土工基础

土工基础 ›› 2023, Vol. 37 ›› Issue (2): 231-235.

• 专题论述 • 上一篇    下一篇

基于广义回归神经网络的压缩模量预测研究

欧阳磊1，邓希萍2，何利军1，刘家印1 ，陈 鹏1   

  1. (1.南昌航空大学土木建筑学院，南昌 330063；2.萍乡市规划勘察设计院，江西萍乡 337000)
  • 收稿日期:2021-08-16 修回日期:2021-08-22 出版日期:2023-04-30 发布日期:2023-04-20
  • 通讯作者: 何利军（1977-），男，博士，讲师，主要研究方向为岩土工程。
  • 作者简介:欧阳磊(1996-)，男，在读硕士研究生，主要研究方向为工程数值模拟与计算。
  • 基金资助:
    广西防灾减灾与工程安全重点实验室开放课题(2016ZDK013)；广西大学工程防灾与结构安全教育部重点实验室开放课题(2016ZDK013)

Compressive Modulus Prediction Using Generalized Regression Neural Network

OUYANG Lei1, DENG Xiping2, HE Lijun1, LIU Jiayin1， CHEN Peng1   

  1. (1.School of Civil Engineering, Nanchang Hangkong University, Nanchang 330063；
    2.Pingxiang City Planning Survey and Design Institute, Pingxiang 337000)
  • Received:2021-08-16 Revised:2021-08-22 Online:2023-04-30 Published:2023-04-20

摘要： 压缩变形是引起沉降的重要影响因素，其值过大会造成重大地质灾害，压缩模量是其重要指标，预测压缩模量对于预防地质灾害具有重要意义。基于广义回归神经网络的基本原理，以常规土物性指标作为输入向量，以压缩模量（ES1-2）为输出向量，网络输出结果的最大相对误差和最小相对误差分别为 10.73%和0%、均方误差为0.683 1、与真实值吻合度很高，故模型可运用到压缩模量的预测。比较不同光滑因子值下的均方误差，0.9时的均方误差较其他取值小，故在参数设置过程中应该正确选取光滑因子的值。为证明基于广义回归神经网络的预测模型的性能，以均方误差为评价标准与其他算法比较，广义回归神经网络预测模型的均方误差小于其他算法，说明基于广义回归神经网络预测模型性能优于其他算法所构建的预测模型。


关键词: 预测, 压缩模量, 广义回归神经网络, 土物性指标, 光滑因子

Abstract: The compression deformation is an important factor that affecting the magnitude of the settlement. If its value is too large, it will induce the major geological hazards. The compression modulus is an important index. Predicting the magnitude of the compression modulus is of great significance for preventing geological hazards. Throughout basic principles of the generalized regression neural network, with the conventional soil property indicators as the input vector and the compressive modulus (ES1-2) as the output vector, the maximum and minimum relative errors of the network output results are 10.73% and 0%, respectively. The mean square error is 0.6831, which is in good agreement with the true value, so the model can be used to predict the compressive modulus. Comparing the mean square error under different smoothing factor values, the mean square error at 0.9 is smaller than other values, so the smoothing factor value should be selected correctly during the parameter setting process. To prove the performance of the prediction model throughout the generalized regression neural network, compared with other algorithms based on the mean square error, the mean square error of the generalized regression neural network prediction model is less than that of other algorithms, indicating that the prediction model throughout the generalized regression neural network has superior performance based on prediction models constructed by other algorithms.

Key words: Prediction, Compression Modulus, Generalized Regression Neural Network, Soil Physical Property Index, Smoothing Factor

中图分类号: