SPATIAL DISTRIBUTION ESTIMATION CONSIDERING CORRELATION AMONG MULTIPLE GEOTECHNICAL PROPERTIES USING GAUSSIAN PROCESS REGRESSION

Abstract

For the safety management of geotechnical structures, it is important to accurately estimate the spatial distribution of geotechnical properties. However, the points where geotechnical properties can be obtained are limited, and it is desirable to use all the available information effectively for the estimation. More accurate estimation is possible if other measurement data that are highly correlated with the geotechnical properties to be estimated are available. In this study, we extend the spatial distribution estimation method using Gaussian Process Regression based on the superposition of multiple random fields, and propose a method that considers the correlation among them for the random component. The spatial distribution of the estimated geotechnical properties of the three types of measured data was separated into trend and random components, and a comparison was made between the random component with and without considering the correlation of multiple properties. When correlations among multiple properties are taken into account, the uncertainty of the estimation is reduced and the targeting property is more accurately estimated.