Research on the prediction of TBM boring parameters based on CNN-BiGRU-RF model
As a new method of underground tunnel boring,tunnel boring machine(TBM)has significant economic benefits,so the prediction of TBM boring parameters is of great significance to ensure the boring efficiency of TBM.The TBM data obtained in the field were cleaned and preprocessed,the model input features were screened using Pearson correlation coefficient method,and a TBM boring parameter prediction model was integrated by random forest(RF)on the basis of bidirectional gated recurrent unit(BiGRU)neural network optimized by convolutional neural network(CNN)to predict the boring parameters of the TBM.The results showed that the cutter speed,cutter torque and penetration,which were most closely related to the total thrust and propulsion rate,were selected as the feature parameters;the constructed CNN-BiGRU-RF model predicted that the mean value of R2 of the boring parameters to the total thrust and the propulsion rate was 0.950 and 0.966,respectively;the average value of MSE was 0.750 and 0.782,respectively;the average value of RMSE was 0.866 and 0.885,respectively;the average value of MAE was 1.054 and 1.007,respectively;and compared with the CNN-BiGRU model,the value of regression evaluation metrics,MSE,RMSE and MAE,was reduced by 2.497,0.966 and 0.386,respectively,and the R2 was improved by 23.4%,which proved that the CNN-BiGRU-RF model had the highest prediction accuracy and generalization.This study provides guidance for the prediction of boring parameters in practical engineering,which helps the promotion of TBM in coal mines and the construction progress of TBM.
CNN-BiGRU-RF modelTBM boring parametersPearson correlation coefficientconvolutional neural network(CNN)bidirectional gated recurrent unit neural network(BiGRU)random forest(RF)time series predicting
NETL
NSTL
万方数据
