基于机器学习回归模型的三峡大坝混凝土强度预测  

Concrete Strength Prediction of the Three Gorges Dam Based on Machine Learning Regression Model

作　　者：徐潇航 胡张莉 刘加平[1] 李文伟 刘建忠 XU Xiaohang;HU Zhangli;LIU Jiaping;LI Wenwei;LIU Jianzhong(School of Materials Science and Engineering,Southeast University,Nanjing 211189,China;China Three Gorges Corporation,Beijing 100038,China;Jiangsu Sobute New Materials Co.,Ltd.,Nanjing 211103,China)

机构地区：[1]东南大学材料科学与工程学院,南京211189 [2]中国长江三峡集团有限公司,北京100038 [3]江苏苏博特新材料股份有限公司,南京211103

出　　处：《材料导报》2023年第2期41-49,共9页Materials Reports

基　　金：国家自然科学基金联合基金项目(U2040222);高性能土木工程材料国家重点实验室开放基金项目(2020CEM011)。

摘　　要：人工神经网络、决策树与支持向量机为目前混凝土强度预测的常用机器学习算法。为实现三峡大坝大体积混凝土原材料筛选以及配比经验的学习与应用,并对大坝维护以及其他水利工程的建设提供指导,本研究基于三峡大坝主体工程混凝土28 d抗压强度数据,构建了原材料性能及配合比与混凝土强度之间的关系,并结合随机森林特征权重排序与统计分析的方法,确定了水泥用量、混凝土温度、水灰比为影响三峡大坝混凝土抗压强度的关键特征参数。探讨了常用机器学习算法对三峡大坝28 d混凝土强度预测效果,依据固定特征参数、通用参数与超参数综合调优后的多种算法的预测结果对比可知,经体系化综合调优的Epsilon支持向量回归(SVR)算法在预测中更优。Artificial neural network,decision tree and support vector machine are commonly used machine learning algorithms for concrete strength prediction.For learning and applying of the experience about the raw material selection and mixture design of Three Gorges Dam mass concrete,and providing guidance for dam maintenance and construction of other water conservancy projects,this study established a relationship between material properties,mixture design and concrete strength based on the 28-day compressive strength data of concrete used in the main project of Three Gorges Dam.Combined with random forest characteristic weight ranking and statistical analysis,cement dosage,concrete temperature and water-to-cement ratio were found to be the three key characteristic parameters dominating the compressive strength of the Three Gorges Dam concrete.Meanwhile,the prediction efficiency of the commonly used machine learning algorithms on 28-day concrete strength of the Three Gorges Dam were discussed.According to the prediction results of the three different kinds of machine learning algorithms after comprehensive optimization of the fixed characteristic parameters,general parameters and hyper-parameters,the Epsilon-support vector regression(SVR)algorithm with systematic comprehensive optimization was found to be the best in prediction.

关 键 词：混凝土 三峡工程 抗压强度 机器学习 模型调优 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP183[自动化与计算机技术—控制科学与工程] TU502.6[建筑科学—建筑技术科学] TU528.01