Intelligent shield machine selection for subway tunnel using machine learning

Abstract

Shield machines are specialized equipment for tunnel construction, and selecting a proper machine is crucial for an efficient and safe tunneling project. This paper presents an intelligent methodology for selecting shield machines in projects, using data from 146 cases. Firstly, main shield parameters are extracted by an improved k-medoids clustering based on grey correlation analysis. Secondly, data quality is ensured by integrating four imputation methods and two outlier filtering methods. Then, the Single Input Multiple Output Recurrent Neural Network with Weights determined by a Hierarchical Agglomerative Clustering module (WHAC-SIMO-RNN) model predicts shield machine type, cutterhead type, opening rate, rated thrust, and breakout torque. The proposed method's adaptability is evaluated by comparing the predicted shield parameters with those used in the three real projects. Result shows that this model framework can achieve a fully intelligent determination process for shield machine selection, providing a reference for future real shield tunneling projects.