Gaussian process model of water inflow prediction in tunnel construction and its engineering applications

Due to the extremely complicated hydrogeological environment, significant symptoms of water inrush can not be detected accurately using normal exploratory methods, which produces hundreds of water inrushes occurred during tunnel construction in karst area. This study aims to present a new water inflow prediction technique without considering the relationship between hydrogeological features and water discharge rate. Therefore, the nonlinear regression Gaussian process analysis is applied to develop a model for predicting water inflow into tunnels. In order to meet the requirement of the data format of Gaussian process regression model (GPR), the basic evaluation index system of water inflow into tunnels and corresponding criterion are set up and quantified based on the statistical information of water inrush cases. To verify its feasibility, The GPR model is applied to Zhongjiashan tunnel on Jilian highway in China. The results of the comparisons indicate that the prediction results obtained from the GPR model are generally in a good agreement with the field-observed results. The proposed Gaussian process, on the whole, performs better than the support vector machine (SVM) and artificial neural network (ANN) in predictive analysis of water inflow into tunnels.