基于多种机器学习模型的混凝土力学性能预测及对比研究  

Prediction and comparative study of concrete mechanical properties based on multiple machine learning models

作　　者：杨光 杨利香[1] YANG Guang;YANG Lixiang(Shanghai Institute of Building Science Co.,Ltd.,Shanghai 200032,China)

机构地区：[1]上海市建筑科学研究院有限公司,上海200032

出　　处：《混凝土与水泥制品》2025年第1期34-40,共7页China Concrete and Cement Products

基　　金：上海市社会发展科技攻关项目(20dz1202103)。

摘　　要：选择了BP神经网络模型、支持向量机、XGBoost模型这3种机器学习模型构建了混凝土原材料用量与28 d抗压强度之间的预测模型,比较了各模型的拟合度、泛化能力和预测误差。结果表明:XGBoost模型在拟合度和泛化能力方面表现较为优秀,是构建混凝土原材料用量与28 d抗压强度之间预测关系的最佳模型。Three machine learning algorithms,namely BP neural network model,support vector machine,and XGBoost model were selected to construct a prediction model between the amount of concrete raw materials and the 28 d compressive strength.The fitting degree,generalization ability,and prediction error of each model were compared.The results show that the XGBoost model performs well in terms of fitting and generalization ability,and it is the best model for predicting the relationship between the amount of concrete raw materials and the 28 d compressive strength.

关 键 词：机器学习 混凝土 力学性能 预测 XGBoost模型 神经网络 支持向量机 

分 类 号：TU528.07[建筑科学—建筑技术科学]