Offset Continuation Full Wavefield Inversion for Poor Velocity Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Full Wavefield Inversion (FWI) methods struggle to update long wavelength anomalies in subsurface velocity models, often getting stuck in local minima due to poor starting models, especially when the kinematic differences between simulated and observed data exceed half the dominant wavelength, leading to inaccurate subsurface imaging.
Innovation Solution
The method employs an offset continuation approach combined with scale separation, using a Gaussian smoothing operator to constrain velocity model updates as a function of depth and offset, and re-parametrizing the starting velocity model to prevent false updates, allowing incremental expansion of maximum offset in each stage of inversion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional FWI is used with a poor starting model, then computational complexity is reduced, but the inversion gets stuck in local minima due to cycle skipping
Solution Approach 1:
The method applies preliminary smoothing to the starting velocity model before inversion, and performs preliminary velocity analysis to establish a better initial model. This preliminary preparation prevents cycle skipping during the inversion process by ensuring the starting model is sufficiently accurate, thereby improving convergence reliability without excessive computational cost.
Solution Approach 2:
The inversion process is segmented into multiple stages with increasing frequency content. The method first inverts for low-frequency components to establish the long-wavelength velocity structure, then progressively adds higher frequency components. This segmentation allows the inversion to converge reliably at each stage before moving to the next, avoiding local minima traps.
2Productivity
If conventional FWI updates all wavelength components simultaneously, then inversion speed is improved, but long wavelength anomalies cannot be recovered when starting model is poor
Solution Approach 1:
The velocity model updates are segmented by wavelength components. The method uses a smoothing operator that selectively updates long-wavelength components based on the starting model quality, while allowing short-wavelength updates only when the long-wavelength structure is sufficiently established. This segmentation ensures accurate recovery of long wavelength anomalies while maintaining overall inversion efficiency.
Solution Approach 2:
The smoothing operator applied to velocity updates is dynamic and adapts based on the current iteration stage and starting model quality. Early in the inversion or when starting models are poor, stronger smoothing is applied to preserve long-wavelength information. As the inversion progresses and the model improves, the smoothing is reduced to allow finer wavelength updates, thus dynamically balancing precision and speed.
3Measurement precision
If strong smoothing is applied to velocity updates, then long wavelength components are preserved, but short wavelength details are lost
Solution Approach 1:
The smoothing operator strength is dynamically adjusted throughout the inversion process. In early iterations or when dealing with poor starting models, stronger smoothing is applied to ensure accurate long-wavelength velocity structure. As the inversion converges and the velocity model improves, the smoothing strength is progressively reduced, allowing short-wavelength velocity variations to be recovered. This dynamic adjustment preserves long-wavelength accuracy while minimizing information loss at shorter wavelengths.
Solution Approach 2:
The inversion process periodically re-evaluates the starting model quality and adjusts the smoothing operator accordingly. At certain intervals, the method may reapply velocity analysis or adjust smoothing parameters based on the current state of the velocity model, ensuring that long-wavelength accuracy is maintained while gradually recovering short-wavelength details as the model improves.
Data Source
AI summary
A computer-implemented method for updating subsurface models including: using an offset continuation approach to update the model, and at each stage defining a new objective function where a maximum offset for each stage is set, wherein the approach includes, performing a first stage iterative full wavefield inversion with near offset data, as the maximum offset, to obtain velocity and density or impedance models, performing subsequent stages of iterative full wavefield inversion, each generating updated models, relative to a previous stage, wherein the subsequent stages include incrementally expanding the maximum offset until ending at a full offset, wherein a last of the stages yields finally updated models, the subsequent stages use the updated models as starting models, and the full wavefield inversions include constraining scales of the velocity model updates at each stage of inversion as a function of velocity resolution; and using the finally updated models to prospect for hydrocarbons.