Multi-source data driven method for assessing the rock mass quality of a NATM tunnel face via hybrid ensemble learning models

• Published in: International journal of rock mechanics and mining sciences (Oxford, England : 1997) England : 1997), 2021-11, Vol.147, p.104914, Article 104914
• Main Authors: Zhou, Mingliang, Chen, Jiayao, Huang, Hongwei, Zhang, Dongming, Zhao, Shuai, Shadabfar, Mahdi
• Format: Article
• Language: English
• Subjects: Algorithms; Data driven; Empirical analysis; Ensemble learning; Excavation; Learning algorithms; Machine learning; Multi-source information; New Austrian Tunnelling Method; Quality assessment; Regression analysis; Rock mass quality; Rock mass rating; Rocks; Sensitivity analysis; Sensitivity enhancement; Tunnel face; Tunnels; Weights & measures
• ISSN: 1365-1609; 1873-4545
• DOI: 10.1016/j.ijrmms.2021.104914

Current assessments of rock mass quality of a NATM tunnel face are important in the practice of tunnel excavation. This study establishes a multi-source database and proposes a data driven method for the assessment. Thirteen multi-source variables describing the tunnel faces are considered as inputs, and the rock mass rating (RMR) values computed by the empirical formula are the target outputs. We adopted two meta machine learning models (classification and regression tree (CART) and multiple layers perceptron (MLP)) and two ensemble learning models (gradient boosting regression tree (GBRT)) and random forest (RF)) to capture the relationships between the inputs and outputs. The tree-structured Parzen estimator (TPE) algorithm is applied to automatically determine the optimized model hyper-parameters. The experimental results suggest that the proposed hybrid ensemble learning models (TPE-RF and TPE-GBRT) perform well at assessing rock mass quality. The feature importance ranks of the input variables are determined by a sensitivity analysis, which enhances the knowledge on assessing the rock mass quality of a tunnel face. [Display omitted]