Prediction of Maximum Surface Settlements of Bai∼Hua Tunnel Section based on Machine Learning

Abstract

Research on the settlement caused by subway tunnel construction has always been an essential issue in tunnel research. However, due to the complexity of soil characteristics and construction parameters, using empirical formulas or numerical simulations to predict the maximum ground settlement is challenging to balance ease of use and accuracy. In recent years, with the rapid development of machine learning theory and computer science technology, machine learning algorithms are increasingly being used to predict the maximum settlement. Random forest (RF) and artificial neural network (ANN) are often used to predict settlement. However, applying the extreme gradient boosting algorithm (XGB) in predicting the settlement is rarely seen. This article compares these three machines learning algorithms, using tunnel geometric parameters, shield construction parameters, and geological parameters as input parameters to predict the maximum ground settlement caused during tunnel construction. Compared with linear regression, the result shows these three machine learning algorithms can achieve higher quality results, and the stability of the RF and the XGB model is better than the neural network model. The XGB method can obtain the best results.