Machine Learning Velocity Models for Accurate Seismic Depth Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating velocity models in seismic migration techniques often assign inaccurate depth values to interpreted horizons at drilled well locations and select unsuitable seismic reflectors, leading to discrepancies between seismic images and well data, which degrades seismic data migration and imaging.
Innovation Solution
A machine learning model is trained to identify seismic reflectors using seismic and sonic log data, selecting velocity knees for improved velocity model building, and iteratively refining the model until a final velocity model is achieved, ensuring accurate depth values and reflector selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used to generate velocity models, then the process is simpler, but the depth values assigned to interpreted horizons are inaccurate
Solution Approach 1:
The patent applies preliminary action by pre-training a machine learning model on sonic log data and seismic reflector information before the actual velocity model building process. This pre-trained model is then used to automatically identify velocity knees and select seismic reflectors during model generation, eliminating the need for manual interpretation and ensuring consistent accuracy across all depth values.
Solution Approach 2:
The patent replaces manual mechanical interpretation methods with an automated machine learning system. The ML model automatically identifies velocity knees, selects seismic reflectors, and generates velocity models without human intervention, substituting the traditional manual picking and interpretation process with an intelligent automated system that maintains high precision.
2Productivity
If manual selection of seismic reflectors is performed, then the process is more controllable, but the processing time increases
Solution Approach 1:
The patent implements self-service by enabling the machine learning model to autonomously perform all steps of velocity model generation including identifying velocity knees, selecting seismic reflectors, and updating the velocity model. The system serves itself by automatically iterating and refining models without requiring manual intervention at each step, dramatically reducing processing time while maintaining quality control through the trained model's consistent decision-making.
3Reliability
If inaccurate reflector selection is made, then the velocity model can be generated quickly, but the seismic imaging quality deteriorates
Solution Approach 1:
The patent applies feedback by using the trained machine learning model to continuously evaluate and select the most appropriate seismic reflectors based on their correlation with velocity knees. The model provides feedback on which reflectors accurately represent subsurface boundaries, ensuring that only high-quality reflectors are used to update the velocity model, thereby maintaining both imaging quality and model accuracy simultaneously.
Data Source
AI summary
Among other things, techniques are described for generating a velocity model for a subsurface structure using a machine learning model. A method can include generating, using seismic data and sonic log data from a subterranean surface, one or more Time-to-Depth Relationship (TDR) curves; generating, using (i) the seismic data and sonic log data and (ii) the one or more TDR curves, a combined set of seismic data and sonic log data; selecting one or more seismic reflectors; generating, using the one or more seismic reflectors, a velocity model update; generating, using the velocity model update and one or more operations of pre-stack depth migration, a candidate final velocity model; determining the candidate final velocity model satisfies a matching threshold; and in response, providing the final velocity model as output.