APPLICATION OF DEEP LEARNING TECHNOLOGIES FOR STUDYING THIN SECTIONS ON THE EXAMPLE OF USINSK OIL FIELD

Abstract

The article is devoted to development of methodological techniques for application of machine learning technologies, including deep learning, to the problems of in-depth analysis of geological and physical parameters based on the results of laboratory studies of core sections. To achieve this goal, we solve the problem of developing a specialized tabular format for describing the core sections of carbonate deposits, formation of a database on the basis of the developed format for further analysis and application of deep and surface training technologies. The permocarbon deposit of Usinsk field located in the Komi Republic was chosen as the object of research. Deep learning technology was applied to obtain a mathematical model for predicting a number of geological parameters from the photos of sections. As the main example, the forecast of eight classes of Danhem, allocated by sections, was considered. The developed format allows presenting all text descriptions of the geological characteristics of the section in a tabular form with a discrete encoding. The table view provides a number of advantages. First, it allows you to perform mathematical and statistical analysis of the description of sections. Second, it is possible to form a database for analysis, using the results of the work of different authors, including photographs of thin sections, thirdly, provides an opportunity to compare and analyze the parameters obtained for the sections with other results of studies of the cores. On the example of permocarbon deposit of Usinsk field, a unique database of 500 sections from 6 wells was formed according to the developed format. In addition to the descriptions of the sections, the database was loaded with information on the results of laboratory studies of various geological and physical parameters obtained on standard core samples from the same intervals as the sections. Using the formed database, the ratio of mineralogical density and permeability with the categorization of points according to the Danhem classification on the permocarbon deposit of the Usinsk field is constructed. The generated database of sections descriptions is related as well to photographs of sections, that, in its turn, allows the use of modern computer vision technologies based on deep learning to analyze and predict the parameters of sections. As a result of the experiments, a model was obtained, which allows distinguishing geological parameters from the photo of the plume. To date, work on updating the database and improving the model continues, but the model is already used as a tool to accelerate the process of sections analysis.

Keywords

• technology of machine learning
• investigations of core
• description of thin sections
• mathematical-statistical analysis
• the classification of danhem