These algorithms can be used not only in practice, but also for scientific research.

TASS, July 15. Russian scientists have created an artificial intelligence system that can predict how much oil can be extracted from a particular field using multistage hydraulic fracturing. This is reported by the press service of the Skolkovo Institute of Science and Technology with reference to an article in the Journal of Petroleum Science and Engineering. 

"Our approach opens up opportunities to create a system that will provide engineers with recommendations on the optimal set of parameters for hydraulic fracturing, or at least information about narrower ranges to find the right set of design parameters," said one of the developers, Skoltech professor Andrey Osiptsov.

Over the past century, geologists have discovered many deposits of oil and other hydrocarbons, "sealed" inside deposits of shale rocks. They contain large reserves of minerals, but until recently it was either impossible or unprofitable to extract them. About 50 years ago, engineers solved this problem by coming up with a technique for hydraulic fracturing.

Its essence lies in the fact that oilmen drill a special network of wells in the oil-bearing rock, into which a specially selected viscous liquid with solid particles is pumped under pressure. As a result, many cracks form in the rocks, through which hydrocarbons can be pumped out.

Hydraulic fracturing is now used for oil production in the United States, Canada and other countries with large deposits of shale rocks. In Russia, it can be used to develop the so–called Bazhenov formations - oil-bearing shale rocks, giant deposits of which were formed at the end of the Jurassic period at the bottom of the sea, which was located on the site of modern Western Siberia. According to Gazprom Neft experts, they contain from 1 to 60 billion tons of oil.

Artificial intelligence-"neftyanik"

In recent years, the technique of hydraulic fracturing has become so complex that it requires calculating all the properties of the field using very accurate computer models. And even such calculations, as practice shows, do not always give a completely optimal result. Because of this, a significant part of the oil remains inside the rock.

Osiptsov and his colleagues have found a solution to this problem. They adapted a machine learning system to predict how much oil can be extracted from an arbitrary rock formation using multistage hydraulic fracturing.

Russian oilmen and scientists have been collecting a detailed database for two years, which includes comprehensive information on production volumes and the nature of the device for about six thousand wells and 20 oil fields in Western Siberia.

Using this data, mathematicians from Skoltech trained the AI system to accurately calculate how a well will behave during production and how much oil can be extracted from it using the current parameters of hydraulic fracturing. In addition, as scientists note, this algorithm can also be adapted to solve the inverse problem – to choose the optimal parameters for hydraulic fracturing.

In addition, the same approach can be used for scientific research, in particular, to determine how the composition of the liquid, the amount of solid particles in it and other work parameters affect the amount of oil extracted and other important aspects of well operation. These calculations will help oil companies more optimally and safely extract hydrocarbons from the bowels of the Earth, scientists conclude.