Random forest-based predictors for driving forces of earth pressure balance (EPB) shield tunnel boring machine (TBM)

Abstract

During the design and excavation stage of the shield tunnel, the prediction of driving forces is pivotal, especially in complex strata. The accuracy of theoretical models for driving loads calculation is insufficient, owing to the academic hypothesis and parameter setting deviation. In this regard, a load prediction approach of an earth pressure balance (EPB) shield tunnel boring machine (TBM) based on a machine learning method (i.e., random forest, RF) was developed. Geological conditions and shield operational data were chosen as features to build the prediction models. The models were then applied to the thrust and torque prediction of an EPB shield TBM in a soil-rock mixed-face ground. The application effect of the RF-based method was evaluated by the coefficient of determination (R2) and the root mean square error (RMSE). Then we discussed the transferability of the built RF model for other EPB shield projects. The results showed that the load prediction capacity of the approach based on the RF algorithm was more accurate than that of theoretical models. Variables importance analysis took the decrease of the mean square error (MSE) as an evaluation indicator. The cover depth of the tunnel ring, the foam quantity, the advance rate, and the rotation speed of the screw conveyor were the top four most influential features for the RF-based thrust and torque prediction model. Overall, this study provides a valuable reference for the safe and efficient EPB shield tunnel excavation in the soil-rock mixed-face ground.