Contents lists available at ScienceDirect Tunnelling and Underground Space Technology journal homepage: www.elsevier.com/locate/tust Modeling the relationship between ground surface settlements induced by shield tunneling and the operational and geological parameters based on the hybrid PCA/ANFIS method D. Bouayad a , F. Emeriault b, ⁎ a Bejaia University, Department of Civil Engineering, Algeria b Grenoble-INP, UJF-Grenoble 1, CNRS UMR 5521, 3SR Lab, Grenoble F-38041, France ARTICLE INFO Keywords: Ground settlement Shield tunneling ANFIS Principal Component Analysis Hierarchical clustering TBM operation parameters Geology ABSTRACT This paper proposes a methodology that combines the Principal Component Analysis (PCA) with Adaptive Neuro Fuzzy based Inference System (ANFIS) to model the nonlinear relationship between ground surface settlements induced by an earth pressure balanced TBM and the operational and geological parameters. The study is based on data recorded during the excavation of contract 2 of the subway line B tunnel in Toulouse (France). Prior to modeling, principal components analysis (PCA) and agglomerative hierarchical clustering (AHC) are used to describe the interrelation pattern between the TBM parameters and geology profiles. At first, a model ANFIS based on 10 selected TBM operation parameters (geology conditions considered as homogenous) is developed and validated by drawing the settlement profiles for different reference points. Secondly, to take into account the effect of geology on the settlements, 5 parameters representing the thicknesses of categories of soil were added as input variables. Then, a model ANFIS using the significant principal components as inputs is developed and validated. The results indicate a high correlation between predicted and measured settlements despite the low amount of data used in the analysis. In addition, the model is able to predict Gaussian troughs for the representative groups identified by the AHC. The results show that the shape of the predicted settlement troughs could be explained by the TBM parameters and soil profiles that characterize each group. 1. Introduction The prediction of ground surface settlements induced by shallow tunnel excavation with tunnel boring machines (TBM) is a complex problem that involves a large number of TBM operation parameters that are strongly related to the encountered geology and to the pilot management of the machine. Hence, it is very difficult to model the nonlinear relationship between the settlement and parameters using conventional methods such as least squares regression (Vanoudheusden, 2006). To model this relation, Bouayad et al. (2014) used the partial least squares regression (PLSR) taking into account the interaction between TBM parameters. Despite a logarithmic transformation performed on the displacement, the developed model does not capture the high nonlinearities. Recent studies have demonstrated that artificial neural networks (ANN) can be efficiently used to model the nonlinear relation between a large number of parameters involved in tunneling construction and the induced ground movements (Suwansawat and Einstein, 2006; Boubou et al., 2010; Mahdevari and Torabi, 2012). More recently, sophisticated models were developed using ANN with evolutionary techniques such as fuzzy logic, and support vector machines algorithms (Mahdevari et al., 2012, 2013; Ocak and Seker, 2013). The technique known as Adaptive Neuro Fuzzy based Inference Systems (ANFIS) is proved to be a powerful tool for modeling complex problems because of its ability in treating imprecision and uncertainty that may affect data in general. It has been successfully applied to various geotechnical problems but the prediction of settlements in- duced by tunneling has been recently explored (Hou et al., 2009; Adoko and Wu, 2012; Bouayad and Emeriault, 2013, 2014). Hou et al. (2009) used ANFIS to predict the maximum surface settlement induced by an earth pressure balanced TBM. They showed that the error between the predicted and measured settlements does not exceed 7%, moreover, the comparison of the latter with the error given by back propagation neural network confirmed that ANFIS was more accurate. This paper proposes a methodology that combines the Principal Component Analysis (PCA) with ANFIS method to model the nonlinear http://dx.doi.org/10.1016/j.tust.2017.03.011 Received 11 April 2016; Received in revised form 25 November 2016; Accepted 29 March 2017 ⁎ Corresponding author. E-mail address: fabrice.emeriault@3sr-grenoble.fr (F. Emeriault). Tunnelling and Underground Space Technology 68 (2017) 142–152 0886-7798/ © 2017 Elsevier Ltd. All rights reserved. MARK