A novel dual-channel deep neural network for tunnel boring machine slurry circulation system data prediction

The slurry circulation system is a crucial component of the Slurry Pressure Balance Tunnel Boring Machine (SPB TBM),with the pressure and flow at the inlet and outlet sections pipelines significant parameters for SPB TBMs.Accurate prediction of these parameters is essential for maintaining face pressure and preventing surface settlement or heave, providing a reference for TBM control adjustments.This research proposes a novel Dual-channel Hybrid Model based on Variational Mode Decomposition and Self-attention Temporal Convolutional Networks (DHM-VSATCN) to address this issue.This multi-input multi-output model is designed to forecast pressure and flow in slurry pipelines accurately.This method encompasses several key components, including data preprocessing,signal decomposition, an enhanced dual-channel deep learning model,a loss function, and evaluation metrics to ensure prediction accuracy. Validation of the model using a real SPB TBM operation dataset demonstrates that the model achieves excellent performance for five pressure and flow rate parameters, with low Mean Absolute Errors (MAE) ranging from 0.0032 to 4.01,R² values above 0.95, and Mean Absolute Percentage Errors (MAPE) consistently below 0.23 %. The comparative analysis highlights the superior performance of the proposed DHM-VSATCN method over models such as SVR, XGB, FTS, ARIMA, RNN, LSTM and iTransformer. Furthermore,in the context of multi-output prediction problems,the proposed dual-channel modeling strategy not only ensures prediction accuracy but also reduces training time compared to existing modeling strategies. The proposed DHM-VSATCN achieves an all-MAPE of only 0.7253 % across five parameters,with a model training time of just 1212.8 s.Therefore, this method is an effective solution for predicting TBM performance and offers valuable insights for other engineering scenarios requiring the prediction of multiple related outputs using the same input.