Kazan University’s technology helps predict location of sweet spots of oil
Deputy Director for Innovation of the Institute of Geology and Petroleum Technologies Vladislav Sudakov talks about the technology.
As he notes, KFU experts started working on neural networks for the oil industry in 2019. After that, the team won a grant of 217 million rubles from the Ministry of Science and Higher Education.
The project includes a software set for the extraction of late stage oil deposits. A significant part of the residual reserves is located in pillars – areas limited by washed high-permeability zones. The problem is to delineate these areas and create effective methods for their development.
Sudakov explains,
“The essence of the project is that, using the entire history of field development and artificial intelligence, as well as laboratory studies of fluids that are produced from the reservoir, we can predict the location of sweet spots – places where there is residual oil and where there is a lot of it . This is where you can get good profitable debits.”
The Deputy Director says that the final oil recovery factor of most fields is 40-50 percent,
“Thanks to our development, we plan to increase this figure by another 20 percent. These are huge volumes of oil that we can produce together with oil companies. We don’t need significant investment to extract this oil. We focus on old fields where the necessary infrastructure already exists.”
He emphasized that the project has no foreign analogues,
“At the moment, we have already prepared a package solution wrapped in an interface. And it can be used for the needs of not only Tatneft but also other companies.”
Among the companies who have already shown interest in the tool are Abu Dhabi National Oil Company, PetroChina, Baker Hughes, and Gazpromneft.
According to Sudakov, the work uses a method of presenting geological and field data from wells and reservoir characteristics of the near-wellbore zone to localize reserves in the form of a multi-channel image, which is used as input data for a convolutional neural network. These algorithms allow for multicomponent analysis that can take into account complex nonlinear dependencies. The result of the calculation using this method is a two-dimensional map of the field, containing the flow rate of oil, water and residual oil reserves for each cell, the scientist noted.
To select efficient development methods, a stochastic analogue of a hydrodynamic simulator is implemented using a generalized Kalman filter. The model is trained using the EM algorithm. In addition, regression models are additionally applied, which are trained in the intervals between iterations of the EM algorithm. The result of the project is obtaining an economic effect due to an increase in oil production and a decrease in water cut due to the localization of pillars of oil / sources of water intrusion.