A Hybrid Numerical-ML Model for Predicting Geological Risks in Tunneling with Electrical Methods

Abstract

In order to ensure construction efficiency and stability during tunnel excavation, it is essential to predict geological risks ahead of tunnel faces. In this study, a geological risk prediction model was developed based on a machine learning (ML) algorithm. The database used to implement the ML model was synthetically acquired from a series of finite-element (FE) numerical analyses, which could simulate electrical resistivity surveys during tunnel excavation. The developed FE model helped obtain resistivity data representing various risky ground conditions (such as typical fault zones, water intrusion, mixed ground, geological transitions, and cavities) encountered during tunnel advancement. Four ML algorithms (support vector machine, k-nearest neighbors, random forest, and extreme gradient boosting) were used to develop the prediction model. The evaluation results showed that the proposed ML prediction models produced highly accurate results. Among the ML algorithms, the prediction model based on the random forest (RF) algorithm exhibited superior performance, with an accuracy of 97.33%. Given the feasibility and efficiency of recognizing hazardous ground conditions, the proposed model is expected to serve as a reliable approach for risk management. Finally, an engineering flowchart was proposed to assist in the application of the study results to actual tunneling sites.