Artificial Intelligence Will Select Wells for Geological and Technical Operations and Calculate Hydraulic Fracturing
The TN-Neurocram neural network is being created at Tatneft. The program is based on artificial intelligence (AI) tools. This is a completely own development of the company, which began to be created last year. Napoleon IT is a consultant for Tatneft in the implementation of AI projects. Oil News writes about this (Ramis AMINOV).
The Modeling Center of PJSC TATNEFT has become a platform for new experiments. For several years now, they have been engaged in geological and technical activities, digitalization of data and the construction of digital twins of development objects. All this makes it possible to assess the current state of field development, to predict the effectiveness of geological and technical operations (Devon News Agency wrote about this here).
AI also needs information to learn from it. And the more data there is, the faster and more efficient the learning process. To do this, the Center learned how to directly connect to digital twins, to the company’s information databases and upload all data to the neural network. For example, production rates, current reserves in areas of the development site, perforation intervals in wells, porosity, oil saturation, etc.
Additionally, software libraries were created that allow this information to be consistently converted into a readable code for the neural network. Unlike existing methods for selecting geological and technological measures at the considered development site, the program counts and issues candidate wells for geological and technical operations. It also quickly calculates the effectiveness of each of them. Due to this, it is predicted that the labor costs of the stage of selection of geological and technical measures will be reduced by 80%, and in general for projects of a comprehensive asset development plan – by 10%.
Now the Modeling Center is intensively working on calculations of hydraulic fracturing (HF), completion of formations, drilling of sidetracks and optimization of well operation modes. These are the most frequent and costly procedures, the effectiveness of which is most logical to improve in the first place. The specialists of the Modeling Center are working on training neural networks so that they can be used at all company facilities.
“Under the current conditions, a specialist looks through the entire fund and selects suitable geological and technical measures according to certain criteria,” explains Zoya LOSHCHEVA, head of the Center. – However, human capabilities are limited. And the neural network is able not only to find the required 100 or 200 wells, but to rank them on geological and hydrodynamic models according to the degree of probable success of the geological and technical operations.
A neural network is a type of machine learning. Among the main problems for scientists, there is still the question of the impossibility of creating one neural network for all tasks. A lot of work is required to integrate elements of artificial intelligence into existing processes, Tatneft notes.
For example, in the current state, traditional models have an advantage over neural networks. The models are based on the physical solution of problems that software algorithms have yet to master. In addition, there is an inevitable heterogeneity of data across different wells and fields. Where there is more information, the program learns faster. In this case, a statistical concept, better known as survivor’s error, may arise. It is also very important to think over the user interface, experts emphasize. Simply put, so that a person understands what he wanted to “say” to AI.
Earlier, Inform-Devon reported that the software systems Geoindicator and Gisneiro, created at Kazan Federal University (KFU), were tested at the fields of Tatarstan. “Geoindicator” allows you to determine the source of watering or oil inflow in production wells based on geochemical methods for studying reservoir fluids. “Gisneiro” automates the interpretation of geophysical well surveys. Kazan scientists have also developed software for oil displacement simulation.
