基于BP神经网络和支持向量回归对比预测新拌自密实混凝土的工作性能  

Comparative Prediction of Fresh Performance of Self-compacting Concrete Based on BP Neural Network and Support Vector Regression

作　　者：李亚冉 肖其远 刘凌晖 LI Yaran;XIAO Qiyuan;LIU Linghui(Department of Civil Engineering,Changzhi Vocational and Technical College,Changzhi 046000,China;School of Civil Engineering,Central South University,Changsha 410075,China)

机构地区：[1]长治职业技术学院土木工程系,山西长治046000 [2]中南大学土木工程学院,湖南长沙410075

出　　处：《材料科学与工程学报》2023年第4期610-619,共10页Journal of Materials Science and Engineering

基　　金：中南大学大型仪器设备共享基金资助项目(CSUZC202118)。

摘　　要：新拌自密实混凝土(SCC)的工作性能调控对其现场施工乃至后期硬化性能至关重要。为了进一步提高通过初始配合比预测新拌工作性参数的精确度,采用BP神经网络和支持向量回归(SVR)两种常见的机器学习计算模型来分析配合比输入变量与坍落扩展度、L型仪通过率和V型漏斗流出时间的潜在函数关系。在计算前按照60%,20%和20%的比例依次将数据集划分为训练集、验证集和测试集。采用三个机器学习性能评估指标,即决定系数R^(2),均方误差MSE和平均绝对百分比误差MAPE来表征和对比它们的预测精度差异。并且通过敏感性分析研究单一变量的变化对结果的影响趋势。结果表明,支持向量回归SVR模型对SCC坍落扩展度,L型仪通过率和V型漏斗流出时间的预测结果相比BPNN神经网络模型更高。以V型漏斗流出时间参数为例,基于SVR的总体数据集合R^(2),MSE和MAPE值分别为0.9213,0.8602和14.519,高于BP神经网络对应的0.9162,1.1281和18.007。参数敏感性分析结果表明,水泥掺量越高会导致坍落扩展度的减小和L型仪通过率,V型漏斗流出时间的增加。与此同时,水胶比和减水剂掺量的提高则会使得SCC扩展度提高。The workability regulation of fresh self-compacting concrete(SCC)is very important for its on-site construction and even hardened performance.To further improve the accuracy of predicting the workability parameters of fresh SCC through the initial mix proportion,two common machine learning calculation models,i.e.BP neural network and support vector regression(SVR)were used to analyze the potential functional relationship between the input variables of mix proportion and slump flow,L-box filling ratio and V-funnel flow time.Before calculation,the data set was divided into training set,verification set and test set according to the proportion of 60%,20%and 20%.Three machine learning performance evaluation indexes,namely determination coefficient(R^(2)),mean square error(MSE)and mean absolute percentage error(MAPE),were used to characterize and compare their prediction accuracy differences.Through sensitivity analysis,the change trends of the single variable on the results were studied.The results show that BP neural network has good learning ability for the slump flow.The prediction results of SVR model for SCC slump flow,L-box filling ratio and V-funnel flow time are higher than those of BPNN model.Taking the V-funnel flow time as an example,the overall data set R^(2),MSE and MAPE based on SVR are 0.9213,0.8602 and 14.519,respectively,which are higher than the corresponding 0.9162,1.1281 and 18.007 of BPNN,respectirely.The results of parameter sensitivity analysis show that the higher the cement content is,the lower the slump flow,the higher the L-box filling ratio and V-funnel flow time are.At the same time,the slump flow of SCC will be improved with the increase of water-to-binder ratio and superplasticizer content.

关 键 词：自密实混凝土 坍落扩展度 L型仪通过率 V型漏斗 BP神经网络 支持向量回归 

分 类 号：TU528[建筑科学—建筑技术科学]