基于机器学习方法的小跨高比SPRC连梁承载力预测  

Bearing capacity prediction of SPRC coupling beams with small span-to-height ratio based on machine learning method

作　　者：田建勃[1] 周文婧 陈黄健 赵勇 赵钦[1] 黄大观 闫靖帅 TIAN Jianbo;ZHOU Wenjing;CHEN Huangjian;ZHAO Yong;ZHAO Qin;HUANG Daguan;YAN Jingshuai(School of Civil Engineering and Architecture,Xi’an University of Technology,Xi’an 710048,China;China Construction First Group Corporation Limited,Beijing 100161,China)

机构地区：[1]西安理工大学土木建筑工程学院,陕西西安710048 [2]中国建筑一局(集团)有限公司,北京100161

出　　处：《地震工程与工程振动》2025年第1期85-94,共10页Earthquake Engineering and Engineering Dynamics

基　　金：国家自然科学基金联合基金重点支持项目(U2368203);国家自然科学基金项目(51608441);陕西省自然科学基础研究计划项目(2022JM-220)。

摘　　要：为了更方便地预测小跨高比钢板-混凝土组合(steel plate-concrete reinforced composite,SPRC)连梁的承载力,通过机器学习(machine learning,ML)的方法对SPRC连梁展开承载力预测模型研究,具有重要意义。首先收集了现有的试验数据建立了SPRC连梁数据库,在此基础上,通过极限学习机(extreme learning machine,ELM)算法、反向传播神经网络(back propagation neural network,BPNN)算法、支持向量机(support vector machine,SVM)算法、K临近(K nearest neighbor,KNN)算法、随机森林(random forest,RF)算法以及极端梯度提升(extreme gradient boosting,XGBoost)算法等6种ML算法进行了数据的回归训练。通过模型性能指标对比分析,发现基于XGBoost算法的预测模型具有最好的鲁棒性和泛化能力,相比于软化拉压杆模型(softened strut-and-tie model,SSTM)具有更高的计算精度和稳定性,并提出了基于ML方法的高精度SPRC连梁承载力预测模型。此外,还对影响SPRC连梁的承载力参数进行了敏感性分析,结果表明各特征参数对于SPRC连梁承载力的影响程度从大到小依次是:钢板配板率(ρp)、连梁截面高度(h)、连梁截面宽度(b)、跨高比(l n/h)、箍筋屈服强度(f_(vy))、纵筋配筋率(ρs)、纵筋屈服强度(f_(sy))、箍筋配箍率(ρt)、钢板屈服强度(f_(py))、混凝土抗压强度(f_(cu))。In order to predict the bearing capacity of steel plate-concrete reinforced composite(SPRC)coupling beams more conveniently.In this paper,it is of great significance to study the bearing capacity prediction model of SPRC coupling beams by machine learning(ML)method.Firstly,the SPRC coupling beam database is established by collecting the existing experimental data.On this basis,six ML algorithms,including extreme learning machine(ELM)algorithm,back propagation neural network(BPNN)algorithm,support vector machine(SVM)algorithm,K-nearest neighbor(KNN)algorithm,random forest(RF)algorithm and extreme gradient boosting(XGBoost)algorithm were used for data regression training.Through the comparative analysis of model performance indicators,it is found that the prediction model based on XGBoost algorithm has the best robustness and generalization ability.Compared with the softened strut-and-tie model(SSTM),it has higher calculation accuracy and stability.A high-precision SPRC coupling beam bearing capacity prediction model based on ML method is proposed.In addition,the sensitivity analysis of the parameters affecting the bearing capacity of SPRC coupling beams is also carried out.The results show that the influence degree of each characteristic parameter on the bearing capacity of SPRC coupling beams is in descending order as follows:steel plate ratio(ρp),coupling beam section height(h),coupling beam section width(b),span-depth ratio(l n/h),stirrup yield strength(f_(vy)),longitudinal reinforcement ratio(ρs),longitudinal reinforcement yield strength(f_(sy)),stirrup ratio(ρt),steel plate yield strength(f_(py)),concrete compressive strength(f_(cu)).

关 键 词：小跨高比 SPRC连梁 机器学习 鲁棒性 承载力预测 

分 类 号：TU398.2[建筑科学—结构工程]