PDF全文阅读

李风增^1 ,2

( 1. 郑州市公路事业发展中心 ，河南 郑州 450015 ;

2. 河南省交通科学技术研究院有限公司 ，河南 郑州 450015)

摘   要： 28 d 抗压强度是混凝土应用过程中重要的强度指标 。本文采用人工神经元网络模型( ANN) 对加入了高炉矿渣和粉煤灰的混凝土 28 d 抗压强度进行了预测，研究了ANN 隐含层数、传递函数类型和优化算法对抗压强度预测结果的影响，发现采用单一隐含层的 7-23- 1 网络结构、隐含层和输出层均采用purelin 传递函数、trainlm 算法进行计算，建立的 ANN 结构最佳，仅需 0. 29 s、经过 3 个历元，预测值的 MSE 就能降到 0. 004 左右，训练、验证和测试阶段 ANN 的预测结果与试验结果回归曲线斜率均大于 97% , 两者十分吻合。

关键词：混凝土； 抗压强度；人工神经元网络； 预测

中图分类号： TV431      

文献标志码： A      

文章编号： 1005-8249 (2023) 03-0020-08

DOI : 10. 19860/j.cnki.issn1005-8249.2023.03.020

LI Fengzeng^1 , 2

(1. Zhengzhou Highway Development Centre , Zhengzhou 450015 , China;

2. Henan Transportation Research Institute Co. ，Ltd. , Zhengzhou 450015 , China)

Abstract: The 28 d compressive strength is an important index in concrete application. In this study, an artificial neural network model (ANN) was used to predict the compressive strength of concrete added with blast furnace slag and fly ash. The effects of number of hidden layer, transfer function types and optimization algorithms on ANN output were researched. The results showed that the 7-23-1 network structure with one hidden layer, purelin transfer function in the hidden and output layer, and trainlm algorithm established the optimal network. The established ANN only takes 0.29 seconds to reduce the MSE of prediction values to about 0.004 with 3 epoches. The slopes of the regression curves between the ANN outputs and the experimental results in the training, verification, and testing stages were greater than 97%, which indicated the consistence between prediction values and experimental ones.

Keywords : concrete ; compressive strength; ANN ; prediction

∗基金项目： 河北省教育厅科研发展基金项目 (2020J0148)。

作者简介： 李风增 ( 1981 —) , 男，硕士，高级工程师 ，国家注册一级建造师，国家注册城市规划师。主要从事道路桥梁工程研究。

收稿日期： 2021 -08-21