基于反向传播神经网络的锈蚀RC柱抗压承载力预测  

Prediction of ultimate bearing capacity of corroded RC columns based on backpropagation neural network

作　　者：辛景舟 蒋黎明 王劼耘 马闻达 李双江 XIN Jingzhou;JIANG Liming;WANG Jieyun;MA Wenda;LI Shuangjiang(State Key Laboratory of Mountain Bridge and Tunnel Engineering,Chongqing Jiaotong University,Chongqing 400074,China;School of Civil Engineering,Chongqing Jiaotong University,Chongqing 400074,China;Guangxi Communication Investment Group Ltd.,Nanning 530022,China)

机构地区：[1]重庆交通大学省部共建山区桥梁及隧道工程国家重点实验室,重庆400074 [2]重庆交通大学土木工程学院,重庆400074 [3]广西交通投资集团有限公司,广西南宁530022

出　　处：《混凝土》2023年第6期19-25,共7页Concrete

基　　金：国家自然科学基金面上项目(52278292,51908111);重庆市自然科学基金(cstc2020jcyj-msxmX0532);重庆市博士后研究项目特别资助(2021XM1016)。

摘　　要：为了解决当前锈蚀钢筋混凝土(Reinforced concrete,RC)柱极限抗压承载力预测中模型普适性差、计算繁琐、精度有限问题,提出了一种基于反向传播神经网络(Backpropagation neural network,BPNN)的锈蚀RC柱极限抗压承载力预测方法。首先,考虑不同的试件尺寸、布筋方式、锈蚀率等因素影响,搜集了既有文献192个试验数据,并分为训练集与测试集;其次,基于训练集数据,通过BPNN算法训练回归各影响因素与锈蚀RC柱极限抗压承载力间的非线性映射关系,得到锈蚀RC柱极限抗压承载力预测模型;最后,基于测试集数据,对该模型的预测精度进行验证,与解析预测模型进行横向对比。结果表明:所提方法的预测结果与各类构件的试验数据吻合较好,决定系数达到0.981,整体精度高,普适性好;相较于解析方法,所提方法的预测精度显著提升,积分绝对误差和平均百分误差分别提升了46%和40%。所提方法可为锈蚀RC柱抗压性能的评估与预测提供参考。An ultimate compressive bearing capacity prediction model is established in order to solve the problems of poor universality,complicated calculation and limited accuracy of corroded reinforced concrete column.The prediction method based on back propagation neural network(BPNN)was proposed.Firstly,by collecting the 192 experimental data associated with the corroded reinforced concrete columns including influence of different section size,reinforcement layout,corrosion rate and other factors from the existing literature.Secondly,the BPNN algorithm is trained to obtain regressionally the nonlinear relationship between the influencing factors and ultimate compressive bearing capacity of the corroded reinforced concrete column calculated with a unified ultimate compressive bearing capacity prediction model.Finally,the prediction accuracy of the model is verified and compared with the analytical prediction model.The analysis results show that the prediction values of the proposed method are in good agreement with the experimental data of various components,and the determination coefficient reaches 0.981.The overall accuracy is high and the universality is good.Markedly improved prediction accuracy of BPNN compared to analytical methods.And the integral absolute error and average percentage error of BPNN are increased by 46%and 40%respectively compared with analytical model.The proposed method can provide a reference for the evaluation and prediction of compressive performance of corroded RC columns.

关 键 词：锈蚀钢筋混凝土柱 反向神经网络 抗压承载力 

分 类 号：TU528.01[建筑科学—建筑技术科学]