Surrogate models based on sparse estimation for geotechnical reliability analysis

This paper presents a method for building surrogate models for geotechnical reliability analysis based on sparse estimation. Sparse estimation, which is called least absolute shrinkage statistical operator (lasso) in statistics, has the property that some of the parameters in surrogate models are driven to zero and leads to simpler models. Building surrogate models can be divided into two processes, model selection and parameter estimation, and the sparse estimation enables to achieve these two processes at the same time. A surrogate model was designed to estimate consolidation settlement value of a specific time based on sparse estimation, and its applicability has been investigated by comparing the results by the surrogate model with those by finite element analysis.