Machine Learning Surrogate for Reservoir Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current reservoir simulation models face challenges in accurately simulating fluid flow through heterogeneous porous media due to their inability to efficiently handle complex geometries and heterogeneity, leading to inaccurate results and high computational costs.
Innovation Solution
The use of machine learning algorithms, specifically neural networks, to approximate the inverse operator of a matrix equation that models fluid flow through porous media, allowing for the creation of a solution surrogate that can be reused across different sub-regions of a hydrocarbon reservoir, thereby improving simulation accuracy and reducing computational expense.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a fine grid discretization is used to capture heterogeneity in porous media, then measurement precision of physical properties is improved, but computational cost increases significantly
Solution Approach 1:
The domain is divided into coarse grid cells, each containing heterogeneous fine-grid structures. This segmentation allows the fine-grid heterogeneity to be captured within representative elementary volumes (REV) while using a coarse overall grid, thus reducing total computational cost while maintaining accuracy for capturing heterogeneity effects.
Solution Approach 2:
Instead of simulating the entire fine-grid model, the patent creates simplified copy representations of the fine-grid heterogeneity within each coarse grid cell using REV models. These copies capture the essential heterogeneity characteristics without requiring full fine-grid resolution, significantly reducing computational cost while preserving accuracy for heterogeneity effects.
2Productivity
If a coarse simulation grid is used to reduce computational expense, then productivity is improved, but measurement precision of physical properties deteriorates
Solution Approach 1:
The patent applies different levels of detail to different parts of the model: coarse-grid regions use simplified REV representations of heterogeneity, while fine-grid regions maintain full resolution. This local quality approach allows accurate heterogeneity capture where needed while using coarser resolution elsewhere, improving overall simulation speed without sacrificing necessary precision.
Solution Approach 2:
The patent introduces a new dimension of modeling by incorporating Representative Elementary Volume (REV) concepts that bridge the scale between coarse simulation grids and fine heterogeneity structures. This dimensional approach allows the model to represent sub-grid heterogeneity effects without requiring fine-grid resolution, thus maintaining measurement precision while improving productivity.
3Ease of manufacture
If TPFA discretization is used for orthogonal grids with isotropic permeability, then ease of manufacture is improved, but manufacturing precision of fluid velocity calculations deteriorates when permeability is anisotropic or grid is non-orthogonal
Solution Approach 1:
The patent transforms the discretization approach by incorporating REV-based effective properties that account for anisotropic permeability and non-orthogonal grid effects. By changing the parameters used in the discretization (using REV-averaged properties instead of simple point values), the method maintains TPFA simplicity while improving velocity calculation accuracy for complex permeability structures and grid configurations.
4Manufacturing precision
If MPFA or Finite Element Methods are used to address complex geometries, then manufacturing precision of fluid velocity is improved, but device complexity and computational cost increase
Solution Approach 1:
The patent uses inexpensive REV-based effective property calculations instead of complex MPFA or Finite Element discretizations. These simplified REV models act as disposable approximations that capture the essential physics of complex geometries and anisotropic permeability without requiring the computational overhead and implementation complexity of advanced discretization methods.
Data Source
AI summary
There is provided a method for modeling a hydrocarbon reservoir that includes generating a reservoir model comprising a plurality of sub regions. At least one of the sub regions is simulated using a training simulation to obtain a set of training parameters comprising state variables and boundary conditions of the at least one sub region. A machine learning algorithm is used to approximate, based on the set of training parameters, an inverse operator of a matrix equation that provides a solution to fluid flow through a porous media. The hydrocarbon reservoir can be simulated using the inverse operator approximated for the at least one sub region. The method also includes generating a data representation of a physical hydrocarbon reservoir can be generated in a non-transitory, computer-readable, medium based, at least in part, on the results of the simulation.