Page 8 - Lime Petrolium Annual Report 2020

LIME PETROLEUM
“LL Merge” – Method and Algorithms for post-stack Inversion
Lime Petroleum AS is participating and sponsoring a R&D project along with Innovation Norge to develop methods to predict lithology and fluids from well and seismic data using Artificial Intelligence (AI) and Machine Learning (ML) techniques. AGGS AS, PSS-Geo AS and Cama Geosciences AS, independent professional geological and geophysical consulting firms, are working as a consortium combining seismic big-data and well data with AI and ML processes integrated in a geological context.
Vintage seismic 3D data cover large areas in mature petroleum provinces. On the Norwegian shelf most of the data is part of the public domain and has on several occasions been merged into large 3D surveys. While these surveys are great tools for regional geological analysis, they also represent an enormous amount of data that can be used in big-data analysis leading to new understandings.
This R&D project concentrates on developing a method and algorithms for post-stack inversion to separate Vp and seismic density from reprocessed and merged public 3D seismic. In a second part of the project Machine Learning algorithms are used to predict pore fluids by combining the previously inverted seismic with well data.
The result will be a local or regional seismic inversion cube premediated for fluid and lithology predictions. The reprocessed data is expected to have a resolution comparable to newly acquired seismic, and with the possibility for stratigraphy and fluid analysis gain detailed insight in the deposition systems
and fluid migration. This data will give explorationists a new regional tool to access prospectivity on a local and regional basis.
The project workflow is divided into two parts. For the first part Artificial Intelligence inversion (annealing global optimi- zation) is used to estimate P- velocity and density from post- stack seismic data. Further, the inverted volumes are used to compute Volume of Clay (Vcl), Porosity (Phi) and Reservoir Quality ((1-Vcl) *Phi). This method has been tested on seismic data of the Cretaceous turbiditic plays in Mid-Norway (60 000 km2) and proved highly successful.
In the second part, fluid detection from seismic data is investigated. Since the post-stack mega merged seismic does not contain S-wave related information, the possibilities of using ML techniques with well data as training input to predict fluid and specific fluid type is investigated. The basic principle of such automation is to find relationships among attributes for identifying fluid. Well-logs have been used as input for training, testing and validation.
PAGE 8