基于均匀设计及BP神经网络的大体积混凝土热学参数反分析
DOI:
作者:
作者单位:

长沙理工大学土木工程学院

作者简介:

通讯作者:

中图分类号:

基金项目:

国家重点基础研究发展计划(973计划)


Inverse analysis on thermal parameters of mass concrete based on uniform design and BP neural network
Author:
Affiliation:

School of Civil Engineering, Changsha University of Science & Technology, Changsha

Fund Project:

National Program on Key Basic Research Project of China

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    为了解决桥梁工程大体积混凝土热学参数失真的问题,提出了一种基于均匀设计理论与BP神经网络的大体积混凝土热学参数反分析方法。首先基于Matlab建立BP神经网络模型,采用tan-sigmod型传递函数,然后利用均匀设计方法确定热学参数样本,通过有限元计算得到混凝土特征点温度计算样本,训练其二者之间的非线性关系。最后将实测温度值输入训练好的神经网络,即可得到各热学参数的反演值。在太洪长江大桥散索鞍支墩承台大体积混凝土施工中反演了导热系数、反应速率及绝热温升,基于反演值的温度计算值与现场实测值吻合较好,指导了温控施工,确保了工程安全。工程实践表明:将均匀设计和BP神经网络相结合的反分析方法,减少了样本数据,在BP神经网络的训练阶段,采用附加动量法明显缩短网络训练时间,提高了效率。

    Abstract:

    In order to solve the problem of distortion of thermal parameters of mass concrete in bridge engineering, an inverse analysis method of thermal parameters of mass concrete based on uniform design theory and BP neural network was proposed. Firstly, BP neural network model is established based on Matlab, use a tan-sigmod transfer function. Then the thermal parameter samples are designed by uniform design method. The finite element software is used to analyze the temperature samples of concrete feature points and train the complex nonlinear relationship between them. Finally, the measured temperature values are input into the trained neural network, and the inversion values of the thermal parameters are obtained. The thermal conductivity, reaction rate and adiabatic temperature rise were inverted during the mass concrete construction of the scaffold saddle support pier of the Taihong Yangtze River Bridge. The calculated temperature based on the inversion value is in good agreement with the actual measured value, which guides the temperature control construction and ensures project safety. Engineering practice shows that the inverse analysis method combining uniform design and BP neural network reduces the sample data. In the training phase of BP neural network, the additional momentum method is used to significantly shorten the network training time and improve the efficiency.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2019-12-24
  • 最后修改日期:2020-02-23
  • 录用日期:2020-03-15
  • 在线发布日期:
  • 出版日期: