Machine Learning Seismic Imaging Reduces Computational Burden
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current migration algorithms for seismic data processing are computationally intensive, making them costly and time-consuming, especially when generating stacked seismic images of complex geological regions with varying particle velocities.
Innovation Solution
A method utilizing a pre-trained machine-learning model, such as a neural network, to approximate the computations required for migration, where seismic data is pre-processed into gathers and used to train the model, reducing the need for extensive computational efforts by transforming time-based data into depth representation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional migration algorithms are used to convert time-based seismic data into depth representation, then accurate seismic imaging is achieved, but computational time and processing cost increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model on a large dataset of seismic data before actual imaging. The model learns the complex mapping from time-based to depth-based representation in advance, so that during actual processing, the pre-computed knowledge can be rapidly applied without performing the full computational migration algorithm again, thus resolving the contradiction between accuracy and computational time
Solution Approach 2:
The patent uses copying by creating a machine learning model that replicates the functionality of traditional migration algorithms. Instead of directly executing the computationally intensive migration process, the system uses the trained model to copy or approximate the migration results much faster, maintaining imaging accuracy while dramatically reducing computational time
2Measurement precision
If complete migration-wavefield inversion is performed to achieve accurate subsurface representation, then imaging quality is improved, but computational complexity and processing resources increase
Solution Approach 1:
The patent applies mechanics substitution by replacing the traditional mechanical/computational migration algorithm with a machine learning-based system. The complex iterative calculations of wavefield inversion are substituted with a trained neural network that performs the same function through pattern recognition and learned mappings, reducing computational complexity while maintaining subsurface representation accuracy
3Measurement precision
If traditional seismic migration methods are used for complex geological regions with varying particle velocities, then accurate depth imaging is achieved, but processing cost and time consumption increase
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model on diverse seismic data including complex geological regions with varying particle velocities. This pre-computation of knowledge allows the model to rapidly handle similar complex cases in production, achieving accurate depth imaging for complex geology without the prohibitive processing costs of traditional methods
Data Source
AI summary
A method may include obtaining seismic data regarding a geological region of interest. The seismic data may include various pre-processed gathers. The method may further include obtaining a machine-learning model that is pre-trained to predict migrated seismic data. The method may further include selecting various training gathers based on a portion of the pre-processed gathers, a migration function, and a velocity model. The method may further include generating a trained model using the training gathers, the machine-learning model, and a machine-learning algorithm. The method may further include generating a seismic image of the geological region of interest using the trained model and a remaining portion of the seismic data.