Stochastic Inversion Dimensionality Reduction Seismic Exploration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Stochastic inversion methods for seismic data analysis, particularly Stein-Variational Gradient Descent (SVGD), face significant computational costs and complexity due to high-dimensional parameter spaces, limiting their applicability in real-time and large-scale seismic exploration applications.
Innovation Solution
The method involves compressing initial subsurface models using dimensionality reduction techniques such as image segmentation and geometric data compression, employing B-spline curves and control points, to reduce the number of dimensions and models required for sampling, thereby decreasing computational demands and maintaining accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If stochastic inversion methods are applied to seismic data analysis, then subsurface property estimation accuracy is improved, but computational cost increases significantly
Solution Approach 1:
The patent applies image segmentation to divide the subsurface model into distinct regions or layers, allowing the inversion process to focus on specific segments rather than processing the entire high-dimensional model simultaneously. This segmentation reduces the computational burden while maintaining estimation accuracy for each segment.
Solution Approach 2:
The patent introduces an intermediate compressed representation of the subsurface model that serves as a mediator between the original high-dimensional model and the inversion process. This intermediate representation reduces dimensionality while preserving essential features, thereby lowering computational cost without significantly compromising accuracy.
2Manufacturing precision
If high-dimensional subsurface models are used, then model detail and accuracy are improved, but computational complexity increases
Solution Approach 1:
The patent extracts and removes redundant or less important dimensions from the high-dimensional subsurface model, keeping only the essential features needed for accurate inversion. This extraction process reduces computational complexity while maintaining the critical detail required for precise subsurface property estimation.
Solution Approach 2:
The patent transforms the high-dimensional model into a lower-dimensional representation by changing the dimensional structure. This dimensionality reduction is achieved through compression techniques that preserve the essential information in fewer dimensions, thereby reducing computational complexity while maintaining model accuracy.
3Reliability
If the number of particles is increased, then sampling quality is improved, but computational time increases
Solution Approach 1:
The patent changes key parameters of the particle representation by compressing them into a lower-dimensional space. This parameter transformation allows for efficient manipulation and updating of particles, enabling high sampling quality with fewer particles and reduced computational time compared to working with full-dimensional particle representations.
Data Source
AI summary
A method includes generating an initial subsurface model having an initial dimensionality and based at least in part on initial seismic data, compressing the initial model subsurface model to reduce a dimensionality of the initial subsurface model and form a compressed subsurface model having a compressed dimensionality that is less than the initial dimensionality, producing an initial plurality of particles from the compressed subsurface model at the compressed dimensionality, selecting particles from the initial plurality of particles, expanding the selected particles to return the selected particles to the initial dimensionality, iteratively updating a value of each particle of the selected particles utilizing synthetic seismic data produced from the initial subsurface model to generate a posterior set of particles, and outputting the posterior set of particles as a target distribution.