Seismic Waveform Inversion Using Partitioned Background Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current seismic inversion techniques face challenges in accurately modeling subsurface formations due to non-convexity in the inversion objective function, leading to local minima issues and difficulties in resolving deep targets with high-frequency absorption, which affects the resolution and accuracy of subsurface imaging.
Innovation Solution
The implementation of a full waveform inversion (FWI) method that iteratively updates a subsurface earth model by minimizing differences between observed and simulated seismic data, employing a partitioning approach to separate background and reflectivity components, and using an offset-dependent matching filter to mitigate local minima and enhance low-wavenumber updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional seismic inversion techniques are used, then the inversion process can be performed with simpler methods, but the resolution and accuracy of subsurface imaging deteriorates due to non-convexity and local minima issues
Solution Approach 1:
The patent partitions the subsurface model into two distinct components: a background model representing low-wavenumber variations and a reflectivity model representing high-wavenumber variations. This segmentation allows each component to be optimized separately, with the background model providing a stable foundation that avoids local minima traps, while the reflectivity model captures detailed subsurface features. The partitioned approach resolves the contradiction by maintaining computational simplicity through separate optimization while achieving high imaging accuracy through complementary modeling of different spatial frequencies.
2Measurement precision
If high-frequency components are emphasized to improve resolution, then the resolution of subsurface features improves, but cycle-skipping issues and local minima problems worsen
Solution Approach 1:
The patent performs preliminary construction of a background model that captures low-wavenumber subsurface variations before introducing high-frequency reflectivity components. This preliminary action establishes a stable, cycle-skipping-free foundation that guides subsequent high-frequency inversion. By preparing the background model first with appropriate low-frequency content, the method enables reliable convergence even when high-frequency components are later emphasized for improved resolution.
Solution Approach 2:
The patent employs frequency-domain optimization where the background model is optimized at lower frequencies to ensure stable convergence, while the reflectivity model is optimized at higher frequencies to capture detailed features. This parameter change strategy in the frequency domain allows the inversion to progressively build from stable low-frequency solutions to high-resolution high-frequency solutions, maintaining reliability throughout the inversion process while achieving improved resolution.
3Device complexity
If the full waveform inversion updates all frequency components simultaneously, then the computational process is simpler, but the low-wavenumber updates are insufficient to mitigate local minima
Solution Approach 1:
The patent segments the inversion objective function into two distinct optimization problems: one for the background model and one for the reflectivity model. This segmentation enables targeted optimization where the background model specifically addresses low-wavenumber updates to prevent cycle-skipping and local minima, while the reflectivity model handles high-wavenumber details. The separated optimization processes are more reliable than simultaneous updates because each can be tuned independently for its specific frequency range.
Data Source
Figure 1
Figure 2A~2D
Figure 3
AI summary
A method can include receiving seismic data of a geologic environment; receiving a background model that is a part of a partitioned model of the geologic environment; predicting reflections using the background model; determining incoherence of an offset-dependent matching filter based at least in part on the reflections and the seismic data; based at least in part on the incoherence, adjusting the background model to generate an adjusted background model; and outputting the adjusted background model.