Adaptive shield TBM selection model integrating AHP with attention mechanism and its engineering application
4 affiliations
- 1School of Mechanical Engineering, Dalian Jiaotong University
- 2School of Transportation Engineering, Dalian Jiaotong University
- 3China Railway Construction Bridge Engineering Bureau Group Co., Ltd
- 4Department of Civil and Environmental Engineering, Imperial College London
Abstract
Shield TBM selection involves multi-criteria coupling, while the traditional analytic hierarchy process (AHP) relies on subjective scoring with insufficient objectivity and consistency verification difficulties. This study proposes an intelligent selection method by enhancing AHP with the Transformer self-attention mechanism. Pre-trained language models enable semantic vectorization, and the multi-head self-attention mechanism captures inter-indicator correlations to generate semantics-informed judgment matrices; beneficial-type membership functions and the weighted average fuzzy composition operator are adopted for fuzzy comprehensive evaluation; attention weight heatmaps are generated for visual analysis. The Yangshan–Jingguan section of Xuzhou Metro Line 4 serves as a case study; two candidate TBMs are evaluated and compared with traditional AHP, alongside parameter sensitivity analysis and multi-case validation across seven projects. The judgment matrix consistency ratio is 0.0114, and the CREC172/173 composite earth pressure balance (EPB) shield is recommended with a score of 95.38. Compared with traditional AHP, the Transformer-based method exhibits an 8.79% difference in the safety criterion weight, more accurately reflecting engineering semantic features such as karst development. Parameter sensitivity analysis shows safety criterion weight variation rate remains below 2%, confirming parameter robustness. Multi-case validation reveals dynamic weight adjustment under varying engineering conditions, with selection results consistent with field-adopted schemes, preliminarily demonstrating its applicability across multiple engineering scenarios.
Related
- Hybrid Explainable Deep Ensemble Framework with self-attention mechanism for early prediction of depression severity using PHQ-9 clinical and behavioral data
- Cross-corpus consensus feature selection for language-invariant Parkinson’s disease detection: a dual-stream acoustic and spectrogram-attention framework
- A Structure-Based hERG Pose Selection Workflow Integrating Induced-Fit Docking, Shape Similarity, Interaction Fingerprints, and Molecular Dynamics
- WITHDRAWN: Combined triglyceride-glucose and frailty index (TyGFI) and risk of endometrial cancer in U.S. women aged >=45: NHANES 2011-2018 analysis integrating data engineering and machine learning with logistic modeling
- Vehicle lane change prediction with explainable soft mask attention mechanism and heterogeneous information encoder
BibTeX
@article{zhang2026adaptive,
title = {{Adaptive shield TBM selection model integrating AHP with attention mechanism and its engineering application}},
author = {Yanchao Zhang and Yundong Ma and Shuai Zheng and Pengyuan Yao and Fu Te},
journal = {Scientific Reports},
year = {2026},
month = jul,
doi = {10.1038/s41598-026-64860-7},
url = {https://doi.org/10.1038/s41598-026-64860-7},
note = {CC BY},
}
RIS
TY - JOUR
TI - Adaptive shield TBM selection model integrating AHP with attention mechanism and its engineering application
AU - Yanchao Zhang
AU - Yundong Ma
AU - Shuai Zheng
AU - Pengyuan Yao
AU - Fu Te
PY - 2026
DA - 2026/07/01
JO - Scientific Reports
DO - 10.1038/s41598-026-64860-7
UR - https://doi.org/10.1038/s41598-026-64860-7
AB - Shield TBM selection involves multi-criteria coupling, while the traditional analytic hierarchy process (AHP) relies on subjective scoring with insufficient objectivity and consistency verification difficulties. This study proposes an intelligent selection method by enhancing AHP with the Transformer self-attention mechanism. Pre-trained language models enable semantic vectorization, and the multi-head self-attention mechanism captures inter-indicator correlations to generate semantics-informed judgment matrices; beneficial-type membership functions and the weighted average fuzzy composition operator are adopted for fuzzy comprehensive evaluation; attention weight heatmaps are generated for visual analysis. The Yangshan–Jingguan section of Xuzhou Metro Line 4 serves as a case study; two candidate TBMs are evaluated and compared with traditional AHP, alongside parameter sensitivity analysis and multi-case validation across seven projects. The judgment matrix consistency ratio is 0.0114, and the CREC172/173 composite earth pressure balance (EPB) shield is recommended with a score of 95.38. Compared with traditional AHP, the Transformer-based method exhibits an 8.79% difference in the safety criterion weight, more accurately reflecting engineering semantic features such as karst development. Parameter sensitivity analysis shows safety criterion weight variation rate remains below 2%, confirming parameter robustness. Multi-case validation reveals dynamic weight adjustment under varying engineering conditions, with selection results consistent with field-adopted schemes, preliminarily demonstrating its applicability across multiple engineering scenarios.
SN - 2045-2322
ER -