Typing of rocks of the Achimov sequence by machine learning methods based on the construction of their volumetric-component model

Abstract

Achimov sequence sediments in northern West Siberia are a classic example of a complex-built reservoir. Traditional petrophysical interpretation methods often fall short for such reservoir due to high geological heterogeneity, which manifests in wide variety in mineral composition and reservoir properties. The main method for enhancing the reliability of geological interpretation of well log data (WLD) in these sediments is robust lithological typing of the rocks. In this article, the authors suggest an approach to lithological typing based on the development of a Volumetric-Component Model (VCM). In the first stage, the researchers built two VCMs. The number of components for each model is determined using two distinct sets of WLD: an extended set (including Gamma Ray, Neutron Porosity, Litho-Density, Elemental Spectroscopy) and a standard set (Gamma Ray, Neutron Porosity, Litho-Density). In the second stage, these VCMs serve as input data for configuring machine learning algorithms aimed at lithological typing of the rocks. This approach improves the accuracy of lithotype predictions in wells without core compared to the traditional statistical analysis performed directly on original well log curves.