摘要
为了解决将粗骨料引入超高性能混凝土(UHPC)后混凝土抗压强度预测和配合比设计复杂的问题,本工作提出了一种基于Stacking集成模型融合的粗骨料超高性能混凝土(CA-UHPC)抗压强度预测方法。首先,对通过文献收集到的175组数据进行预处理,随后分别基于随机森林(RF)、支持向量机(SVM)和极端梯度提升树(XGBoost)算法进行贝叶斯超参数优化以及模型的训练与评估。采用均方根误差(RMSE)、平均绝对误差(MAE)、以及决定系数(R^(2))来对比分析不同模型的预测精度。在此基础上建立了基于Stacking融合多种算法的CA-UHPC抗压强度预测模型,并对其进行了泛化性能验证以及可解释性和参数分析。结果表明,与单一机器学习算法相比,基于Stacking集成多种机器学习方法的模型在预测精度上有所提升,且模型与工程实践经验吻合较好,具有较高的合理度和可靠性。此外,通过模型参数分析可知,在进行CA-UHPC制备时,最佳的纤维掺量在2.0%~2.5%之间,最佳的骨料掺量为胶凝材料总量的0.2~0.4倍,增加硅粉用量可有效提高CA-UHPC的抗压强度。本工作的研究成果可为CA-UHPC的配合比设计提供理论支撑。
To solve the problem of compressive strength prediction and proportion design after introducing coarse aggregates into ultra high performance concrete(UHPC),the paper proposes a novel method for predicting the compressive strength of coarse aggregate ultra-high performance concrete(CA-UHPC)based on the Stacking integrated model fusion.Initially,175 datasets sourced from the literature were preprocessed.Subsequently,Bayesian hyperparameter optimization,model training and evaluation were conducted based on random forest(RF),support vector machine(SVM),and extreme gradient boosting tree(XGBoost)algorithms.Root mean square error(RMSE),mean absolute error(MAE),and coefficient of determination(R^(2))were employed to analyze the prediction accuracies of these models.On this basis,a CA-UHPC compressive strength prediction model integrating multiple algorithms by the stacking method was proposed.The results demonstrate that the stacking model outperforms individual algorithms in prediction accuracy.Moreover,the interpretability analysis of the model aligns well with engineering practical experience,thus rendering it highly reasonable and reliable.In addition,the parameters analysis shows that the optimum fiber content is between 2.0%and 2.5%,and the optimum aggregate content is 0.2-0.4 times of the cementitious material content for CA-UHPC.Besides,increasing the silica fume content can effectively improve the compressive strength of CA-UHPC.This model may have potential for the proportion design of CA-UHPC.
作者
唐博文
梁梓豪
丁平祥
范志宏
TANG Bowen;LIANG Zihao;DING Pingxiang;FAN Zhihong(CCCC Fourth Harbor Engineering Institute Co.,Ltd.,Guangzhou 510230,China;Key Laboratory of Construction Materials,CCCC,Guangzhou 510230,China;Key Laboratory of Harbor&Marine Structure Durability Technology,Ministry of Transport,Guangzhou 510230,China)
出处
《材料导报》
EI
CAS
CSCD
北大核心
2024年第S02期469-474,共6页
Materials Reports
基金
国家重点研发计划(2022YFB2603000)。
