基于BP神经网络的胶凝砂砾石细观参数标定研究  

Research on calibration of cemented sand-gravel microscopic parameters based on BP neural network

作　　者：李飞 梅应春 LI Fei;MEI Yingchun(Anhui Transport Consulting and Design Institute Co.,Ltd.,Hefei 230088,China;Anhui Traffic Survey and Design Institute Co.,Ltd.,Hefei 230011,China)

机构地区：[1]安徽省交通规划设计研究总院股份有限公司,安徽合肥230088 [2]安徽省交通勘察设计院有限公司,安徽合肥230011

出　　处：《江淮水利科技》2025年第1期27-30,40,共5页Jianghuai Water Resources Science and Technology

摘　　要：针对传统室内试验无法获取胶凝砂砾石细观层面参数的问题,提出基于BP神经网络的胶凝砂砾石细观参数标定研究。通过胶凝砂砾石单轴压缩试验获取其宏观参数,并利用参数敏感性分析对细观参数进行计算,选取敏感度较高的细观参数,包括骨料相、水泥砂浆相、界面过渡区相的弹性模量和抗拉强度。在此基础上,通过BP神经网络对胶凝砂砾石的上述参数进行标定研究,并对模型进行评估,其平均绝对误差为0.425,均方误差为0.434,均方根误差为0.633,平均绝对百分比误差为1.46%,决定系数为0.989,模型运行时间为84 s。评估结果表明模型能够准确捕捉胶凝砂砾石的细观参数,并展现出较强的泛化能力,可为胶凝砂砾石细观参数的研究提供参考。To address the limitations of traditional laboratory tests in obtaining meso-level parameters of cemented sand and gravel(CSG)materials,this paper proposed a BP neural network-based calibration approach for meso-parameters of CSG.Macroscopic parameters were obtained from uniaxial compression tests on CSG,and parameter sensitivity analysis was used to calculate meso-parameters,selecting those with higher sensitivity for the BP neural network-based CSG meso-parameter calibration study.The model was evaluated with a mean absolute error of 0.425,mean squared error of 0.434,root mean squared error of 0.633,mean absolute percentage error of 1.46%,and a coefficient of determination of 0.989.The model runtime was 84 seconds,indicating that it accurately captures the meso-parameters of CSG and exhibits strong generalization ability,there by providing a reliable basis for the study of CSG meso-parameters.

关 键 词：胶凝砂砾石 BP神经网络 细观参数 宏观参数 参数标定 

分 类 号：TV42[水利工程—水工结构工程]