上海交通大学学报(自然版) ›› 2018, Vol. 52 ›› Issue (11): 1508-1515.doi: 10.16183/j.cnki.jsjtu.2018.11.013

• 学报(中文) • 上一篇    下一篇

基于RS-MIV-ELM模型的基坑水平位移影响因素分析和预测

钟国强1,2,王浩1,张国华1,覃卫民1,王成汤1,2,熊俊峰1,2   

  1. 1. 中国科学院武汉岩土力学研究所 岩土力学与工程国家重点实验室, 武汉 430071; 2. 中国科学院大学, 北京 100049
  • 通讯作者: 王浩,男,研究员,博士生导师,电话(Tel.):027-87197320;E-mail: hwang@whrsm.ac.cn.
  • 作者简介:钟国强(1990-),男,山东省曲阜市人,博士生,目前主要从事岩土工程自动化监测及变形预测和风险评估研究.
  • 基金资助:
    国家自然科学基金资助项目(41472288, 41172287, 51579235)

Analysis and Prediction of Factors Affecting Horizontal Displacement of Foundation Pit Based on RS-MIV-ELM Model

ZHONG Guoqiang,WANG Hao,ZHANG Guohua,QIN Weimin WANG Chengtang,XIONG Junfeng   

  1. 1. State Key Laboratory of Geomechanics and Geotechnical Engineering, Institute of Rock and Soil Mechanics, Chinese Academy of Sciences, Wuhan 430071, China; 2. University of Chinese Academy of Sciences, Beijing 100049, China

摘要: 为了预测基坑的测斜最大水平位移及深度,提出了基于粗糙集(RS)属性约简、平均影响值(MIV)和极限学习机(ELM)的组合模型RS-MIV-ELM.在系统分析、量化变形影响因素的基础上,利用RS属性约简算法和基于ELM的MIV算法(ELM-MIV)分别去除影响因素集中的冗余因素和相关性极小的因素,以简化模型输入变量;采用简化的影响因素集训练ELM模型,并用ELM模型对其他测点位移进行预测.验证结果表明,RS-MIV-ELM模型的训练速度、预测精度和泛化能力均比全因素ELM模型和基于最简集的BP神经网络模型RS-MIV-BP具有较大的提高,其均方根误差和平均相对误差仅为全因素ELM模型和RS-MIV-BP模型的1/2~2/3.

关键词: 基坑, 水平位移预测, 极限学习机, 属性约简, 影响因素筛选

Abstract: In order to predict the maximum horizontal displacement and depth of the inclinometer in the foundation pit, an RS-MIV-ELM model based on rough set attribute reduction, mean impact value and extreme learning machine was proposed. The model was based on systematic analysis and quantification of deformation factors, the attribute reduction algorithm of rough set (RS) and the mean impact value based on the extreme learning machine algorithm (ELM-MIV) were used to remove the redundant factors and small correlation factors respectively. Then, the extreme learning machine (ELM) model was trained by the simplified influence factor set, and the model was used to predict the displacement of other measuring points. Experimental results show that the training speed, prediction accuracy and generalization ability of the proposed model are better than those of the all factors ELM model and the BP neural network model based on the simplest set. The root mean square error or average relative error of RS-MIV-ELM model is about 1/2~2/3 of the two contrast models.

Key words: foundation pit, horizontal displacement prediction, extreme learning machine (ELM), attri-bute reduction, screening influencing factors

中图分类号: