Scientists from Skoltech and Saint Petersburg State University have developed a new, more accessible method for determining the mineral composition of rocks in unconventional reservoirs, such as the Bazhenov Formation in Western Siberia. The machine learning-based technique uses standard well logging data combined with information on the thermal properties of rocks. It provides accurate mineralogy data along the entire length of the wellbore, offering a practical and cost-effective alternative to expensive and point-specific laboratory core analysis or downhole spectroscopy. The study was published in the journal Advances in Geo-Energy Research.
In oil and gas geology, reservoirs are rocks that contain oil and gas and can release them during field development. Unconventional reservoirs, like the Bazhenov Formation studied by the researchers, are more complex: They are heterogeneous, rich in organic matter, and require special approaches to study. Mineral composition directly affects porosity, permeability, and other rock properties that determine the success of hydrocarbon exploration and production.
Traditional methods like core analysis or downhole spectroscopy are either too localized to be representative or too expensive for continuous application. The machine learning-based method proposed by the scientists enables a detailed mineralogy profile along the entire wellbore using standard logging and thermal measurement data, making rock analysis simultaneously accurate, continuous, and economically accessible.
The developed approach can predict with high accuracy the mass and volume fractions of key minerals such as clay, calcite, dolomite, pyrite, quartz-feldspar-mica, and siderite, while also accounting for organic carbon content. The method is based on a gradient boosting algorithm integrated into a special regressor chain structure that considers the relationships between the minerals being determined. The model demonstrates high accuracy comparable to specialized and expensive instruments. To validate the results, the predicted volume fractions of minerals were used in a theoretical model to calculate rock thermal conductivity, and the calculated values showed strong agreement with experimental measurements.
“The key challenge when working with unconventional reservoirs, such as the Bazhenov Formation, is their high heterogeneity and complex mineral composition,” commented Batyrkhan Gainitdinov, the lead author of the paper and a PhD student in the Petroleum Engineering program at Skoltech. “Our model, trained on standard logging data enriched with thermal measurements, showed that even without expensive specialized studies, the mineralogical profile along the wellbore can be reconstructed with good accuracy. We were able to quantitatively demonstrate the contribution of thermal data: Adding it reduced the prediction error for mineral volume fractions.”
“The comprehensive approach we developed — from data preprocessing and model selection to physical validation through thermal conductivity calculations — creates a foundation for practical application,” added Dmitry Koroteev, the research supervisor and a professor at Skoltech’s Petroleum Center. “This method can be used for rapid data interpretation during drilling, identifying promising intervals in complex reservoirs, and optimizing enhanced oil recovery techniques, ultimately helping to reduce economic costs in exploration and field development.”
