Automated Velocity Model Updating via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual identification of velocity boundaries in seismic data migration is time-consuming and inefficient due to the reliance on human operators for data-intensive tasks, limiting the speed and accuracy of subsurface exploration.
Innovation Solution
A method using a computer processor to automatically determine velocity boundaries by generating seismic migration gathers, selecting a cross-correlation lag value, and updating the initial velocity model with a trained machine-learning model, reducing the need for manual intervention and enhancing the efficiency of subsurface imaging.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual velocity boundary identification is used, then measurement precision can be maintained, but productivity decreases due to time-consuming operations
Solution Approach 1:
The patent replaces manual mechanical velocity boundary identification with an automated machine learning system. The trained model processes seismic migration gathers and automatically determines velocity boundaries, eliminating the need for human operators to manually analyze each boundary, thus significantly reducing processing time while maintaining accuracy
Solution Approach 2:
The system performs self-service by automatically selecting cross-correlation lag values and updating velocity models without requiring continuous human intervention. The machine learning model independently analyzes seismic data, selects optimal parameters, and generates updated velocity models, enabling the system to operate autonomously throughout the velocity boundary identification process
2Productivity
If automated velocity model updating is implemented, then productivity increases, but device complexity increases due to machine learning components
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model using human-picked boundary data and augmented data before actual velocity model updating. This preliminary training phase prepares the model to automatically process new seismic data, allowing the system to achieve high productivity in the actual application while managing complexity through pre-computed model parameters
Solution Approach 2:
The system uses copying by creating augmented boundary data from human-picked boundary data through augmentation operations. This creates synthetic training data that replicates the complexity of real geological structures, enabling the model to learn and reproduce accurate velocity boundaries without requiring direct complex analysis during deployment
3Measurement precision
If human operators manually determine velocity boundaries, then measurement precision is maintained, but loss of time increases significantly
Solution Approach 1:
The patent implements feedback by using human-picked boundary data to train and validate the machine learning model. The model continuously refines its velocity boundary predictions by comparing against known accurate boundaries from human annotation, ensuring that measurement precision is maintained while the automated system processes data much faster than manual methods
Solution Approach 2:
The system changes parameters by automatically selecting optimal cross-correlation lag values from multiple generated seismic migration gathers. Instead of manually adjusting parameters, the machine learning model evaluates different lag values and selects the ones that produce the most accurate velocity boundaries, maintaining precision while reducing time consumption
Data Source
AI summary
A method may include obtaining an initial velocity model regarding a subterranean formation of interest. The method may further include generating various seismic migration gathers with different cross-correlation lag values based on a migration-velocity analysis and the initial velocity model. The method may further include selecting a predetermined cross-correlation lag value automatically using the seismic migration gathers and based on a predetermined criterion. The method may further include determining various velocity boundaries within the initial velocity model using a trained model, wherein the trained model is trained by human-picked boundary data and augmented boundary data. The method may further include updating, by the computer processor, the initial velocity model using the velocity boundaries, the automatically-selected cross-correlation lag value, and the migration-velocity analysis to produce an updated velocity model. The method may further include generating an image of the subterranean formation of interest using the updated velocity model.