TBM Load Prediction Based on CX-SAM-GRU Model

Zhen-liang ZHOU , Man WANG , Yi-mei YUE , Hao-kai LI

Science Technology and Engineering ›› 2026, Vol. 26 ›› Issue (31) : 13665 -13677.

PDF
Science Technology and Engineering ›› 2026, Vol. 26 ›› Issue (31) : 13665-13677 DOI: 10.12404/j.issn.1671-1815.2600094
research-article
TBM Load Prediction Based on CX-SAM-GRU Model
Author information +
History +
PDF

Abstract

In order to mitigate the risk of equipment damage or jamming disasters caused by overloading when a tunnel boring machine (TBM) excavates in adverse geological conditions, which can negatively affect the project schedule, data from the Tianshan Shengli Tunnel project was used to investigate TBM load prediction. First, outliers in the cyclic segment data were removed utilizing the interquartile range (IQR) criterion, and the data was subsequently normalized. Then, feature parameters with strong causal relationships to two load parameters, cutterhead torque and thrust, were selected as input features through the independent component analysis-based linear non-Gaussian acyclic model (ICA-LiNGAM) method. Finally, a gated recurrent unit (GRU) prediction model incorporating a self-attention mechanism (SAM) was constructed, and the internal decision-making mechanism of the model was analyzed utilizing the Shapley additive explanations (SHAP) method. The results show that the prediction accuracy and computational efficiency of the model can be improved through feature screening and the introduction of a self-attention mechanism. A goodness-of-fit (R2) of 0.899 1 and 0.955 9 for the TBM load cutterhead torque and thrust, respectively, is achieved by the CX-SAM-GRU prediction model, and good generalization ability is exhibited under different engineering geological conditions. It is indicated by the importance analysis that the prediction results of cutterhead torque are largely influenced by thrust and advance rate, whereas the prediction results of thrust are largely influenced by cutterhead water spray flow, cutterhead rotation speed, and propel pump pressure. The robustness of this method for multi-step prediction results of thrust is superior to that of cutterhead torque, and the cutterhead torque prediction is generally more sensitive to closer time steps. It is concluded that a theoretical reference for TBM load prediction in practical engineering can be provided, and support for safe on-site construction can be offered.

Graphical abstract

Keywords

TBM load / time series prediction / model interpretation / causal feature screening / self-attention mechanism

Cite this article

Download citation ▾
Zhen-liang ZHOU, Man WANG, Yi-mei YUE, Hao-kai LI. TBM Load Prediction Based on CX-SAM-GRU Model. Science Technology and Engineering, 2026, 26(31): 13665-13677 DOI:10.12404/j.issn.1671-1815.2600094

登录浏览全文

4963

注册一个新账户 忘记密码

References

Funding

PDF (29041KB)

0

Accesses

0

Citation

Detail

Sections
Recommended

/

〈 〉