Multi-step prediction model enhanced by adaptive denoising and encoder-decoder for shield machine cutterhead torque in complex conditions

•A multi-step prediction model for cutterhead torque is proposed.•Adaptive denoising and encoder-decoder are used to improve the model performance.•The model was validated using site data from the Heyan Road tunnel project.•This model has promising performance in complex geological and working condi...

Full description

Saved in:
Bibliographic Details
Published in:Tunnelling and underground space technology Vol. 158; p. 106398
Main Authors: Xu, Deming, Wang, Yuan, Huang, Jingqi, Xu, Shujun, Zhou, Kun
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.04.2025
Subjects:
Cutterhead torque
Encoder-decoder model
Long short-term memory
Multi-step prediction
Shield machine
Shield machine
Long short-term memory
Cutterhead torque
Encoder-decoder model
Multi-step prediction
ISSN:0886-7798
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•A multi-step prediction model for cutterhead torque is proposed.•Adaptive denoising and encoder-decoder are used to improve the model performance.•The model was validated using site data from the Heyan Road tunnel project.•This model has promising performance in complex geological and working conditions. Cutterhead torque reflects the obstruction extent of geological environment on the shield machine, and its prediction can assist operators to adjust control parameters to improve construction efficiency and avoid machine jamming. However, tunneling in complex geological or working conditions often results in high cutterhead torque fluctuations and noise, which seriously affects the accuracy of torque prediction. This study proposes a multi-step prediction model for cutterhead torque enhanced by adaptive denoising and encoder-decoder. In this model, a novel adaptive denoising method for cutterhead torque is employed to improve prediction accuracy under complex conditions. Moreover, by introducing encoder-decoder method, the processing capability for multi-time dimensional data and multi-step prediction performance of LSTM neural networks are further improved. The effectiveness of proposed model is verified through an application to the Heyan Road River Crossing project. The results of this study can assist operators in achieving precise adjustment of control parameters under complex conditions.
ISSN:0886-7798
DOI:10.1016/j.tust.2025.106398