Prediction of the slurry pressure and inversion of formation characteristics based on a machine learning algorithm during tunnelling in a fault fracture zone

The reasonable setting of the slurry pressure is very important for the safety of shield tunnel construction. In view of the common geological problems of high water pressure and multiple fracture zones in water conveyance tunnels in China, the instability mechanism of the shield excavation face in these formations remains unclear. In this paper, based on the Pearl River Delta Water Resources Allocation Project, various machine learning algorithms are introduced, and the predicted values of the slurry pressure obtained with different machine learning algorithms are evaluated and compared. The optimal machine learning algorithm suitable for this project is determined. Then, using this algorithm, considering the influence of the fault fracture zone, the concept of the formation influence coefficient is proposed, which is used as the input of the prediction model. The influences of the formation coefficient distribution and the influence width of the fracture zone on the accuracy of slurry pressure prediction are examined. Finally, based on this model, the slurry pressure of nearly 100 groups is predicted and verified. The results show that (1) if the influence of the fault fracture zone is considered and a reasonable influence width of the fault fracture zone can be determined, the mean absolute error (MAE) of the predicted slurry pressure is much higher than that of the predictions without considering the fault fracture zone, and the prediction accuracy can be improved by approximately 25%; (2) when the formation influence coefficient is obtained via the method of the theoretical distribution and the influence width of the fault fracture zone is 17 rings, the prediction accuracy of the slurry pressure reaches the optimal value; and (3) through on-site refined investigation, it can be determined that the influence width of the fault fracture zone is 19 rings. The research results of this paper can provide an effective method for setting the shield slurry pressure under highwater pressure conditions in the fault fracture zone.

Automated summary

In this paper, a machine learning-based method for predicting the slurry pressure in shield tunnel construction is proposed. By considering the influence of fault fracture zones and setting the formation influence coefficient, the accuracy of the prediction is significantly improved.