Back to search

PETROMAKS2-Stort program petroleum

Pre-stack implementation of the AI-based seismic inversion algorithm with artificial constraints

Alternative title: Pre-stack implementering av den AI-baserte seismiske inversjonsalgoritmen med kunstige begrensninger

Awarded: NOK 2.0 mill.

Finding natural resources requires studying the underground, which takes months of qualified specialists’ work and many high-tech technologies. Each available technology has limitations. Integration of them and their analysis help suggest the subsurface content with better confidence. However, the risk of failure is still high. What if we could click on a “magic button” that would analyse available data and reveal the subsurface content in a minute? In 1980, we could for example only dream about some “button” that could convert subsurface-acquired seismic data to relevant colour attributes to help us understand or guess the rocks with similar properties. Today, we have software that gives us attributes in seconds. However, to this day we still do not have that “button” that lets us predict the type of rocks and fluid contained in their pores. We can only make assumptions about the possible content of the subsurface. The fundamental limitation of the available rock properties prediction lies in the first physical principles from back in 1900. Today, we can overcome this limitation with big computation resources and AI algorithms by rewriting the fundamental description of the conventional physical process. PSS-Geo aims to enable companies working within geology-related fields predict the type of subsurface rocks and fluids in minutes and not months as it is now. We expect that the results of our research will provide such solutions to others.

PSS-Geo has recently developed an AI-based algorithm to estimate accurate rock properties from post-stack seismic data. This research project aims to extend the algorithm to pre-stack seismic data for better rock properties estimation. No algorithms and methods are available in conventional physics to estimate the non-linked key rock properties of Velocity and Density from seismic data. Oil and gas exploration mainly uses these properties for further lithology and fluids prediction. All existing approaches to seismic inversions are based on these fundamental assumptions: the linearization of the Zoeppritz equation, the estimation of impedances or elastic parameters, and the extraction through regressions of linked Velocity and Density. Accurately estimating the Velocity and Density removes the gap between pure seismic structural interpolation and quantitative interpretation (QI). Using only a conventional physical approach leaves QI as supplementary with less trustable methods due to fundamental assumptions/limitations. Applying the proposed technique for high-resolution shallower seismic will significantly improve geohazard studies for seabed installations. It substantially reduces the required time and resources and achieves a high-resolution velocity field of the full frequency band. FWI is, in comparison, less efficient and can only generate results of a limited frequency band. In seabed mining (SBM), our technique will directly allow the mapping of rock properties with information on the spreading of geo bodies or the accumulation of specific minerals. The project aims to develop an AI-based pre-stack seismic inversion algorithm with a radically new solution approach. This new algorithm will benefit the estimation accuracy of rock properties and substantially reduce the man-hand influence and processing costs. Patented technology and peer-reviewed publications will be the basic principles behind this R&D project.

Funding scheme:

PETROMAKS2-Stort program petroleum