Thrust force requirement prediction using HOA-XGBoost for TBM tunneling in squeezing ground

Abstract

Tunnel Boring Machine (TBM) has been widely used in deep and long tunnels due to its highly efficient advantage. However, TBM can be subject to adverse geology, especially the soft and weak ground, so the TBM jamming, which seriously affects the construction schedule, could happen. Thrust force requirement evaluation avoiding TBM jamming is therefore important for TBM tunnel construction at the early stage or machine designing phase. In this research, some intelligent machine learning models are constructed to conveniently assess the enough thrust force based on the thousands of numerical simulations and existing nomograms. And the detailed model performance comparisons are carried out and concluded that XGBoost model outperforms the random forest (RF), multiple of linear regression (MLR), general regression neural network (GRNN), and support vector regression (SVR) models. Furthermore, the expression coefficient in MLR can reveal the relationship between the input variables and output. For further improving the XGBoost's accuracy, the hiking optimization algorithm (HOA), sparrow search algorithm (SSA), and northern goshawk optimization (NGO) are adopted here and used to adjust XGBoost's hyperparameter combination, the HOA-XGBoost (MAE=0.0191, RMSE=0.0232, R2=0.9853 in the testing set) is finally the most precise model and also behaves the most stable and robust among all the built models through the 5-fold cross-validation processes.