Scholarlink

<p>The advance rate (AR) of a tunnel boring machine (TBM) governs construction scheduling, cost control, and overall project efficiency; thus, its accurate prediction is essential for effective resource allocation and mitigation of delays arising from geological and operational variability. This study develops an optimal soft-computing framework by comparatively evaluating support vector regression (SVR), feedforward neural networks (FFNN), gene expression programming (GEP), gated recurrent units (GRU), long short-term memory (LSTM), and bidirectional LSTM (BiLSTM) models. A dataset comprising 1,197 TBM operational records was utilized, incorporating cutterhead rotation speed (CRS), mean thrust (F/A), mean cutterhead torque (T/D³), upper earth pressure (UEP), lower earth pressure (LEP), and torque penetration index (TPI). Multicollinearity among predictors was quantified using the variance inflation factor (VIF), while feature sensitivity was assessed via the cosine amplitude method. Model performance was evaluated using eight statistical indices, three reliability measures, regression error characteristic (REC) curves, generalizability assessment, and the Wilcoxon signed-rank test. Comparative analysis demonstrated the superior predictive capability of the BiLSTM model, achieving accuracy exceeding 98.60% across training, testing, and validation phases. Reliability indices confirmed its robustness. Nevertheless, curve-fitting analysis indicated mild overfitting during testing (2.49) and validation (1.98), examined through the interaction between feature multicollinearity and sensitivity.</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

Operational Parameter–Based Prediction of Shield TBM Advance Rate Using Explainable Computational Intelligence

  • 期刊论文

  • 发表日期：

    2026年3月20日

查看全文
  • Jitendra Khatti,
  • Pijush Samui,
  • Swapnil Mishra

摘要

The advance rate (AR) of a tunnel boring machine (TBM) governs construction scheduling, cost control, and overall project efficiency; thus, its accurate prediction is essential for effective resource allocation and mitigation of delays arising from geological and operational variability. This study develops an optimal soft-computing framework by comparatively evaluating support vector regression (SVR), feedforward neural networks (FFNN), gene expression programming (GEP), gated recurrent units (GRU), long short-term memory (LSTM), and bidirectional LSTM (BiLSTM) models. A dataset comprising 1,197 TBM operational records was utilized, incorporating cutterhead rotation speed (CRS), mean thrust (F/A), mean cutterhead torque (T/D³), upper earth pressure (UEP), lower earth pressure (LEP), and torque penetration index (TPI). Multicollinearity among predictors was quantified using the variance inflation factor (VIF), while feature sensitivity was assessed via the cosine amplitude method. Model performance was evaluated using eight statistical indices, three reliability measures, regression error characteristic (REC) curves, generalizability assessment, and the Wilcoxon signed-rank test. Comparative analysis demonstrated the superior predictive capability of the BiLSTM model, achieving accuracy exceeding 98.60% across training, testing, and validation phases. Reliability indices confirmed its robustness. Nevertheless, curve-fitting analysis indicated mild overfitting during testing (2.49) and validation (1.98), examined through the interaction between feature multicollinearity and sensitivity.