Researchers from Skoltech and Gazpromneft STC have patented a machine learning-based method that allows optimizing hydraulic fracturing technology, which oil and gas companies use to intensify well production. Based on the analysis of painstakingly collected data on thousands of wells, AI makes an individual forecast of optimal hydraulic fracturing parameters for a particular well.

Originally presented in the January issue of the Journal of Petroleum Science and Engineering and another article in the same journal, the development was recently registered by Rospatent. Hydraulic fracturing is the injection of a mixture of water, chemical additives and solid granules of a "wedging" material into an oil or gas—bearing rock under high pressure in order to create and fix cracks in the formation, due to which the inflow of hydrocarbons into the well increases. This operation is now carried out on almost any new well, but it can be performed in different ways. The main parameters include the composition and volume of the injected liquid, the injection rate, the characteristics of the granules and other technical details.

Usually, the values of all these parameters are set based on the results of a complex and long simulation. Andrey Osiptsov, the scientific director of the project, Professor at Skoltech, explains: "A physical model of the hydraulic fracturing process is taken, and parameter values supported by certain hypotheses and reasoning are fed to it — the model predicts how the productivity of the well will change with such parameters. This process is repeated many times to select the most successful set of parameter values, but such resource-intensive modeling takes days or even weeks, and we are talking about only one well, and there are many wells in one field."

"Instead, we trained artificial intelligence to predict well performance based on a unique database of field data that we collected and carefully verified — it contains 92 characteristics of the well, the surrounding rock and hydraulic fracturing for 6 thousand wells from 23 fields," says Anton Morozov, a graduate student at Skoltech. Since the productivity of these wells after hydraulic fracturing is already known for certain (and recorded in the same database in the form of values of 16 more characteristics), an AI trained on these data can predict productivity based on certain initial conditions.

"Being able to predict the productivity of a well after hydraulic fracturing based on the parameters of this operation, we can solve the so-called inverse problem. Given: characteristics of the well and reservoir. Find out: at what parameters of hydraulic fracturing will the largest oil or gas production be achieved?" explains project manager, senior engineer at Skoltech Albert Weinstein and clarifies that using AI and optimization algorithms, this task is solved very quickly. Dmitry Popkov, a graduate student at Skoltech, adds: "The method has been successfully tested at the Priobskoye field in Western Siberia and is applicable to any oil well."

"In the future, the solution can be improved and adapted for more careful and environmentally responsible extraction of fossil fuels in the spirit of ESG principles," Viktor Duplyakov, a graduate student at Skoltech, comments on the results of the work and explains that instead of striving to maximize total production at any cost, artificial intelligence can be tasked with balancing oil recovery and metrics significant for the environment, such as the amount of fresh water and chemicals used, or diesel fuel used to power pumps and greenhouse gases released into the atmosphere. "Perhaps all these metrics could even be combined into a single coefficient of environmental friendliness or energy efficiency of a single hydraulic fracturing operation," Egor Shel, an employee of Gazpromneft STC, shared his thoughts on possible future research.