Parallel Adaptive Data Partitioning for Reservoir Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing reservoir simulation methods face inefficiencies due to improper load balancing and increased computational costs as fluid properties change, leading to suboptimal parallel performance and iterative convergence issues.
Innovation Solution
The method involves generating a reservoir model, partitioning it into domains based on processing costs and connectivity levels, and dynamically re-partitioning to optimize parallel processing, using techniques such as streamline tracing and node weighting to distribute workload efficiently across multiple processing elements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If the reservoir model is partitioned into fixed domains, then the parallel processing structure is simple, but the load balance deteriorates as fluid properties change during simulation
Solution Approach 1:
The patent implements dynamic repartitioning of the reservoir model during parallel simulation by detecting changes in fluid properties and re-distributing grid blocks across processing elements. This allows the partitioning structure to adapt to changing computational workloads, maintaining optimal load balance without requiring a completely complex fixed structure
Solution Approach 2:
The system changes partitioning parameters based on simulation progress by monitoring fluid property changes and adjusting domain boundaries accordingly. This allows the same basic partitioning framework to maintain efficiency through parameter adaptation rather than structural redesign
2Productivity
If dynamic repartitioning is implemented to maintain load balance, then parallel processing efficiency is improved, but computational overhead and complexity increase
Solution Approach 1:
The patent employs feedback mechanisms by monitoring fluid property changes during simulation and using this information to trigger repartitioning operations. This feedback loop ensures repartitioning occurs only when necessary, balancing efficiency improvement against computational overhead
Solution Approach 2:
The system performs preliminary detection of fluid property changes before initiating repartitioning, allowing it to anticipate when load balance will deteriorate and prepare optimal partitioning strategies in advance, reducing the overall complexity of real-time decision-making
3Device complexity
If nodes are partitioned without considering processing costs, then the partitioning algorithm is simple, but load balance deteriorates leading to iterative convergence issues
Solution Approach 1:
The patent applies local quality by assigning different weights to different nodes based on their specific processing costs, which are determined by local fluid properties and connectivity characteristics. This ensures that each region is partitioned according to its unique computational requirements, improving convergence reliability without requiring a uniformly complex algorithm
4Loss of energy
If the reservoir model is not repartitioned during simulation, then computational overhead is low, but load balance deteriorates as fluid properties change
Solution Approach 1:
The patent implements periodic repartitioning by monitoring fluid property changes at defined intervals or triggers during simulation. This periodic adjustment maintains load balance quality without requiring continuous repartitioning, thereby limiting computational overhead while preserving productivity
Data Source
AI summary
A computer implemented system and method for parallel adaptive data partitioning on a reservoir simulation using an unstructured grid includes a method of simulating a reservoir model which includes generating the reservoir model. The generated reservoir model is partitioned into multiple sets of different domains, each one corresponding to an efficient partition for a specific portion of the model.