GPU-Accelerated Reservoir Simulation Linear Solvers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current compositional reservoir simulators face challenges in accurately simulating giant subsurface reservoirs due to the sheer volume of data and complexity, requiring faster and more accurate linear solution computations to handle thousands of time-steps and millions to billions of grid blocks, which is not efficiently managed by conventional processors.
Innovation Solution
The use of a computer system with a central processing unit (CPU) and a graphical processing unit (GPU) for preconditioned conjugate-gradient extrapolation to solve linearized systems of equations, optimizing computations by partitioning reservoir models into smaller tasks and utilizing multi-threaded GPU architecture for parallel processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional processors are used to solve linearized systems of equations in compositional reservoir simulation, then the simulation can be performed, but the processing time becomes excessively long and cannot keep up with real-time data acquisition rates
Solution Approach 1:
The patent replaces conventional CPU-based linear solvers with GPU-accelerated solvers. The GPU performs preconditioned conjugate-gradient extrapolation to solve large linearized systems of equations, achieving near-order-of-magnitude speedup compared to traditional CPU-based methods. This substitution of computational architecture directly addresses the processing time bottleneck.
Solution Approach 2:
The patent segments the reservoir model into multiple grid blocks and partitions the linear system into smaller sub-systems that can be processed in parallel on different GPU cores. This segmentation enables concurrent computation of multiple reservoir zones, significantly reducing total processing time while maintaining simulation accuracy.
2Measurement precision
If the reservoir model is divided into more grid blocks to increase spatial resolution and accuracy, then measurement precision improves, but the number of unknowns and computational complexity increases exponentially
Solution Approach 1:
The patent changes the computational parameters by using GPU architecture with its specific memory hierarchy and parallel processing capabilities. This allows efficient handling of large-scale linear systems with hundreds of millions to billions of unknowns through optimized memory access patterns and parallel matrix operations, making high-resolution simulations computationally feasible.
Solution Approach 2:
The patent introduces a preconditioner as an intermediary component in the conjugate-gradient extrapolation process. This preconditioner transforms the difficult-to-solve linear system into an equivalent system that converges faster, reducing the number of iterations required and making large-scale simulations more manageable computationally.
3Measurement precision
If compositional simulation includes more fluid components and inorganic substances for enhanced accuracy, then the simulation accuracy improves, but the number of unknowns per grid block increases
Solution Approach 1:
The patent transitions from sequential CPU processing to parallel GPU processing, adding a spatial dimension of parallelism. Multiple fluid components and inorganic substances are processed simultaneously across thousands of GPU cores, allowing the system to handle increased numbers of unknowns without proportional increases in wall-clock time.
Data Source
AI summary
A computer-based system performs iterative linear solution of giant systems of linear equations with the computational acceleration capabilities of GPU's (Graphical Processing Units). Processing is performed in a heterogeneous (hybrid) computer environment composed of both computer data processing units (CPU's) and GPU's. The computational acceleration in processing provides an order of magnitude speed improvement over other methodology which utilizes only CPU's. The present invention enables reservoir studies to be carried out within time constraints, and real-time reservoir simulations to be made while keeping pace with online data acquisition.