Artificial intelligence in oil prospecting

Credit: Maksim Bogodvid| RIAN

Armen Zakharyan’s company Cervart has developed an AI-based program that evaluates oil reserves at the exploration stage and during the well’s operation. The company’s clients already include Russian oil giants Lukoil and Gazprom Neft; now Chervart is working with a small American company operating a complex field in Texas.

It is generally believed that artificial intelligence or machine learning should always imply neural networks, but I no longer use them because neural networks cannot predict. They can only show what was before,” the entrepreneur says.

Armen Zakharyan has been developing neural networks for about ten years, from 2001 to 2010, using data from all of Lukoil’s West Siberian oilfields – about 30,000 wells in explored fields. The developer was involved in predicting oil production rates after hydraulic fracturing. An oil well flow rate is the volume of fluid delivered through the well per unit of time. Hydraulic fracturing increases oil flow to a well from petroleum-bearing rock formations; this well development process costs about $100,000.

The hydraulic fracturing has the effect of 51%; in 49% cases it is water instead of oil. Developers and predictors could increase the effect by 5% by processing data using neural networks and that was considered a lot. Lukoil conducts around 500 fracturing projects per year and spends around $50 mio which is high enough to start thinking about a new and more precise method to predict oil reserves.

Cervart (Cervaello artifatto means ‘artificial intelligence’ in Italian) was founded in 2012. Armen Zakharyan left Lukoil to create a company and make his own money from clients’ contracts. Last year, the company earned around RUR 1.5 mio ($23K), which provided return on investment. Cervart partnered with Skoltech (Skolkovo Institute of Science and Technology) that also implements AI in the oil industry. It was in Skoltech that Zakharyan was introduced to Iskandia.

Zakharyan developed an AI-based program that can predict the prospects of an oilfield. The AI is built on the Big Data provided by the client. The Big Data is used to create and adaptive model of reserves that can also integrate new data. The model can work with any data. Thus, for predicting the oil flow rate, the data collected include location of the well, the depth of the reservoir, years in operation (for old wells), the amount of earlier produced oil, etc. The amount of data is quite large. The system is fully self-sufficient; the program’s algorithms can independently process and mark the data.

Zakharyan has been testing the program for about ten years now and claims that its results are as good and sometimes better than those of neural networks. Fifty-six percent is not precise enough for these predictions. Artificial neural networks are efficient for image recognition but not for predictions because they cannot create new situations that were not included in the teaching data selection. Neural networks were invented back in 1943 by two American scientists, neuropsychologist Warren McCulloch and mathematician Walter Pitts.

The AI-based program developed by Zakharyan is less expensive than a similar program created by Schlumberger, the world’s largest provider of oilfield services. Schlumberger’s license for appraising one oil well costs $300,000, while Cervart’s costs $2,000. Zakharyan is confident that the mass introduction of his company’s product could drive Schlumberger from the market. In addition, his program works faster basing on the same amount of big data. A year ago, Cervart received a contract for providing forecasting services from a small US oil company Iskandia, which has oilfields in Permian Basin Houston, Texas.

“Their oilfields are somewhat special; standard simulators like Schlumberger cannot appraise them, so Iskandia approached us. We can calculate basically anything – shale or any other type of oil. These oilfields are noted by producing 97 tons of water per three tons of oil, which usually occurs at the end of the extraction process; I have not seen such oilfields before”, Zakharyan noted.

Cervart’s AI-based program, unlike Schlumberger’s, makes its forecasts considering physical and informational principles. For instance, the formation pressure is measured physically as it is highly important as regards assessing the well flow rate. This indicator, however, is never considered when making forecasts. Another parameter is the formation permeability which provides information about holes and possible losses in the well. But it cannot be measured as it is reciprocal and cannot be calculated accurately. But when a model is calculated, the formation permeability indicator defines the well flow rate, but this is wrong as this data is less accurate. Forecasts primarily require quality and accuracy of the information and not abstract physical laws.

Zakharyan expects to receive a contract from Salym Petroleum for appraisal of the Salym oilfield, which has its oil in clay bodies which is extremely uncommon as it is usually found in sandstone. This makes modeling such oilfields in traditional simulators impossible; however, the adaptive method developed by Cervart can cope with it.

Cervart would like to work not only with oil producers but also with the Federal Agency for Mineral Resources (Rosnedra) as well as banks such as Rosbank, which is making large investments in developing AI. For Ronedra and Rosbank, Cervart could evaluate client data’s reliability.

By Natalia Kuznetsova

Previous ArticleNext Article