关键词: 水泥粉煤灰碎石（CFG）桩, 地基承载力, 高斯过程, 机器学习, 预测

Abstract: Gaussian Process（GP） is a newly developed machine learning technology and has become a power tool for solving highly nonlinear problems.Aiming at the fact that it is still difficult to reasonably determine the bearing capacity of composite foundation with Cement-flyash-gravel（CFG） pile，the model based on GP is proposed for forecasting bearing capacity of composite foundation with CFG pile.According to few training samples，the nonlinear mapping relationship between bearing capacity of composite foundation with CFG pile and influencing factors is established by GP model.GP model is applied to a real engineering.The results of case study show that GP method is feasible，effective and simple to implement.It has merits of self-adaptive parameters determination and excellent capacity for solving non-linear small samples problems.The good performance of GP model makes it very attractive for application in the foundation engineering.

Key words: cement-flyash-gravel pile, capacity of foundation, Gaussian process, machine learning, forecast