Prediction of shield machine posture using the GRU algorithm with adaptive boosting: A case study of Chengdu Subway project

Abstract

Shield machine deviation from the design tunnel axis (DTA) causes dislocation and damage of the segments and may lead to poor tunnel quality, which is a primary concern in tunnel construction. Therefore, it is necessary to predict the shield machine posture dynamically and assist the operator in adjusting the tunneling parameters in advance. Based on the tunneling data of five earth pressure balance (EPB) shield machines, a novel method for predicting shield machine posture, mainly composed of adaptive boosting (AdaBoost) and gated recurrent unit (GRU) algorithms, is proposed in this paper. In parallel, a data preprocessing algorithm is developed for the original tunneling parameters, including three phases: data extraction, data compilation, and data normalization. The hyperparameters of the model were determined using the grid search and cross-validation technology. The actual deviation case test shows that once the model predicts that the shield machine posture will deviate significantly from DTA, it can issue a warning in advance and assist the machine operators in optimizing tunneling parameters for a better trajectory. Then, the model prediction results were compared with the benchmark algorithms. The results reveal that the GRU algorithm is conductive to capture the trend of the time sequence, and the AdaBoost algorithm is beneficial for improving the fitting ability of the regression model. Finally, we found some association rules of tunneling parameters that affect the posture of shield machine.