Comparison of the Efficiency of Robust Solvers for Automated Prediction of Geological and Geophysical Parameters

Abstract

Summary There is often a need to predict the values of certain parameters based on a limited set of values of the observed feature from a certain local neighborhood in geological modeling. There are optimal solutions for various special cases, for example, for the case of a priori established spatial stationarity of the studied value, but in the General case, as a rule, there are a number of complicating factors that do not allow us to talk about the presence of any one optimal solution. These factors include: • extremely small sample size • presence of outliers in the source data • errors associated with measurement errors and rounding errors • complex multi-modal distribution of values • spatial non-stationary distribution of values (non-stationary mean, variance, distribution law) • possible heteroscedasticity of the distribution (proportionality of variance growth with average growth), including local • the presence of smooth gradations and abrupt changes in the spatial distribution of the values of the studied quantity The presence of such restrictions makes traditional methods of classical parametric statistics inapplicable, the use of a geostatistical approach, and even calls into question the applicability of some nonparametric methods. This article offers an analysis of possible approaches to forecasting geological and geophysical parameters, taking into account all the problems and limitations discussed above.