A novel data-driven approach for proactive risk assessment in shield tunnel construction

Abstract

Underestimation of risks during tunnelling may result in substantial economic losses and even fatal accidents. This study develops a data-driven approach for evaluating construction risk levels during tunnelling. Two computational models including the deep forest algorithm (DF) and fuzzy set pair analysis (FSPA) are fused, where the DF is employed for predicting shield operational parameters and the FSPA is utilized to evaluate the risk level based on the predicted operational data. Furthermore, a linear combination of the subjective and objective weights is adopted in FSPA. The proposed method is then applied to an intercity railway tunnel project in Guangzhou, China. The analysis results align well with the in-situ engineering observations for the first 600 rings. In addition, it effectively predicts a relatively high risk (level IV) during the construction of rings 1571 to 1580. The proposed method offers a reliable and feasible tool for proactively assessing the risk levels in tunnelling.