基于HEMNG模型的混凝土抗压强度预测  

Predicting concrete compressive strength based on HEMNG model

作　　者：周继发 曾晓辉[1] 谢友均[1] 龙广成[1] 唐卓 周智 ZHOU Jifa;ZENG Xiaohui;XIE Youjun;LONG Guangcheng;TANG Zhuo;ZHOU Zhi(School of Civil Engineering,Central South University,Changsha 410075,China;School of Civil and Architecture Engineering,Hainan University,Haikou 570208,China)

机构地区：[1]中南大学土木工程学院,湖南长沙410075 [2]海南大学土木建筑工程学院,海南海口570208

出　　处：《铁道科学与工程学报》2025年第2期875-886,共12页Journal of Railway Science and Engineering

基　　金：国家自然科学基金资助项目(52078490,11790283);国家重点研发计划项目(2023YFB2604304-2)。

摘　　要：基于集成学习理论,首次将人工神经网络和极端梯度提升算法进行集成,提出一种全新的算法:HEMNG(hybrid ensemble model based on neural networks and gradient boosting),旨在更准确地预测混凝土抗压强度。采用303组混凝土配合比数据进行建模,以水胶比、砂率、浆骨比、粉煤灰替代比例和养护龄期5个可解释特征作为输入,抗压强度为输出。为了分析HEMNG模型在抗压强度预测中的优势,采用人工神经网络、极端梯度提升、支持向量机、随机森林等模型进行比较,并将模型迁移到全新数据中,以探究其在未知数据上的泛化能力。基于训练良好的HEMNG模型进行敏感性研究,量化3个重要特征对抗压强度的影响。结果表明:HEMNG模型采用5个可解释特征,可准确、可靠地预测抗压强度,在测试集中预测值与实际值的拟合度为0.961,均方根误差为2.704,模型预测精度和泛化能力均明显优于其他模型;将HEMNG模型迁移到新数据中,强度预测值与实际强度值较为吻合,最大绝对误差仅为7 MPa,模型表现出良好的稳健性;根据模型敏感性研究显示,存在一个最佳砂率使抗压强度达到最大;增大水胶比会降低混凝土抗压强度,最佳砂率会随水胶比增大而减小;随着浆骨比的增大,最佳砂率会表现出先增大后减小的趋势,模型能量化分析各参数对抗压强度的影响。开发的HEMNG模型为评估混凝土抗压强度提供了新的思路和方法。Based on ensemble learning theory,artificial neural networks and the extreme gradient boosting algorithm were integrated for the first time,resulting in the proposal of a novel algorithm called HEMNG(Hybrid Ensemble Model based on Neural Networks and Gradient Boosting),aiming at more accurately predicting concrete compressive strength.A dataset of 303 concrete mix proportions for modeling,with five interpretable features including water-cement ratio,sand ratio,paste-aggregate ratio,fly ash replacement proportion,and curing age as model inputs,and compressive strength as the output.The HEMNG model was compared with artificial neural networks,extreme gradient boosting,support vector machines,random forests,etc.to analyze the advantages of the HEMNG model in predicting compressive strength.The model was migrated to new data to explore its generalization ability on unknown data.Based on the well-trained HEMNG model,a sensitivity study was carried out to quantify the influence of three important features on the compressive strength of the concrete.The results are as follows.The compressive strength of concrete can be reliably and accurately predicted using the HEMNG model with five explainable features.The fitting degree between the predicted and actual values in the test set is 0.961,and the root mean square error of 2.704.This is better than other models in prediction accuracy and generalization.By migrating the HEMNG model to new data,the predicted compressive strength values are relatively consistent with the actual strength values,the maximum absolute error is only 7 MPa,indicating good robustness.Based on the sensitivity study,optimal sand ratio exists,which maximizes the compressive strength;increasing the water-binder ratio decreases the compressive strength of the concrete,and the optimal ratio decreases with the water-binder ratio becoming higher.It shows an upward trend and then a downward trend as the paste-aggregate ratio increases,and the model can quantify the influence of each parameter on compressive

关 键 词：混凝土 抗压强度 预测 集成学习 可解释特征 敏感性分析 

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