Prediction of ground surface settlement by shield tunneling using XGBoost and Bayesian Optimization

Abstract

Ground surface settlement caused by shield tunneling is a complex problem caused by multiple factors. Machine learning models can help in the nonlinear intelligent prediction of ground surface settlement caused by shield tunneling. At present, the Artificial Neural Network model and the Support Vector Machine model are the most widely used models for settlement prediction. Due to the black-box characteristics of these two models, they are inherently deficient in interpretability, which means it is difficult to provide guidance for engineering. To solve the problem of poor interpretability in the prediction of ground surface settlement using these two models, an ensemble learning algorithm called the XGBoost model is introduced. In order to select hyperparameters in XGBoost more efficiently, the Bayesian optimization is used for parameter search. In this study, 533 cases of ground surface settlement monitoring data from a shield tunnel construction project in a city were used. Compared with the prediction results of the ANN model and the SVM model, the XGBoost model has the advantages of prediction accuracy and interpretability, especially for the prediction of out-of-limit settlement points.