An Innovative Wavelet Transform Convolutional Approach for Multi-Step Cutterhead Torque Prediction in Shield Tunneling Machines

Accurate prediction of cutterhead torque, which serves as a critical operational parameter for shield tunneling machines, is paramount to ensuring safe excavation processes. However, complex geological conditions induce substantial torque fluctuations, posing significant challenges for prediction accuracy due to the coexistence of high-frequency noise interference and low-frequency baseline drift in torque signals. To address this challenge, we propose a hybrid Wavelet Transform-Convolutional Transformer (WTCT) model for time series prediction. The WTCT framework innovatively integrates wavelet decomposition with attention mechanisms: Input parameters are decomposed into distinct frequency components, enabling targeted feature extraction through convolutional operations on high-frequency details while preserving low-frequency trends via Transformer layers. This dual-path architecture effectively captures multi-scale temporal dependencies while suppressing noise contamination. Validated against real-world datasets spanning diverse geological strata, the model achieves superior performance with prediction accuracy reaching 98.52%. The demonstrated robustness to geological variations confirms its practical value for intelligent control systems in tunnel boring operations.