Parallel Multilevel Simulation for Reservoir Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current reservoir simulation methods face challenges in efficiently solving sparse matrix equations on parallel computers due to uncontrollable growth in connections between unknowns, leading to increased communication costs and reduced scalability.
Innovation Solution
A method that generates a data representation with interconnection weights, sets weights below a threshold to zero, and iteratively creates lower-level matrices to restrict the solution, reducing the number of connections and eliminating unconnected cells, thereby generating a preconditioner for Krylov-type iterative methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If parallel computing systems are used to solve reservoir simulation equations, then simulation speed and productivity are improved, but the number of connections between unknowns grows uncontrollably, leading to increased communication costs and reduced scalability
Solution Approach 1:
The patent applies segmentation by dividing the computational domain into subdomains that can be processed in parallel, while using multigrid techniques to coarsen the grid hierarchy. This segmentation allows the simulation to be distributed across multiple processors, improving productivity while the multigrid coarsening controls the growth of connections by creating a hierarchical structure where coarse levels have fewer inter-processor dependencies.
Solution Approach 2:
The patent introduces an additional dimension by implementing a multilevel hierarchical structure with coarse and fine grid levels. This dimensional addition allows the system to manage connections more efficiently by solving problems at multiple scales, where coarse levels handle global communication and fine levels handle local detail, thus controlling connection growth while maintaining parallel computing benefits.
2Adaptability or versatility
If the number of connections between unknowns is reduced to control communication costs, then scalability is improved, but the accuracy of the simulation model may be compromised
Solution Approach 1:
The multigrid method segments the problem into different resolution levels, allowing accurate representation of local features on fine grids while using coarser grids for global behavior. This segmentation maintains simulation accuracy by preserving essential physics at appropriate scales while reducing the overall number of connections that require communication across processors.
Solution Approach 2:
By adding the multilevel hierarchical dimension, the patent enables accurate simulation with reduced connections. The coarse-fine level structure allows the system to capture both global reservoir behavior and local flow details without requiring all possible connections to be explicitly represented, thus improving scalability while maintaining accuracy.
3Loss of time
If global communication is reduced in parallel computing, then communication costs and computation time are decreased, but the ability to solve complex reservoir models is limited
Solution Approach 1:
The patent segments the computational work into independent tasks that can be executed with minimal global communication. By dividing the reservoir model into subdomains and using multigrid coarsening, the system reduces the frequency and volume of global communication events while maintaining the ability to solve complex models through localized computation and hierarchical coordination.
Solution Approach 2:
The multilevel hierarchical dimension enables the system to solve complex reservoir models with reduced global communication. Coarse levels provide global context with minimal communication, while fine levels handle detailed local computations independently, allowing complex models to be solved efficiently by distributing work across the hierarchical structure rather than requiring constant global synchronization.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Exemplary embodiments of the present techniques provide methods and systems for coarsening a computational mesh, for example, for use in a reservoir simulation. An exemplary method of performing a reservoir simulation, includes generating a data representation in a storage system, wherein the data representation includes an interconnection weight that represents the magnitude of an interconnection between each of a number of computational cells in a computational mesh. A threshold value is compared to each interconnection weight and any interconnection weight that is equal to or less than the threshold value is set to zero.