基于MIC-LSTM的盾构施工地表变形动态预测

李增良

doi:10.3973/j.issn.2096-4498.2021.02.005

隧道建设 >

2021 , Vol. 41 >Issue 2: 199 - 205

DOI: https://doi.org/10.3973/j.issn.2096-4498.2021.02.005

研究与探索

基于MIC-LSTM的盾构施工地表变形动态预测

李增良

展开

（中铁二十局集团有限公司，陕西西安 710016）

李增良（1976—），男，湖南平江人，2000年毕业于中南大学，机械制造及自动化专业，本科，高级工程师，主要从事物资设备及盾构施工技术与管理工作。E-mail： 852499699@qq.com。

网络出版日期: 2021-03-05

基金资助

中国博士后科学基金资助项目(2020M673525)

收起

Dynamic Prediction of Surface Deformation Induced by Shield Tunneling Based on Maximal Information CoefficientLong ShortTerm Memory

LI Zengliang

Expand

(China Railway 20th Bureau Group Co., Ltd., Xi′an 710016, Shaanxi, China)

Online published: 2021-03-05

Fold

摘要

在盾构施工过程中准确预测施工引起的地表变形，对于保障盾构施工的顺利掘进具有重要意义。基于此，提出盾构施工地表变形MIC-LSTM动态预测模型。首先，确定影响地表变形的主要因素，并采用最大信息系数法（MIC，maximal information coefficient）确定各个影响因素和地表变形之间的相关程度，进而对各个影响因素赋权；其次，将赋权后的各个影响因素和盾构中心处过去最近10个监测时刻的地表变形数据作为输入变量、未来3个监测时刻的变形数据作为输出变量来构建长短期记忆（LSTM，long shortterm memory）神经网络动态预测模型；最后，为验证所构建的MIC-LSTM动态预测模型的实用性，依托昆明地铁5号线盾构施工项目，将预测结果与LSTM、RNN（recurrent neural network）以及BP（back propagation）神经网络的预测结果进行对比。研究结果表明：所构建的盾构施工地表变形动态预测模型具有较高的预测精度。

关键词： 地铁隧道; 盾构; 地表变形; 动态预测; 最大信息系数; 长短期记忆神经网络

本文引用格式

李增良 . 基于MIC-LSTM的盾构施工地表变形动态预测[J]. 隧道建设, 2021 , 41(2) : 199 -205 . DOI: 10.3973/j.issn.2096-4498.2021.02.005

Abstract

Accurate prediction of surface deformation caused by shield tunneling is very important in ensuring successful shield tunneling. Accordingly, a dynamic prediction model for surface deformation induced by shield tunneling based on maximal information coefficient (MIC)long shortterm memory (LSTM) is proposed. First, the main factors that affect the surface deformation are determined, and the MIC method is used to determine the degree of correlation between each influencing factor and surface deformation. Then, an LSTM neural network dynamic prediction model is established by considering the weighted influencing factors and recent 10 surface deformation data at the center of the shield as input variables. The next three deformation data are considered as output variables. Finally, to verify the practicability of the constructed MICLSTM dynamic prediction model, its prediction results for a shield project in Kunming Metro Line No. 5 are compared with those of the LSTM, recurrent neural network, and backpropagation neural network. The results show that the constructed dynamic prediction model for shield tunnelinginduced surface deformation demonstrates high prediction accuracy.

Key words： metro tunnel; shield; surface deformation; dynamic prediction; maximal information coefficient; long shortterm memory neural network

Options

文章导航

〈

〉

模态框（Modal）标题

摘要

本文引用格式

Abstract