Machine Learning Surrogates for Reservoir Simulation Heterogeneity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current reservoir simulation models face challenges in accurately simulating hydrocarbon 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 techniques, specifically neural networks, to generate solution surrogates for simulating hydrocarbon reservoirs by approximating inverse operators and effective permeability values, allowing for the creation of a coarse-scale approximation of phase permeability and constitutive relationships, which can be stored and reused in subsequent simulations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a fine grid discretization is used to capture heterogeneity, then measurement precision is improved, but computational cost increases significantly
Solution Approach 1:
The reservoir is divided into multiple sub-domains or regions based on heterogeneity characteristics. Each sub-domain is processed independently with appropriate grid resolution, allowing fine grid to be applied only where necessary rather than uniformly across the entire reservoir, thus reducing overall computational cost while maintaining accuracy in critical areas.
Solution Approach 2:
Different grid resolutions and simulation approaches are applied to different regions of the reservoir based on local heterogeneity characteristics. Areas with high heterogeneity use fine grid discretization, while homogeneous areas use coarser grids, optimizing the balance between accuracy and computational efficiency.
2Use of energy by moving object
If a coarse simulation grid is used, then computational cost is reduced, but accuracy of fluid flow modeling deteriorates
Solution Approach 1:
Solution surrogates are pre-computed and stored for representative sub-domains with various boundary conditions and heterogeneity patterns. During actual simulation, these pre-computed surrogates are reused and combined to approximate solutions for complex geometries, providing accurate results at reduced computational cost by avoiding repeated full-scale simulations.
Solution Approach 2:
Solution surrogates and effective permeability values are pre-calculated and stored in databases before actual reservoir simulation. This preliminary computation allows rapid querying and combination during simulation without performing computationally expensive calculations in real-time, maintaining accuracy while improving efficiency.
3Ease of manufacture
If traditional discretization methods are used for complex geometries, then ease of manufacture is maintained, but manufacturing precision deteriorates
Solution Approach 1:
Complex reservoir geometries are segmented into multiple simpler sub-domains that can be modeled using standard discretization methods. Each sub-domain is processed independently with appropriate boundary conditions, and results are combined to obtain the overall solution, maintaining both ease of implementation and accuracy for complex geometries.
Solution Approach 2:
Solution surrogates act as intermediaries between simple analytical models and complex numerical simulations. They capture the effects of complex geometries and heterogeneity while providing computationally efficient representations that can be integrated into standard simulation workflows, bridging the gap between simplicity and accuracy.
Data Source
AI summary
There is provided a method for modeling a hydrocarbon reservoir that includes generating a reservoir model that has a plurality of sub regions. A solution surrogate is obtained for a sub region by searching a database of existing solution surrogates to obtain an approximate solution surrogate based on a comparison of physical, geometrical, or numerical parameters of the sub region with physical, geometrical, or numerical parameters associated with the existing surrogate solutions in the database. If an approximate solution surrogate does not exist in the database, the sub region is simulated using a training simulation to obtain a set of training parameters comprising state variables and boundary conditions of the sub region. A machine learning algorithm is used to obtain a new solution surrogate based on the set of training parameters. The hydrocarbon reservoir can be simulated using the solution surrogate obtained for the at least one sub region.