摘要
混凝土抗压强度是评估混凝土结构性能与耐久性的重要指标,其发展规律受多种材料组成与养护条件的综合影响。传统强度预测通常依赖大量试验与经验公式,既耗时又难以准确揭示多因素作用下的强度变化规律。为此,文中基于机器学习方法构建预测模型,系统研究原材料配比及养护龄期对混凝土抗压强度的影响。具体而言,文中收集并整理了包含水泥、矿渣粉、粉煤灰、水、减水剂、粗骨料、细骨料、龄期等特征变量在内的试验数据集,并引入XGBoost、随机森林(RF)、高斯过程回归(GPR)以及多层感知机(MLP)等典型模型进行对比。结果表明,XGBoost在训练集与测试集上均展现出最低的预测误差和最高的决定系数,表现出优异的准确性与稳健性。通过特征重要性分析显示,龄期和水泥用量是影响强度的关键因素,其次为减水剂和水胶比,而粉煤灰和矿渣粉在后期强度增长中起到协同作用。研究结果为混凝土强度预测提供了一种高效、可靠的途径,并为配合比优化设计与高性能混凝土研发提供了数据驱动的理论支撑。
Concrete compressive strength is a key indicator of structural performance and durability,influenced by material composition and curing conditions.Traditional prediction methods rely on experiments and empirical formulas,which are time-consuming and often inaccumte under multi-factor interactions.In this study,a dataset including cement,slng,fly ash,water,superplasticizer,aggregates,and age was used to train machine learning models.Comparative analysis of XG-Boost,Random Forest(RF),Gaussian Process Regression(GPR),and Multi-Layer Perceptron(MLP)shows that XG-Boost achieved the best performance with the lowest errors and highest R2values.Feature importance analysis identified age and cement as the dominant factors,followed by superplasticizer and water-to-binder ratio,while slag and fly ash contributed to later-age strength.The results demonstrate that machine learning provides an efficient and reliable approach for predicting compressive strength and optimizing mix design.
作者
徐道宏
Xu Daohong(Fangyuan Construction Group Co.,Ltd.,Taizhou Zhejiang 318000,China)
出处
《山西建筑》
2026年第2期115-119,共5页
Shanxi Architecture
基金
浙江省建设厅建设科研项目:地下增层开挖卸载条件下群桩基础协同行为研究(项目编号:2022K185)
台州市科技计划项目:智能化基坑风险预测系统开发与应用研究(项目编号:24gyb01)。
