A multi-stage learning method for excavation torque prediction of TBM based on CEEMD-EWT-BiLSTM hybrid network model

Accurate and reliable prediction of cutterhead torque sequence is of great significance to ensure safe and efficient tunnel boring machine (TBM) propulsion. Based on this, a torque prediction method combining complete ensemble empirical mode decomposition (CEEMD), empirical wavelet transform (EWT), and bidirectional long short-term memory (Bi-LSTM) models is proposed. First, CEEMD and EWT were used to reduce the complexity of the original torque sequence. CEEMD was used to decompose the original torque sequence into multiple intrinsic mode functions (IMF) and residual sequences, and the main IMF components were further decomposed by EWT. Then, the Bayesian optimization Bi-LSTM model is used to predict the decomposed sub-sequences, and the final predicted torque value is obtained by superimposing the predicted results. Finally, the measured data in different surrounding rock excavation processes are used to verify the prediction results, which show that the proposed method has high prediction accuracy and generalization adaptability. The experimental results show that the MAE value and RMSE value of the proposed method are within 85 kN & sdot;m, and the MAPE value is less than 3.5% for different classes of surrounding rock.