Optimization of TBM tunnelling efficiency based on stacking ensemble learning

An optimization model for the TBM tunneling rate was established based on data from the TBM work area in the Qinling Tunnel of the Citing Han-Jiwei Tunnel, utilizing the stacking ensemble learning method. RF regression, DT regression, XGBoost regression, LightGBM regression, linear regression, and ridge regression serve as the foundational classifiers. Ridge regression serves as the meta-classifier to address the bias present in multiple learning algorithms applied to the training set. An analysis was conducted on the factors influencing TBM tunnelling efficiency, resulting in the identification of 14 parameters that affect the tunnelling rate. The RF algorithm was employed to rank the weights of these parameters. From this ranking, the five parameters with the highest weights, along with four parameters associated with rock bursts, were selected, culminating in a total of nine parameters designated as the input for the tunnelling rate model utilizing stacking ensemble learning. The evaluation of model performance was conducted utilizing mean square error (MSE) and mean absolute error (MAE) as metrics. The stacking model demonstrated a mean squared error of 0.1071 and a mean absolute error of 0.2771, surpassing the performance of individual learners and indicating enhanced accuracy and robustness in optimizing TBM tunneling efficiency.