号:: 1673-2049(2026)01-0075-10

作者:: 张玉琢¹，席萌¹，刘进隆²，张信龙^3,4，刘发起⁴; 1. 沈阳建筑大学管理学院，辽宁沈阳 110168； 2. 东南大学土木工程学院，江苏南京 211189； 3. 中国建筑东北设计研究院有限公司，辽宁沈阳 110000； 4. 哈尔滨工业大学结构工程灾变与控制教育部重点实验室，黑龙江哈尔滨 150090

Author(s):: ZHANG Yuzhuo¹, XI Meng¹, LIU Jinlong², ZHANG Xinlong^3,4, LIU Faqi⁴; 1. School of Management, Shenyang Jianzhu University, Shenyang 110168, Liaoning, China; 2. School of Civil Engineering, Southeast University, Nanjing 211189, Jiangsu, China; 3. China Northeast Architectural Design and Research Institute Co., Ltd., Shenyang 110000, Liaoning, China; 4. Key Lab of Structures Dynamic Behavior and Control of the Ministry of Education, Harbin Institute of Technology, Harbin 150090, Heilongjiang, China

关键词:: 机器学习; 集成学习; 钢管再生混凝土; 轴压承载力; SHAP

Keywords:: machine learning; ensemble learning; recycled aggregate concrete-filled steel tubular; axial compressive bearing capacity; SHAP

分类号:: TU392.3

DOI:: 10.19815/j.jace.2024.11076

文献标志码:: A

摘要:: 为快速、准确地计算圆钢管再生混凝土短柱的轴压承载力，了解设计变量对轴压承载力的影响机制，提出了一种基于集成学习算法的智能预测模型。通过收集现有文献资料建立包含120组试验样本的数据库，对设计变量与轴压承载力进行Spearman相关性分析，基于单一学习算法和集成学习中的Bagging与Boosting算法构建轴压承载力预测模型。在此基础上，引入拟合优度、均方根偏差和平均绝对偏差对模型性能进行综合评价，利用SHAP方法定量分析各输入变量对轴压承载力的影响机制。结果表明：轴压承载力与截面直径、混凝土抗压承载力与钢管抗压承载力之间的Spearman秩相关系数均超过0.9，钢管抗压承载力和混凝土抗压承载力是影响圆钢管再生混凝土短柱轴压承载力的关键因素；随机森林（RF）模型的综合性能最佳，拟合优度、均方根偏差和平均绝对偏差分别为0.973、20.001和12.758；相较于现行规范和规程中的3种计算方法，RF模型基于平均相对误差的预测精度分别提高了12.79%、19.87%和9.30%；与单一学习模型相比，集成学习模型在提升预测准确性方面表现出显著优势；研究成果可为钢管再生混凝土柱的设计提供参考。

Abstract:: In order to quickly and accurately calculate the axial compressive bearing capacity of recycled aggregate concretefilled circular steel tubular stub columns and to investigate the influence mechanism of design variables on the axial compressive bearing capacity, an intelligent prediction model based on ensemble learning algorithms was proposed. A database comprising 120 experimental samples was established by collecting existing literature, and Spearman correlation analysis was applied to assess the relationship between design variables and axial compressive bearing capacity. Prediction models for axial compressive bearing capacity were developed using both single and ensemble learning algorithms, such as Bagging and Boosting. On the basis, the model performance was comprehensively evaluated using the coefficient of determination, root mean square error, and mean absolute error. The SHAP method was employed to quantitatively analyze how each input variable affects the axial compressive bearing capacity. The results show that the Spearman rank correlation coefficients between the axial compressive bearing capacity and the section diameter, the axial compressive capacity of concrete and the axial compressive capacity of steel tubular are all greater than 0.9. Notably, the axial compressive capacity of concrete and the axial compressive capacity of steel tubular emerge as critical factors influencing the capacity of recycled aggregate concretefilled circular steel tubular stub columns. Among the models, the random forest (RF) model demonstrates superior performance, and the coefficient of determination, root mean square error, and mean absolute error are 0.973, 20.001 and 12.758, respectively. Compared to the three calculation methods in the current specifications and procedures, the RF model improves the prediction accuracy based on the average relative error by 12.79%, 19.87%, and 9.30%, respectively. The ensemble learning model significantly outperforms single model predictions, offering substantial improvements in prediction accuracy. The research findings can provide a reference for the design of recycled aggregate concretefilled circular steel tubular stub columns.

《建筑科学与工程学报》[ISSN:1673-2049/CN:61-1442/TU]

文章信息/Info

参考文献/References:

相似文献/References:

备注/Memo

常用功能

导航/Navigate

工具/Tools

统计/Statistics