Multi-source data-driven prediction of geological cross-section during shield tunnelling

Abstract

This paper proposes a multi-source data-driven framework to estimate the geological profile through incorporating three geological datasets: borehole data, inverse distance weighting (IDW)-enhanced data, and shield parameters related data (shield-enhanced data). IDW-enhanced database is constructed using IDW method, which can obtain the spatial feature of strata based on the spatial relationship between the target area and surrounding boreholes. Borehole data is enhanced by shield parameters to refine geological cross-section and establish shield-enhanced database. Four machine learning methods are employed to assess the effectiveness of the proposed framework. The cross-validation using true boreholes is also adopted to compare the performance of machine learning models. The best model using a support vector machine (SVM) is finally selected to predict the geological profile. The multi-source data-driven framework increases the accuracy of prediction of geological cross-section from 0.77 to 0.938. The proposed framework is successfully applied to the shield tunnelling project in Guangzhou, China.