Deformation prediction of foundation pit using Gaussian Process machine learning

Abstract

Prediction of deformation of foundation pit by means of conventional method such as mechanics analysis or numerical method often has a large error because the deformation process of foundation pit is a highly complicated nonlinear evolution process. A novel method based on Gaussian Process (GP) machine learning is proposed for solving the problem of deformation prediction of foundation pit. GP is a power tool for solving high dimensional, nonlinear, small sample problems. A GP model for deformation prediction of foundation pit is proposed firstly. The nonlinear reflective relationship between deformations in deep foundation pit and their effective factors is built easily by learning the historical knowledge using the GP model. Furthermore, the other GP model for displacement time series prediction of foundation pit is also proposed. The results of cases studies show two models are feasible. The models both have advantages in low calculation cost and higher precision comparing to the traditional methods. The results of studies also show that GP are more suitable for solving small samples prediction problems comparing to artificial neural networks and Grey method.