中低强度等级混凝土抗压强度的BP神经网络模型预测研究  

Prediction of compressive strength of medium and low strength grade concrete by BP neural network model

作　　者：赵明亮 水中和[3] 周华新 崔巩 ZHAO Mingliang;SHUI Zhonghe;ZHOU Huaxin;CUI Gong(State Key Laboratory of High Performance Civil Engineering Materials,Nanjing 210008,China;Jiangsu Sobute New Materials Co.,Ltd.,Nanjing 211103,China;Stage Key Laboratory of Silicate Materials for Architectures,Wuhan University of Technology,Wuhan 430070,China)

机构地区：[1]高性能土木工程材料国家重点实验室,江苏南京210008 [2]江苏苏博特新材料股份有限公司,江苏南京211103 [3]武汉理工大学硅酸盐建筑材料国家重点实验室,湖北武汉430070

出　　处：《混凝土》2021年第3期35-38,共4页Concrete

基　　金：国家重点研发计划(2017YFB0310905-06)。

摘　　要：针对混凝土公司经常面临的原材料波动和配合比调整,造成混凝土抗压强度难以稳定控制的问题,通过建立BP神经网络模型预测混凝土7、28 d抗压强度,以胶凝材料总量、矿粉比例、粉煤灰比例、硅灰比例、水胶比、V_(砂)/V_(浆体)、扩展度、V_(粗骨料)/V砂浆作为输入层,7、28 d抗压强度作为输出层。结果表明,BP神经网络具有很强的非线性映射能力,模型预测计算的7、28 d强度值和实际值的相关系数达到了0.98031、0.95564,预测精度高。Aiming at the problem that the concrete compressive strength is difficult to control steadily due to the fluctuation of raw materials and the adjustment of mix proportion,in this paper,the compressive strength of concrete for 7 d and 28 d are predicted by BP neural network model.The total amount of cementitious materials,the proportion of mineral powder,the proportion of fly ash,the proportion of silica fume,water-cement ratio,V-sand/V-mortar,expansion degree and V-coarse bone/V-mortar are selected as input layers.The compressive strength of 7 d and 28 d are used as the output layer.The results show that the BP neural network has strong non-linear mapping ability.The correlation coefficients of the expected and actual values of 7 d and 28 d predicted by the model reach 0.98031 and0.95564,which means high prediction accuracy.

关 键 词：BP神经网络 抗压强度 配合比 波动 

分 类 号：TU528.01[建筑科学—建筑技术科学]