Abstract: To address the limitations of single models in feature extraction and prediction accuracy for tunnel boring machine (TBM) advance rate prediction, this study proposes an intelligent prediction model based on a TCN-BiGRU-Transformer multi-module fusion architecture. By integrating the multi-scale feature extraction capability of temporal convolutional networks (TCN), the long-term sequential dependency modeling advantage of bidirectional gated recurrent units (BiGRU), and the self-attention mechanism of Transformer, the model achieves precise prediction of the advance rate for double-shield TBMs. Using construction data from the Eping Water Diversion TBM tunnel project, input features including total thrust, cutterhead torque, cutterhead rotation speed, and advance rate during the rising phase of a complete TBM excavation cycle were selected. A dataset containing 3,308 samples was constructed using STL decomposition for data augmentation to predict the stable-phase advance rate. The results demonstrate that the TCN-BiGRU-Transformer model achieves the highest prediction accuracy, with an R² of 96.26%, and the lowest errors, with MAE and MAPE values of 2.18 mm/min and 4.18%, respectively, significantly outperforming comparative models. Furthermore, the model maintains high prediction accuracy across different rock mass grades, with R² consistently exceeding 95%, validating the effectiveness of the multi-module fusion architecture in improving TBM advance rate prediction and its adaptability to complex geological conditions. The research findings provide technical support for intelligent TBM-assisted Tunnelling and hold significant implications for tunnel construction under challenging geological conditions.