Reservoir Turning Bands Simulation Distributed Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The turning bands method for geostatistical reservoir simulation faces a computational bottleneck as the number of points in the reservoir model scales into the billions, making it prohibitively slow to run on a single machine, and existing methods do not efficiently handle the distribution of computing resources for parallel processing.
Innovation Solution
A distributed computing system is employed to parallelize certain steps of the turning bands method, where a client application specifies simulation parameters and sends data to a distributed computing system, which splits the reservoir model into tiles and delegates sub-nodes to run unconditional and conditional simulations independently, allowing for significant reduction in computational time by leveraging the parallelizability of independent distribution sampling and interpolation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the turning bands method is run on a single machine, then the simulation maintains accuracy and correctness, but the computational time becomes prohibitively slow when the number of points scales into the billions
Solution Approach 1:
The reservoir model is divided into multiple independent tiles, each of which can be processed separately. The turning bands simulation is partitioned across multiple computing nodes, with each node handling specific tiles or subsets of the model. This segmentation allows parallel processing while maintaining the accuracy of the simulation through independent computation of each segment.
2Area of stationary object
If the reservoir model size is increased to capture larger formation areas, then the model completeness improves, but the computational complexity and time required for simulation increases significantly
Solution Approach 1:
Large reservoir models are divided into smaller tiles that can be processed independently across multiple computing nodes. This reduces the computational complexity of each individual processing unit while maintaining the ability to model large formation areas through aggregation of multiple tile results.
Solution Approach 2:
The problem is transformed from a single-machine sequential processing model to a multi-node parallel processing architecture. By adding the dimension of distributed computing across multiple machines, the system can handle larger reservoir models without proportionally increasing the computational complexity of individual processing units.
3Device complexity
If traditional single-machine computing is used, then the system architecture remains simple, but the productivity and output rate of reservoir model generation is limited
Solution Approach 1:
The reservoir modeling task is segmented into independent units (tiles) that can be processed in parallel across multiple computing nodes. This segmentation enables the system to achieve high productivity through parallel processing while maintaining manageable system architecture through standardized tile-based processing units.
Solution Approach 2:
The distributed computing system uses universal computing nodes that can each independently process multiple tiles. This multi-functionality allows the system to scale productivity by adding more nodes while maintaining a relatively simple architecture where each node performs the same standardized operations.
Data Source
AI summary
A reservoir model for values of a formation property is simulated using a turning bands method with distributed computing. A distributed computing system simulates the reservoir on separate machines in parallel in several stages. First, line distributions are simulated independently on turning bands. The reservoir model is partitioned into tiles and unconditional simulations are run on each tile in parallel using the corresponding simulated turning bands. The unconditional simulations within each tile are conditioned on known formation values to generate conditional simulations. Conditional simulations are aggregated across tiles to create the simulated reservoir model.