Distributed Sequential Gaussian Simulation for Large Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional Sequential Gaussian Simulation (SGS) algorithms are limited by single-machine processing, which prevents them from scaling to support large models due to memory and CPU constraints, making it difficult to estimate petrophysical properties across entire volumes efficiently.
Innovation Solution
The method involves parallelizing SGS by dividing the simulation into smaller volumes and distributing them across multiple computers in a cloud environment, allowing each computer to process smaller segments simultaneously, using techniques like overlapping bands and neural networks to maintain performance and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional Sequential Gaussian Simulation is run on a single machine, then the algorithm can maintain sequential processing accuracy, but it cannot scale to support large models due to memory and CPU limitations
Solution Approach 1:
The patent divides the large simulation model into multiple smaller segments or blocks that can be processed independently. Each block contains a portion of the total model locations, allowing parallel processing across multiple machines while maintaining the sequential gaussian simulation algorithm within each block. This segmentation enables the system to handle models with billions of locations by distributing the computational load across a cluster of computers.
2Productivity
If the simulation is parallelized across multiple computers, then processing time is reduced from hours to minutes, but the system complexity increases due to distributed computing requirements
Solution Approach 1:
The simulation domain is segmented into multiple blocks that are distributed across different computers in a cluster. Each computer processes its assigned block independently using the sequential gaussian simulation algorithm, enabling parallel execution. This segmentation strategy reduces processing time from hours to minutes while managing system complexity through modular block processing.
Solution Approach 2:
The patent transitions from single-machine sequential processing to multi-machine parallel processing by adding the dimension of spatial distribution across a computing cluster. Blocks are assigned to different machines based on their spatial locations, utilizing the network dimension to enable concurrent processing of geographically distributed model segments.
3Quantity of substance
If the entire volume is processed at once on a single machine, then data consistency is maintained, but memory constraints prevent processing of large models
Solution Approach 1:
The patent segments the large model into multiple smaller blocks, each containing a manageable number of locations that fit within the memory capacity of individual machines. This segmentation allows processing of models with billions of locations by distributing them across multiple blocks, with each block processed independently on separate machines, thereby overcoming single-machine memory constraints.
Data Source
AI summary
A method for processing a well data log may comprise adding one or more boundary areas to the well data log, dividing the well data log into one or more segments using the one or more boundary areas, processing each of the one or more segments on one or more information handling systems, and reforming each of the one or more segments into a final simulation. A system for processing a well data log may comprise one or more information handling systems in a cluster. The one or more information handling systems may be configured to perform the method for processing the well data log.