Deep Learning Salt Geometry Detection for Subsalt Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for subsalt imaging in hydrocarbon prospect exploration, such as in the Gulf of Mexico, rely on manual iteration between full waveform inversion (FWI) and reverse time migration (RTM) to update salt geometry, which is time-consuming and uncertain, especially in areas with little visual evidence for salt boundaries.
Innovation Solution
A deep learning-based approach for salt body detection is integrated into the FWI-RTM workflow, automating the salt geometry update process by training a convolutional neural network to predict salt masks, reducing the need for manual intervention and accelerating the imaging process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual interpretation and FWI iteration are used to update salt geometry, then imaging accuracy is improved, but processing time increases significantly
Solution Approach 1:
A deep learning model is trained in advance on a dataset of seismic images with manually annotated salt masks to learn salt body detection patterns. This pre-trained model can then rapidly predict salt geometries in new seismic data without requiring iterative manual interpretation, significantly reducing processing time while maintaining accuracy
Solution Approach 2:
The deep learning model acts as an intermediary between seismic data acquisition and final imaging interpretation. It automatically generates salt body predictions that guide the FWI process, replacing the need for continuous manual interpretation loops and accelerating the workflow
2Reliability
If unlimited iterations of FWI are performed to gradually shape salt geometry, then imaging quality is improved, but computational cost becomes impractical
Solution Approach 1:
The deep learning model performs salt body detection on a subset of seismic data or at key stages of the imaging workflow, rather than requiring complete iterative FWI loops. This partial automation provides sufficient salt geometry information to guide imaging without the need for exhaustive iterations, balancing quality and efficiency
3Measurement precision
If manual picking of salt horizons is performed after each FWI loop, then salt geometry is updated, but uncertainty remains in areas with little visual evidence
Solution Approach 1:
The deep learning model is trained on diverse seismic data encompassing various salt body types, geological structures, and imaging conditions. This universal training enables the model to generalize and make reliable predictions even in areas with poor visual evidence, such as salt flanks and base boundaries, where manual interpretation is highly uncertain
Data Source
AI summary
A method includes receiving seismic data and an initial velocity model, generating a first seismic image based at least in part on the seismic data and the initial velocity model, training a machine learning model to predict salt masks based at least in part on seismic images, merging the initial velocity model and the first salt mask to generate a first modified velocity model, generating an updated velocity model based at least in part on the first modified velocity model, generating a second seismic image based at least in part on the updated velocity model, predicting a second salt mask based at least in part on the second seismic image and the updated velocity model, using the trained machine learning model, and merging the updated velocity model and the second salt mask to generate a second modified velocity model.