IDENTIFICATION OF LITHOLOGICAL AND FACIES FEATURES OF SEDIMENTS BASED ON MACHINE LEARNING METHODS

In the context of increasing computing power and the general depletion of open hydrocarbon reserves, the issue of detailing permanently operating geological and hydrodynamic models of deposits becomes more relevant. Building detailed models of hydrocarbon deposits is a time-consuming task that requires a large number of specialists of various profiles. The distribution of reservoir properties of sediments, mainly formed by lithological and facies conditions, and tectonic features of the development of the studied object have the greatest influence on the geological structure of the territory. The identification of these factors is associated with the interpretation of a wide range of data — borehole information (core material and data from geophysical studies of wells), geophysical work (seismic, gravity and electrical exploration). The automation of the primary data interpretation allows specialists to save time on the formation of general patterns of the territory's structure, as well as to detail the resulting models using more initial information. This paper presents the experience in developing an algorithm for processing data from geophysical studies of wells to isolate lithological and facies bodies in a section. The study was conducted on a site located in the Shaimsky oil and gas region of the West Siberian megaprovincial oil and gas area. The complex genesis of the productive strata led to the formation of traps with lithological, stratigraphic, and tectonic elements. Productive Jurassic and Lower Cretaceous sediments have different origins and are confined to both continental and marine environments. The effectiveness of the proposed approach was evaluated based on data from nine wells, where a preliminary lithological and facies analysis of sediment samples was conducted. Support vector machines and k-nearest neighbors were selected as the main algorithms for data classification. The results obtained demonstrated the potential of this technique. However, further analysis of different algorithmsand transformation of the initial logging curves is required to confirm its efficiency. The data collected during the operation of the machine learning model can significantly reduce the time needed for information interpretation. Lithofacies zones identified in the intervals contribute to improving the efficiency of hydrocarbon reserve development by influencing operational facilities and enabling targeted methods for enhanced oil recovery.