Scholarlink

<p>Accurate prediction of maximum tunneling-induced ground surface settlement (<i>S</i><sub><i>max</i></sub>) was investigated using numerical simulations and machine learning techniques for cohesive soils under Greenfield conditions. A database of 900 numerical simulations was generated by systematically varying three governing dimensionless parameters, including tunnel depth-to-diameter ratio (<i>C/D</i>), stiffness ratio (<i>E/S</i><sub><i>u</i></sub>), and strength ratio (<i>γD/S</i><sub><i>u</i></sub>). Two predictive models were developed using gene expression programming (GEP) and a hybrid sine–cosine optimized artificial neural network (SCA-ANN). Model evaluation demonstrated that SCA-ANN achieved superior predictive accuracy with <i>R</i><sup><i>2</i></sup> = 0.984 and <i>RMSE</i> = 1.689, compared with GEP yielding <i>R</i><sup><i>2</i></sup> = 0.942 and <i>RMSE</i> = 3.366. The mean ratio of measured-to-predicted values was closer to unity for SCA-ANN (<i>λ</i> = 0.986) than for GEP (<i>λ</i> = 1.581). Taylor diagram analysis confirmed the improved agreement and reduced variability of SCA-ANN predictions. Model uncertainty and reliability were assessed using Monte Carlo simulations, showing that over 90% of SCA-ANN predictions fell within ± 12% error. Feature importance and physical consistency were evaluated using SHapley Additive exPlanations and Fourier Amplitude Sensitivity Test, identifying <i>C/D</i> as the dominant parameter. The proposed models provided accurate and reliable tools for predicting tunneling-induced settlements in geotechnical design.</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

Numerical and Machine Learning-Driven Approaches for Predicting Tunneling-Induced Surface Settlements in Cohesive Soils under Greenfield Conditions

  • 期刊论文

  • 发表日期：

    2026年4月23日

查看全文
  • Arif Khan,
  • Meisam Mahboubi Niazmandi,
  • Sajjad Gholipour,
  • Vahid Hatami Dezdarani,
  • Roya Sedaeesoula

摘要

Accurate prediction of maximum tunneling-induced ground surface settlement (Smax) was investigated using numerical simulations and machine learning techniques for cohesive soils under Greenfield conditions. A database of 900 numerical simulations was generated by systematically varying three governing dimensionless parameters, including tunnel depth-to-diameter ratio (C/D), stiffness ratio (E/Su), and strength ratio (γD/Su). Two predictive models were developed using gene expression programming (GEP) and a hybrid sine–cosine optimized artificial neural network (SCA-ANN). Model evaluation demonstrated that SCA-ANN achieved superior predictive accuracy with R2 = 0.984 and RMSE = 1.689, compared with GEP yielding R2 = 0.942 and RMSE = 3.366. The mean ratio of measured-to-predicted values was closer to unity for SCA-ANN (λ = 0.986) than for GEP (λ = 1.581). Taylor diagram analysis confirmed the improved agreement and reduced variability of SCA-ANN predictions. Model uncertainty and reliability were assessed using Monte Carlo simulations, showing that over 90% of SCA-ANN predictions fell within ± 12% error. Feature importance and physical consistency were evaluated using SHapley Additive exPlanations and Fourier Amplitude Sensitivity Test, identifying C/D as the dominant parameter. The proposed models provided accurate and reliable tools for predicting tunneling-induced settlements in geotechnical design.