Ground Surface Deformation Prediction Using Machine Learning Approaches

This study aims to investigate the use of machine learning algorithms to predict the settlements of the ground surface triggered by tunneling using a shielding method. First, a 2D tunnel calculation scheme was proposed for nonlinear construction stage analysis using the “contraction” load, which can be used to simulate the soil volume loss. Further, this model was verified by comparing the solution with the analytical method and numerical simulation, which uses standard modeling approaches, including in the spatial formulation. Then, after briefly describing the main theoretical substantiation of machine learning techniques, the 2D scheme was used to create the dataset. Finally, the dataset was processed by the machine learning algorithms: linear regression, decision tree, random forests, polynomial regression, ridge regression and neural network, which showed the best forecasting ability (R2 = 0.985; RMSE = 0.000986). The results indicate that machine learning can provide a robust approach for predicting ground settlements, mitigating risks associated with tunneling operations. Additionally, this research highlights the potential of integrating advanced computational methods with traditional engineering analyses, enhancing the accuracy and efficiency of settlement prediction. Future work will explore the applicability of these algorithms to different soil and geological conditions, aiming to further improve predictive models in geotechnical engineering. By establishing a strong correlation between simulated parameters and machine-learning outcomes, this study opens avenues for more intelligent and adaptive tunneling practices.