Translation. Region: Russian Federal
Source: Novosibirsk State University – Novosibirsk State University –
An effective software module for seismic data analysis was developed by a student Faculty of Geology and Geophysics, Novosibirsk University Vladislav Korchuganov under the supervision of NSU Associate Professor Anton Duchkov. The module includes seismic acoustic and synchronous inversion procedures, as well as lithoclassification using machine learning. The uniqueness of the development is that the module is able to automatically find all the necessary parameters, completely eliminating manual adjustment by a specialist. There is no similar comprehensive solution today either on the Russian or foreign market. Despite the active work of several research groups, none of them has yet managed to achieve full automation. The young researcher described his development in his master’s thesis on “Improving the efficiency of volumetric lithotype forecasting based on the results of synchronous amplitude inversion.”
Seismic exploration is a method of exploration geophysics that uses artificially excited elastic waves to study the geological structure of the Earth. This method is used to search for oil and gas traps at depths of up to several kilometers. For oil and gas industry purposes, seismic exploration is based on reflected waves, that is, those waves that are reflected from acoustically contrasting boundaries in the rock mass are studied.
The waves are recorded by special sensors, after which the obtained data undergoes a series of processing and interpretation procedures. The result is a volumetric model of the studied subsurface area, on the basis of which conclusions can be drawn about the geological structure and the presence of promising objects in terms of oil and gas content.
If you imagine such data visually, then for most deposits they look like a “layered pie,” in which each layer is a sedimentary rock approximately 50–100 meters thick, formed over millions of years. The geologist’s task is to find in this “pie” those layers that contain oil and gas.
— From a technical point of view, seismic exploration data is a three-dimensional array consisting of billions of individual points. The volume of such an array (in the industry they are called “seismic cubes”) can easily exceed 15-20 GB. Obviously, working with such large data requires serious IT competence. Currently, the domestic market of Russia is actively developing projects to develop software packages for industrial interpretation of seismic exploration data. Companies are investing heavily to replace imported systems that have become the industry standard. One of such projects is a new generation of software developed by NSU jointly with an industrial partner. My qualification work arose from the need to implement a number of procedures for this software package. In it, I implemented seismic cube inversion procedures. To put it simply: the original seismic data can answer the question “where exactly are the layers?”, but do not allow you to immediately understand “what exactly is contained in these layers?”. Usually, this is done by a geologist, collecting and carefully analyzing a lot of additional information. My algorithms make it possible to partially automate this process by combining data from wells with seismic cubes, which makes it possible to understand more quickly and accurately what exactly is hidden in the subsoil, said Vladislav Korchuganov.
The young researcher joined the team of new generation software developers three years ago. At first, he studied programming and the basics of seismic exploration, after which he fully joined the team. During his master’s degree, Vladislav Korchuganov conducted research aimed at prototyping a software module that became part of the overall development. He had to start “on paper”, using specialized literature, since there were no available software packages on the domestic market that implemented these procedures.
In addition to the basic implementation of procedures, Vladislav Korchuganov decided to optimize their execution: he applied parallelization and preconditioning procedures for the task in a sparse form, which allowed him to speed up the calculations many times over. In his work, the young researcher applied machine learning methods: in particular, classification algorithms for unbalanced data.
— All of the above innovations make my implementation stand out from the solutions available on the market. Machine learning algorithms were used to automate the interpretation of inversion results. As a result, the code I implemented in Python was translated into C by the development team and integrated into the overall structure of the software package, — explained Vladislav Korchuganov.
During industrial tests, the software module demonstrated its high efficiency: synchronous inversion for real data from the Orenburg Region field allowed achieving high convergence of well and calculated elastic properties. The use of the developed classification scheme allowed increasing the key metrics of the “collector” class forecast by three times for the area under study.
— At this stage, our team of developers has closed the basic functionality required by the oil and gas industry. Next, we plan to implement advanced algorithms for interpreting seismic exploration data, such as Ji-Fi inversion, geostatistical inversion, etc. Our own developments in this industry will also be implemented, — said Vladislav Korchuganov.
Please note: This information is raw content directly from the source of the information. It is exactly what the source states and does not reflect the position of MIL-OSI or its clients.