基于随机森林的高性能混凝土抗压强度预测  

Characteristic screening and prediction of high-performance concrete compressive strength based on random forest method

作　　者：吴贤国[1] 刘鹏程 陈虹宇[2] 曾铁梅 徐文 WU Xianguo;LIU Pengcheng;CHEN Hongyu;ZENG Tiemei;XU Wen(School of Civil Engineering&Mechanics,Huazhong University of Science and Technology,Wuhan 430074,China;School of Civil and Environmental Engineering,Nanyang Technological University,639798,Singapore;Wuhan Metro Group Co.,Ltd.,Wuhan 430030,China;College of Architecture and Civil Engineering,Xinyang Normal University,Xinyang 464000,China)

机构地区：[1]华中科技大学土木工程与力学学院,湖北武汉430074 [2]南洋理工大学土木工程与环境学院,新加坡639798 [3]武汉地铁集团有限公司,湖北武汉430030 [4]信阳师范学院建筑与土木工程学院,河南信阳464000

出　　处：《混凝土》2022年第1期17-20,24,共5页Concrete

基　　金：国家重点研发项目(2016YFC0800208);国家自然科学基金(51378235,51778262,71571078,51308240)。

摘　　要：高性能混凝土抗压强度的准确预测是配合比设计优化的关键步骤之一,基于机器学习算法的预测结果容易受到输入变量的影响。提出一种经过特征筛选的抗压强度随机森林预测模型。以水胶比、水泥用量、水泥强度、砂石、粉煤灰掺量和外加剂用量作为原材料指标,通过试验收集56份数据样本。通过变量重要性度量剔除重要性较低的特征,再利用优化后的输入指标进行强度预测,并与未经过特征筛选的随机森林模型和BP神经网络模型的性能进行比较。结果表明,水泥掺量对于高性能混凝土强度的贡献最大,而外加剂的影响较小;随机森林模型的预测精度较高(R^(2)=0.969 09),误差较小(RMSE=0.014 922);基于变量重要性度量的特征筛选对于提高预测精度具有重要意义。The accurate prediction of compressive strength of high-performance concrete is one of the key steps of mix proportion optimization.The prediction results based on machine learning algorithm are easily affected by the input variables.In this study,a random forest prediction model of compressive strength selected by feature screening is proposed.56 data samples were collected through the experiment with water cement ratio,cement dosage,cement strength,sand,fly ash dosage and admixture dosage as raw material indexes.The features with lower importance are eliminated by the measure of variable importance,and then the strength is predicted by the optimized input variables.Moreover,the performance of the model is compared with that of the random forest model and BP neural network model without feature selection.The results show that the contribution of cement content to the strength of HPC is the largest,while the influence of admixture is small.The prediction accuracy of random forest model is high(R^(2)=0.969 09),and the error is small(RMSE=0.014 922).The feature selection based on the measure of variable importance is of great significance to improve the prediction accuracy.

关 键 词：高性能混凝土 随机森林 变量重要性度量 强度预测 

分 类 号：TU528.01[建筑科学—建筑技术科学]