Prediction Model for Shield Tunneling Roll Angle and Pitch Angle: A PCA-PSO-LGBM Approach

Accurately predicting shield roll angle and pitch angle based on different geological conditions and excavation parameters is significant for adjusting the shield’s posture in advance and ensuring safe excavation. Based on this, this paper proposed a prediction model using Principal Component Analysis (PCA)–Particle Swarm Optimization (PSO)–Light Gradient Boosting Machine (LGBM). The composite strata’s primary physical and mechanical parameters were obtained through the strata area weighting method, combined with selected adjustable excavation parameters to form the input parameters of the model. The results showed that the PCA-PSO-LGBM model performs exceptionally well in predicting both roll angel and pitch angle, with mean absolute errors of only 0.068° and 0.048°, respectively, and determination coefficients reaching 0.98 or higher for both. A comparative validation with other models confirmed the superiority of the PCA-PSO-LGBM model. Additionally, the proposed strata area weighting method comprehensively considered the influence of geological factors on the roll angle and pitch angle. Sensitivity analysis revealed that the pitch angle is generally more sensitive to excavation parameters than the roll angle. During the adjustment of shield tunnelling parameters, priority should be given to adjusting hinged pull, hinged thrust, lower cylinder propulsion pressure, and cutterhead speed.