First Author: WU Hao
Corresponding Author: YUE Qiang
Journal: Journal of Basic Science and Engineering

Abstract

Understanding deformation of tunnel surrounding rock is a prerequisite for recognizing the dynamic response of surrounding rock and support structures, as well as for elucidating the mechanism of spatial deformation. Establishing reliable deformation classification criteria is therefore essential for stability assessment and effective design of supporting structures.

This paper proposes a deformation classification prediction model for tunnel surrounding rock based on an improved one-dimensional convolutional neural network (1DCNN) integrated with a support vector weighting (SVW) module. The model considers major influencing factors and feature types of tunnel deformation, including support strength, surrounding rock lithology, and burial depth. Using these indicators, a classification framework for four deformation grades was constructed.

A dataset of 159 groups of surrounding rock deformation cases from domestic tunnel projects was compiled. Full normalization and feature weighting were applied before feeding the data into the improved CNN. A fully connected layer and Softmax classifier were then employed for grade classification. The improved 1DCNN automatically captured hidden feature patterns of tunnel deformation. Training strategies included learning rate decay and Dropout regularization, ensuring robustness against overfitting.

Results show that the proposed 1DCNN+SVW model achieved an accuracy of 90.8%, outperforming traditional machine learning approaches in precision and generalization ability. Comparative analysis with other methods further confirmed its higher accuracy and stability. Application to the Erlang Mountain Tunnel project demonstrated that predicted deformation grades were consistent with field observations, thereby validating the model’s accuracy and practical applicability.

This study enhances the theoretical framework and reliability of tunnel deformation prediction, providing important technical support for tunnel engineering.

Keywords: Tunnel engineering; Deformation; Prediction; Deep learning; Improved one-dimensional convolutional neural network (1DCNN)