Prediction of shield tunnelling parameters based on temporal convolutional network

This paper establishes a novel deep learning model for real-time prediction of shield cutter head torque during tunnelling. A hybrid model was developed which integrates wavelet transform (WT) and temporary convolution network (TCN). We comprehensively assess the predictive performance of the model by R² and RMSE indexes. The validity of the model is verified by Fuzhou Binhai Expressway Project. Another two excellent recurrent neural networks including LSTM and GRU were used for comparison. The results show that the WT-TCN model achieves the highest prediction accuracy, R² and RMSE are 0.995 and 17.078. And the accuracy improvement in RMSE can reach 33.7% and 57.7% respectively. The prediction of tunneling parameters based on WT-TCN algorithm can significantly improve the judgment level of shield tunneling state, which is beneficial for optimizing and adjusting construction parameters.