Domain Decomposition for High-Frequency Elastic Full Waveform Inversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current elastic Full Waveform Inversion (FWI) methods face significant memory limitations when processing seismic wave propagation, making it challenging to perform high-frequency simulations on a single Graphics Processing Unit (GPU) due to high memory requirements, which restricts the accuracy and resolution of subsurface imaging.
Innovation Solution
Implementing a method that utilizes domain decomposition across multiple GPUs, employing an implicit finite-difference approach to reduce memory requirements and data exchange overhead, allowing for parallel processing and overcoming the limitations of a single GPU card by splitting the subsurface offset domain into subdomains and optimizing stencil size and data exchange.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If elastic Full Waveform Inversion (FWI) is performed with high-frequency simulations on a single GPU, then the accuracy and resolution of subsurface imaging is improved, but the memory requirements become prohibitively high
Solution Approach 1:
The patent divides the computational domain into multiple subdomains that can be processed in parallel across multiple GPUs. The subsurface offset domain is segmented such that each GPU handles a specific portion, reducing the memory burden on any single device while maintaining high-frequency simulation capabilities. This segmentation allows the system to achieve high resolution imaging without requiring excessive memory on a single GPU.
2Quantity of substance
If domain decomposition is implemented across multiple GPUs, then memory requirements are reduced and parallel processing is enabled, but data exchange overhead increases
Solution Approach 1:
The patent optimizes the domain decomposition strategy by carefully controlling the data exchange interfaces between subdomains. Only the necessary boundary data is exchanged between GPUs, minimizing communication overhead. The stencil size and data exchange regions are optimized to reduce the amount of data that needs to be transferred while maintaining computational accuracy, thus reducing the energy loss associated with data exchange.
3Productivity
If the subsurface offset domain is split into subdomains for parallel processing, then processing capability is improved, but data exchange costs increase
Solution Approach 1:
The computational domain is segmented into multiple subdomains that can be processed in parallel, improving overall processing capability. The patent carefully designs the segmentation strategy to minimize the surface area of interfaces between subdomains, thereby reducing the data exchange costs while maintaining high parallel processing efficiency.
4Measurement precision
If high-frequency simulations are performed, then subsurface imaging resolution is improved, but memory requirements exceed single GPU capacity
Solution Approach 1:
The patent implements domain decomposition that divides the high-frequency simulation workload across multiple GPUs. Each GPU processes a specific subdomain with reduced memory requirements, while the collective system maintains the high resolution capabilities needed for accurate subsurface imaging. This approach enables high-frequency simulations without requiring any single GPU to have excessive memory capacity.
Data Source
AI summary
A method for imaging a formation of a subsurface includes receiving input data d related to the subsurface, generating synthetic data u related to the subsurface, by applying an implicit finite-difference approach to a reflectivity model r, updating a velocity model V based on the input data d and the synthetic data u, and generating an image of the formation in the subsurface based on the updated velocity model, wherein the formation is used to locate natural resources.


