Abstract: During TBM tunnelling,relying solely on the subjective experience of the lead driver to determine tunnelling parameters may cause some problems such as low construction efficiency,jamming,severe cutterhead abrasion,and collapse of the surrounding rock.In this paper,the least squares method was used to integrate the gate recurrent unit (GRU) and random forest (RF) for the development of a TBM tunnelling parameters predictive model (GRU-RF model),and the grey relational analysis method was employed to screen the input features of the model.The average of goodness of fit (R²),mean absolute percentage error (MAPE),and root mean square error (RMSE) of thrust,rotational speed and penetration predicted by the GRU-RF model were 0.81,8.32%,and 0.74,respectively,and the average relative error(RE) was almost zero.The bidirectional long short-term memory (BiLSTM) model,back propagation neural network (BPNN) model,GRU-BPNN model,and BPNN-RF model also were selected to compare and analyze the prediction error of each tunnelling parameter.The analysis results showed that the GRU-RF model had the highest prediction accuracy and generalization ability.And the integration of a traditional machine learning model and a deep learning model using the least squares method can construct a predictive model with strong predictive performance.Finally,the necessity of using the grey relational analysis method to select the input features of the prediction model was proved.This study provides guidance for the prediction of actual engineering tunnelling parameters and contributes to the advancement of intelligent TBM construction.