S-Wave Velocity Modeling Using ML-Guided Seismic Boundaries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing seismic imaging methods, particularly those involving migration algorithms, are computationally intensive and struggle with accurately determining S-wave velocity models, especially in complex geological regions, leading to incomplete or inaccurate subsurface representations.
Innovation Solution
A method utilizing a trained machine-learning model to determine S-wave velocity boundaries by analyzing migration gathers and cross-correlation lag values, combined with P-wave velocity models, to generate a combined velocity model that enhances seismic imaging accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If migration algorithms are used to convert time-based seismic data into depth representation, then subsurface imaging is achieved, but computational cost increases significantly
Solution Approach 1:
The patent applies preliminary action by using machine learning models to predict S-wave velocity boundaries before performing the computationally intensive migration process. The trained model processes seismic data and attributes to generate velocity boundary predictions in advance, which then guide the migration algorithm to focus computational resources only on critical regions, thereby reducing overall computational cost while maintaining imaging accuracy.
2Measurement precision
If complete migration-wavefield inversion is performed, then accurate depth representation is obtained, but processing time increases
Solution Approach 1:
The patent extracts the most computationally demanding aspect of the inversion process by using a machine learning model to predict S-wave velocity boundaries separately from the main migration workflow. This extracted prediction step provides guidance to the migration algorithm, allowing it to achieve accurate depth representation without performing complete wavefield inversion, thereby significantly reducing processing time.
Solution Approach 2:
The machine learning model performs preliminary prediction of velocity boundaries before the migration process begins. This preliminary action prepares the system by identifying key geological features and velocity discontinuities in advance, enabling the migration algorithm to focus computational efforts on these critical regions rather than processing the entire subsurface volume uniformly, thus reducing overall processing time.
3Measurement precision
If S-wave velocity models are determined using traditional methods, then velocity information is obtained, but accuracy in complex geological regions deteriorates
Solution Approach 1:
The patent changes the approach from traditional physics-based inversion methods to a data-driven machine learning approach. The model is trained on seismic data, velocity ratios, and geological attributes to learn complex non-linear relationships between these parameters and S-wave velocity boundaries. This parameter transformation enables accurate velocity modeling in complex geological regions where traditional methods fail, as the model can capture intricate geological patterns through learned feature representations.
Solution Approach 2:
The patent creates a composite modeling approach by combining multiple data sources (seismic data, velocity ratios, geological attributes) and multiple processing techniques (machine learning prediction, migration algorithm) into an integrated workflow. This composite method leverages the strengths of each component: the ML model's ability to handle complex patterns and the migration algorithm's accuracy in depth imaging, resulting in superior S-wave velocity modeling performance in complex environments.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method may include obtaining a P-wave velocity model (476) and velocity ratio data (411) regarding a geological region of interest. The method may further include generating, based on the P-wave velocity model (476) and the velocity ratio data (411), an initial S-wave velocity model (420) regarding the geological region of interest. The method may further include determining various velocity boundaries within the initial S-wave velocity model (420) using a trained model. The method may further include updating the initial S- wave velocity model (420) using the velocity boundaries, an automatically- selected cross-correlation lag value based on various seismic migration gathers, and a migration- velocity analysis to produce an updated S-wave velocity model (470). The method further includes generating a combined velocity model for the geological region of interest using the updated S-wave velocity model (470) and the P-wave velocity model (476).