Scholarlink

<p>Accurate prediction of ground settlement is essential for risk management in urban twin-tunnel projects, where complex geological heterogeneity and tunnel interaction effects pose significant challenges. This study develops an advanced machine learning framework incorporating Bayesian-optimized Support Vector Regression (SVR), Random Forest (RF), Deep Neural Networks (DNN), and Back-Propagation Neural Networks (BPNN), with key methodological innovations including a novel three-dimensional soil layer parameter calibration method and explicit quantification of twin-tunnel interaction effects using monitoring data from the Dongguan Metro Line 1 project. The optimized DNN model attained exceptional prediction accuracy for the leading tunnel (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^{2}=0.92\)</EquationSource> </InlineEquation>), while incorporating settlement data from the leading tunnel significantly improved following tunnel predictions (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(R^{2}=0.80\)</EquationSource> </InlineEquation>), demonstrating the critical importance of capturing geological heterogeneity and tunnel interactions. This framework offers a robust alternative to conventional methods, with direct implications for enhancing construction safety, optimizing monitoring strategies, and reducing risk in metro tunneling projects.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine Learning-Based Prediction of Ground Settlement in Twin-Tunnel Excavation: A Case Study from Dongguan Metro Line 1

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • Ziwei Xiao,
  • Xiaowen Zhou,
  • Changhui Zhang,
  • Yunchang Xiao,
  • Chenliang Wei,
  • Jian Chen

摘要

Accurate prediction of ground settlement is essential for risk management in urban twin-tunnel projects, where complex geological heterogeneity and tunnel interaction effects pose significant challenges. This study develops an advanced machine learning framework incorporating Bayesian-optimized Support Vector Regression (SVR), Random Forest (RF), Deep Neural Networks (DNN), and Back-Propagation Neural Networks (BPNN), with key methodological innovations including a novel three-dimensional soil layer parameter calibration method and explicit quantification of twin-tunnel interaction effects using monitoring data from the Dongguan Metro Line 1 project. The optimized DNN model attained exceptional prediction accuracy for the leading tunnel ( \(R^{2}=0.92\) ), while incorporating settlement data from the leading tunnel significantly improved following tunnel predictions ( \(R^{2}=0.80\) ), demonstrating the critical importance of capturing geological heterogeneity and tunnel interactions. This framework offers a robust alternative to conventional methods, with direct implications for enhancing construction safety, optimizing monitoring strategies, and reducing risk in metro tunneling projects.