基于机器学习的再生骨料混凝土梁抗剪承载力预测  

Prediction of shear capacity of recycled aggregate concrete beams with machine learning

作　　者：王欣悦 李秀领[1,2] 郭强 吕相蓉[1,2] 孙广 WANG Xinyue;LI Xiuling;GUO Qiang;LÜXiangrong;SUN Guang(School of Civil Engineering,Shandong Jianzhu University,Jinan 250101,China;Key Laboratory of Building Structural Retrofitting and Underground Space Engineering,Ministry of Education,Jinan 250101,China;The Second Construction Limited Company of China Construction Eighth Engineering Division,Jinan 250000,China;China Construction Second Engineering Bureau Co.,Ltd.,Beijing 100000,China)

机构地区：[1]山东建筑大学土木工程学院,山东济南250101 [2]建筑结构加固改造与地下空间工程教育部重点实验室,山东济南250101 [3]中建八局第二建设有限公司,山东济南250000 [4]中国建筑第二工程局有限公司,北京100000

出　　处：《混凝土》2024年第4期71-76,共6页Concrete

基　　金：国家自然科学基金面上项目(51278290);山东省自然科学基金(ZR2020ME245);山东省重点研发计划(重大科技创新工程)(2021CXGC011204)。

摘　　要：为解决当前再生骨料混凝土(RAC)梁缺乏统一抗剪承载力计算模型、相关试验工作量大且难以得出规律性结论等问题,建立了基于机器学习的RAC梁抗剪承载力预测模型。根据既有文献收集468根RAC矩形梁试件的抗剪性能试验数据,通过研究截面宽度、截面有效高度、再生粗骨料取代率、立方体抗压强度、轴心抗拉强度、剪跨比、纵筋配筋率和配箍特征值对再生骨料混凝土梁抗剪承载力的影响,结合逻辑回归(LR)、决策树(DT)、AdaBoost(AB)、支持向量机(SVM)以及人工神经网络(ANN)5种机器学习算法进行学习和训练,建立再生骨料混凝土梁抗剪承载力预测模型并比较预测效果,分析不同机器学习算法预测精度。研究结果表明:ANN算法与AdaBoost算法均能准确预测出再生骨料混凝土梁抗剪承载力,决定系数R2大于0.9,平均绝对误差MAE分别为18.66、15.96。根据精度统计指标,建议再生骨料混凝土梁的预测计算优先使用ANN算法和AdaBoost算法。最后,基于收集试验数据与回归分析,提出RAC梁抗剪承载力建议式。In order to solve the problems that the current recycled aggregate concrete(RAC)beams lack a unified calculation model of shear capacity,the related test workload is large and it is difficult to draw regular conclusions,a prediction model of shear capacity of RAC beams based on machine learning is established.The shear performance test data of 468 RAC rectangular beams are collected according to the existing literature.By studying the effects of section width,section effective height,recycled coarse aggregate replacement rate,cube compressive strength,axial tensile strength,shear span ratio,longitudinal reinforcement ratio and stirrup characteristic value on the shear capacity of RAC beams,combined with five machine learning algorithms of logical regression(LR),decision tree(DT),AdaBoost(AB),support vector machine(SVM)and artificial neural network(ANN),the prediction model of shear bearing capacity of RAC beams is established,the prediction effect is compared,and the prediction accuracy of different machine learning algorithms is analyzed.The results show that ANN and AdaBoost can accurately predict the shear capacity of RAC beams,the determination coefficient R2 is more than 0.9,and the average absolute error MAE is 18.66 and 15.96 respectively.According to the accuracy statistical index,it is suggested that ANN and AdaBoost should be preferred in the prediction and calculation of RAC beams.Finally,based on the collected test data and nonlinear regression analysis,the proposed formula of shear capacity of RAC beam is proposed.

关 键 词：机器学习 数据库 再生骨料混凝土梁 抗剪承载力预测 

分 类 号：TU528.041[建筑科学—建筑技术科学]