Ocean Framework, C#, .NET, Python (AI models)
Reservoir numerical simulation typically operates at two distinct scales. High-resolution geological models capture the heterogeneity of the porous medium with fidelity, while reservoir simulation grids with coarser discretization make execution times compatible with operational workflows and uncertainty studies. Scale transfer (ST) is the procedure that converts petrophysical properties, such as porosity and permeability, from the fine grid to the coarse grid, condensing multiple values into a single representative value per cell.
In homogeneous reservoirs, this procedure is relatively well behaved. In highly heterogeneous reservoirs, such as the Brazilian pre-salt carbonates, scale transfer becomes one of the greatest challenges in reservoir simulation: reducing resolution can compromise information essential to the quality of the results, leading to less reliable production forecasts.
To mitigate this information loss, the UNISIM/CEPETRO group at Unicamp developed scale transfer methodologies tailored to carbonate reservoirs, based on reservoir sectorization and on the application of distinct ST formulations per region, complemented by clustering techniques, property-specific methods, and the adjustment of relative permeability pseudo-curves.
The project's challenge was twofold: to bring this academic methodology into an operational setting for the Oil & Gas industry, while incorporating artificial intelligence into the process, expanding its applicability and making its computational cost competitive against the direct use of high-fidelity models. All of this integrated into Petrel, the platform where geological and simulation models actually live.
DeepSoft developed DeepShift Petrel, a Petrel plug-in that connects the DeepShift software to the platform's native environment. DeepShift is the product resulting from the R&D project led by Solpe, Repsol Sinopec Brasil, and Unicamp, materializing in code the AI-assisted scale transfer methodology proposed by the UNISIM/CEPETRO group.
Built on the Ocean Framework, the plug-in allows reservoir engineers to execute the complete scale transfer workflow directly from Petrel: selecting the high-resolution geological model, defining the target coarse grid, configuring the sectorization and clustering strategies and the per-property ST methods, running DeepShift as the computation engine, and receiving the results as native properties in the Petrel project.
The coupling with artificial intelligence enters at the stages where the traditional methodology requires expensive or case-by-case decisions, such as identifying regions with similar behavior, choosing ST formulations suited to each cluster, and adjusting relative permeability pseudo-curves, enabling the methodology to scale to real industry scenarios.
The scale transfer methodology developed by UNISIM/CEPETRO became available as a computational tool integrated into Petrel, enabling its use by technical teams in real workflows at operating companies.
The inclusion of artificial intelligence at critical stages of the process, such as sectorization, clustering, and pseudo-curve adjustment, expanded the applicability of the methodology in highly heterogeneous carbonate reservoirs.
The entire scale transfer cycle, from selecting the fine model to inspecting the resulting properties on the coarse grid, runs inside Petrel, eliminating manual exports and preserving traceability between the geological model and the simulation model.
DeepShift Petrel is the direct result of an R&D project carried out in partnership with Repsol Sinopec Brasil and Unicamp, reinforcing DeepSoft's role as a strategic executor that turns academic research into software applicable to industry.