Researchers from Perm have enhanced oil extraction efficiency.

In oil extraction, forecasting reservoir properties is a crucial task that enables the assessment of a field's potential and informs decisions regarding its efficient development. Typically, geophysical studies are conducted to determine the characteristics of rock formations, such as porosity, density, and permeability. These findings are then used to create a 3D model of the reservoir, providing insights into the oil and gas reserves contained within. However, the structure and properties of reservoirs can vary significantly, and this heterogeneity often hinders the accuracy of data obtained through traditional methods. Researchers at Perm Polytechnic University have developed an approach for modeling porosity in the oil and gas sector using artificial intelligence. This innovation is expected to enhance forecasting accuracy and improve the efficiency of field development by 56%.

2025-02-28 10:00:10https://naked-science.ru/article/column/povysili-effektivnost-nef

The article with the results has been published in the journal "Geosystem Engineering." The research was funded by the Ministry of Education and Science of Russia.

Reservoirs are rock formations that contain voids capable of holding, retaining, and releasing fluids (oil, gas, or water) during extraction. Modeling their properties is one of the key tasks in evaluating deposits, where accurate forecasting of reservoir porosity is of particular importance. Traditionally, core samples of rocks are studied, and geophysical surveys of wells are conducted for this purpose. Specifically, radioactive, electrical, and acoustic logging allows for the physical measurement of density, porosity, and permeability of rocks. However, in complex geological conditions, the technical limitations of these methods and the heterogeneity of the layers reduce the accuracy of predictions. Neural networks and machine learning can enhance the quality of forecasts and the precision of 3D modeling of deposits.

Scientists from Perm Polytechnic University proposed an approach to assess the porosity of reservoirs using machine learning algorithms developed based on existing results from geophysical studies of wells. The obtained data were integrated into a 3D model of the deposit, which allowed for a refined distribution of porosity and a recalculation of oil reserves.

The polytechnic researchers conducted studies on a deposit with a complex structure, where porosity varies from 0.7% to 24%, and permeability ranges from negligible values to 2.364 μm². To train the algorithm, they compiled a database using the results of geophysical studies from 238 wells across six deposits. In addition to these, they also included results from laboratory studies of core samples to determine porosity.

“We conducted comprehensive work on data collection, training, and tuning the algorithm to enhance its accuracy and ensure the adaptability of development under specific conditions. The machine learning model we built was used to refine the geological model of the deposit and recalculate oil reserves. We performed porosity predictions for 22 wells. As a result, we noted a 56% increase in accuracy compared to the standard method,” says Sergey Krivoshchekov, an associate professor at the Department of Oil and Gas Geology at PNIPU and a candidate of technical sciences.

The refinement of the 3D model using the developed algorithms helped reveal that there is an overall moderate increase in hydrocarbon reserves across the deposit. This is attributed to the increase in average porosity values compared to the initial model.

“We identified additional locations with oil reserves that were previously not utilized in extraction. This allowed us to adjust the production plan to include new zones. The developed approach enables more efficient use of deposit resources, reducing costs and increasing production volumes,” explains Georgy Shiversky, a graduate student at the Department of Oil and Gas Geology at PNIPU.

The work of PNIPU scientists has demonstrated the potential of using machine learning algorithms for modeling and forecasting porosity in conditions of high geological heterogeneity. The developed approach allows for the automation and enhancement of the quality of well property predictions, optimizing the development of oil fields. In the near future, such technologies will become standard tools in subsurface research, merging accumulated geological knowledge with the latest advancements in data analysis and artificial intelligence.