Prediction of Tunnel Squeezing Classification Based on Improved One-Dimensional Convolutional Neural Network

The change of tunnel squeezing is the premise to understand the dynamic role of surrounding rock and support structure and its spatial and temporal evolution mechanism.The accurate prediction of tunnel squeezing classification is an important basis for assessing the stability of rock and the effectiveness of support structure.A prediction model of tunnel squeezing classification with 1DCNN and SVW fused deep network is proposed.According to the main influencing factors and characteristic types of tunnel squeezing,the framework for identifying the tunnel squeezing intensity is established by selecting five evaluation indices,i.e.,strength stress ratio,tunnel burial depth,rock quality index,tunnel equivalent diameter and support stiffness.159 groups of typical tunnel engineering case data are collected and adopted as the sample data.The global mean pooling layer and SVW are used to improve the fully connected and Softmax layers in traditional CNN for rank classification.The implicit typical features of tunnel squeezing are automatically extracted using the improved 1DCNN.Deep learning training techniques are applied to prevent model overfitting,such as Dropout regularization and decaying learning rate.The comparison with the results of other methods proves that this method has better accuracy and robustness.The model is completely data-driven to achieve deep,complex and subtle relationship learning with limited data sets.The model is applied to predict the convergence deformation classification of the surrounding rock of the Doxiongla highway tunnel,and the prediction results are consistent with the actual conditions in the field,which further verified the accuracy and applicability of the model in this paper.The results can be used to improve the theoretical level and reliability of tunnel squeezing prediction and provide a good reference for similar projects.