Prediction model of tunneling speed for double-shield TBM based on CNN-Transformer-ARG

An intelligent prediction model integrating convolutional neural networks (CNN), Transformer, and an adaptive residual gating (ARG) mechanism was proposed, so as to accurately predict the tunneling speed of double-shield tunnel boring machines (TBMs). The model utilized a dual-layer CNN to extract local features of tunneling parameters from different perspectives, employed a Transformer to capture global features among these parameters, and incorporated the ARG mechanism to dynamically weight the extracted local and global features. Based on the monitoring data from historical tunneling segments, the mean, maximum, and minimum tunneling speeds of future tunneling segments were predicted. The model was validated using 927 sets of tunneling data extracted from a mountain rail transit project in Sichuan Province. The results indicate that the model achieves a mean square error of 0.07, a mean absolute error of 0.21, a root mean square error of 0.26, and a coefficient of determination of 0.86, all outperforming the other three comparative models. By assigning weights to the multi-source features extracted by the model, key information is emphasized, thereby improving prediction accuracy. This validates the effectiveness of the ARG mechanism in multi-source models and provides insights for similar structured models.