BP神经网络在再生混凝土强度预测中的应用  

Application of BP neural network in forecast to recycled concrete strength

作　　者：梁炯丰[1,2] 严兰兰 胡鹏[1,2] 杨勇[1,2] 

机构地区：[1]东华理工大学江西省新能源工艺与装备工程技术研究中心,江西南昌330013 [2]东华理工大学建筑工程学院,江西南昌330013 [3]江西杭萧钢构股份有限公司,江西南昌330013

出　　处：《混凝土》2017年第6期9-12,16,共5页Concrete

基　　金：国家自然科学基金资助项目(51368001);江西省重点研发项目(20161BBH80045);江西省青年科学基金资助项目(20161BAB211028);江西省科技支撑计划资助项目(20151BBG70012);中国博士后基金资助项目(2014M562132);中南大学博士后基金资助项目;江西省新能源工艺与装备工程技术研究中心开放基金项目(JXNE-2014-08)

摘　　要：为了提供早期预测再生粗骨料混凝土强度的有效方法,从8篇文献中收集了47组样本,借助Matlab R2015a平台,基于BP神经网络,建立了以单位体积的水、水泥、砂、碎石、再生粗骨料用量作为输入,以再生混凝土28 d棱柱体抗压强度作为输出的含单隐层的3层神经网络模型,其结构为5-21-1。对网络进行训练后的仿真结果表明,预测的最大相对误差为18.69%,预测误差小于5%的占样本总量的78.72%。预测结果表明用BP神经网络模型预测再生混凝土的强度是可行的。It aims at providing an effective method for early prediction of recycled coarse aggregate concrete strength.We collect 47 sets of data from 8 different published literature sources.With Matlab R2015a platform, based on BP neural network,we establish a three-layer neural network model with a single hidden layer which structure is 5-21-1.The input parameters are the per unit volume of water,cement, sand, gravel, recycled coarse aggregate.The output is the 28d prism compressive strength.The simulation results after training show that the predicted maximum relative error is 18.69% ,the prediction error which less than 5% accounted for 78.72% of the total sample.The prediciton results show that BP neural network has good potential to be used as a tool for predicting the compressive strength of recycled aggregate concrete.

关 键 词：BP神经网络 再生粗骨料 再生混凝土 强度 预测 

分 类 号：TU528.01[建筑科学—建筑技术科学]