GEOPHYSICAL STRUCTURE ANALYSIS BASED ON A SELF-LEARNING ALGORITHM

Abstract

This article is devoted to the development of methods for geophysical structuroscopy of rock masses in conditions of a lack of a priori information about the properties of objects. The author proposes the use of self-learning algorithms based on stochastic approximation to automatically determine the number of structural classes and their distribution parameters. The practical effectiveness of this approach is demonstrated using a study of a landslide slope in the village of Kok-Zhangak, where the a priori model was adjusted during the training process and the actual thickness of Quaternary deposits was determined. The use of the proposed recurrent procedures significantly improves the reliability of geological structure recognition compared to the classical Bayesian approach.