Full waveform inversion method and apparatus using unsupervised machine learning
AssigneeCgg Services SAS
InventorNho Hùng Dinh
Life of this patent
Estimating the term…
Abstract
A quality control method and apparatus generate maps for cycle skipping analysis of velocity models that may be the input or the output of a full waveform inversion. A comparison of observed data and corresponding synthetic data generated using a candidate velocity model yields feature images of shots. The feature images are then grouped in clusters using unsupervised machine learning. Deviation of the feature images from an ideal non-cycle skipped image allows to measure the quality of the candidate velocity model.
Claims17
A seismic data processing method for improving a velocity model of a subsurface formation by mitigating cycle skipping, the method comprising:
- obtaining observed seismic data acquired over the subsurface formation, and corresponding synthetic seismic data prepared by a processor using a candidate velocity model of the subsurface formation;
- generating feature images by calculating phase differences and amplitude differences between normalized samples of the observed seismic data and the corresponding synthetic seismic data, wherein pixels are flagged as cycle skipped when an amplitude difference exceeds a threshold and a phase reversal is detected;
- applying by the processor a k-means clustering algorithm to the feature images to generate a class map comprising k clusters and corresponding centroids thereof, wherein the k clusters include a non-cycle skipped cluster representing reference seismic behavior;
- generating a cycle-skipping risk map by calculating pixel-based differences between the feature images and a centroid of the non-cycle skipped cluster;
- updating the candidate velocity model by performing a full waveform inversion constrained by the cycle-skipping risk map to reduce cycle skipping in identified subsurface regions;
- and
- recomputing synthetic seismic wavefields using the updated velocity model to generate a physical representation of the subsurface formation for hydrocarbon exploration.
A seismic data processing apparatus configured to improve a velocity model of a subsurface formation by mitigating cycle skipping, the apparatus comprising:
- an interface configured to obtain observed seismic data acquired over the subsurface formation;
- and
- a processor connected to the interface and configured:
- to prepare synthetic seismic data corresponding to the observed seismic data, using a candidate velocity model of the subsurface formation;
- to generate feature images by calculating phase differences and amplitude differences between normalized samples of the observed seismic data and the corresponding synthetic seismic data, wherein pixels are flagged as cycle skipped when an amplitude difference exceeds a threshold and a phase reversal is detected;
- to apply a k-means clustering algorithm to the feature images and to generate a class map comprising k clusters and corresponding centroids thereof, wherein the k clusters include a non-cycle skipped cluster representing reference seismic behavior;
- to generate a cycle-skipping risk map by calculating pixel-based differences between the feature images and a centroid of the non-cycle skipped cluster;
- to update the candidate velocity model by performing a full waveform inversion constrained by the cycle-skipping risk map to reduce cycle skipping in identified subsurface regions;
- and
- to recompute synthetic seismic wavefields using the updated velocity model to generate a physical representation of the subsurface formation for hydrocarbon exploration.
Dependencies are inferred from the claim text.