Machine Learning-Based Prediction of Drilling and Blasting Tunnel Initial Support Patterns

When the construction encounters a different geological condition from the design, the altered support patterns of China's NATM tunnel are designed by experienced experts based on the comprehensive evaluation of advanced geology forecast and their observation of the tunnel face. However, the current decision methods require expertise, and there are often subjective discrepancies. This study proposes a data-driven method for the support pattern decision. The inputs are thirteen variables describing the tunnel faces and the support pattern of the corresponding tunnel face recorded on the construction site are the outputs. We selected five classic machine learning classification algorithms (support vector machine(SVM), random forest, logistic regression, Gaussian process, and Naïve Bayes) through trial calculations and the optimal parameters of the models were obtained by using the grid search cross-validation technique. The results show that the voting classifier constructed with these five algorithms has good accuracy in the prediction of the support patterns. The feature importance rank of input variables is determined by sensitivity analysis, which enhances our understanding of the relationship between surrounding rock and support.

Keywords: Machine learning, rock tunnel, prediction, support patterns, tunnel face, decision support.