Diffusion Model Subsurface Formation Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning approaches for reconstructing subsurface formations from seismic data face challenges due to the non-unique and ill-posed nature of the inverse problem, leading to uncertainty in accurately mapping subsurface formation properties to seismic images.
Innovation Solution
A diffusion model is employed as a generative model to quantify uncertainty, training on labeled seismic image samples to estimate noise and iteratively update itself, allowing for accurate inference of reservoir parameters such as salt content, faults, porosity, and permeability by removing noise from input seismic images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional machine learning approaches are used to directly map subsurface formation properties to seismic images, then the reconstruction process can be performed, but the mapping is non-unique and leads to uncertainty due to the ill-posed nature of the inverse problem
Solution Approach 1:
The diffusion model performs preliminary actions by iteratively removing noise from seismic images before final reconstruction. The model progressively denoises the input seismic data through multiple diffusion steps, establishing a clean intermediate representation that resolves ambiguities before the final subsurface property mapping is completed
Solution Approach 2:
The diffusion model acts as an intermediary between raw seismic images and subsurface formation properties. It introduces a probabilistic intermediate representation that captures the uncertainty inherent in the inverse problem, allowing the system to explore multiple plausible reconstructions and select the most probable solution
2Measurement precision
If a diffusion model is used to quantify uncertainty and iteratively remove noise from seismic images, then the accuracy of geological inferences is improved, but the computational complexity and processing time increase
Solution Approach 1:
The diffusion model segments the reconstruction process into multiple discrete diffusion steps, where each step removes a portion of the noise. This segmentation allows the complex inverse problem to be broken down into manageable iterations, with each step contributing to the final accurate reconstruction while maintaining computational tractability
Solution Approach 2:
The diffusion model employs periodic iterative action by repeatedly applying noise removal operations in cycles. Each iteration refines the reconstruction by eliminating additional noise components, and the periodic nature of these iterations allows the system to progressively converge to the accurate geological inference while managing computational load through structured repetition
Data Source
AI summary
Some implementations include a method for controlling a computer to geologically characterize a space relative to a borehole. The method may include configuring a diffusion process applied to information and data about samples of reservoir parameters. The method also may include determining, via the diffusion process, a probability distribution of the reservoir parameters in the space relative to the borehole.