Reconstructing Low-Frequency Seismic Data via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Seismic data analysis faces challenges in accurately determining subsurface features due to hardware limitations that result in the loss of low-frequency components, leading to artifacts and errors in velocity models.
Innovation Solution
A computer-implemented method and system that use machine learning to reconstruct low-frequency seismic data by training a model based on initial and secondary frequency data, then update the training velocity model to generate accurate subsurface feature representations using full waveform inversion (FWI).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If hardware limitations are accepted in seismic data collection, then device complexity is reduced, but low-frequency components are lost leading to measurement precision degradation
Solution Approach 1:
The patent applies preliminary action by training a machine learning model in advance using synthetic seismic data with known low-frequency components. This pre-trained model is then used to reconstruct low-frequency data from incomplete field measurements, eliminating the need for complex hardware capable of directly recording low-frequency signals while recovering the lost measurement precision.
Solution Approach 2:
The patent uses copying by generating synthetic seismic data that replicates the characteristics of full-bandwidth data including low-frequency components. This synthetic data serves as a template to train the machine learning model, which then copies the low-frequency information pattern to reconstruct missing data from incomplete measurements.
2Measurement precision
If low-frequency data is lost during measurement, then measurement precision deteriorates, but the quantity of collected data is reduced
Solution Approach 1:
The patent introduces an intermediary machine learning model that bridges the gap between incomplete high-frequency measurements and the missing low-frequency data. This intermediary reconstructs the lost low-frequency components by learning the relationship between frequency bands, thereby improving velocity model accuracy without requiring additional physical data collection.
3Loss of time
If traditional FWI is used without low-frequency data, then processing time is reduced, but reliability of subsurface feature determination deteriorates due to cycle-skipping artifacts
Solution Approach 1:
The patent applies preliminary action by reconstructing low-frequency data before performing full waveform inversion. This pre-reconstruction step provides the necessary low-frequency information that guides the FWI process, preventing cycle-skipping artifacts and improving the reliability of subsurface feature determination while maintaining reasonable processing times through efficient machine learning inference.
Data Source
AI summary
A computer-implemented method for obtaining reconstructed seismic data for determining a subsurface feature, includes: determining an initial training velocity model, training a machine learning model based on first training seismic data and second training seismic data generated from the training velocity model, the first training seismic data corresponding to one or more first frequencies, the second training seismic data corresponding to one or more second frequencies lower than the one or more first frequencies, obtaining, based on measured seismic data and the machine learning model, reconstructed seismic data corresponding to the one or more second frequencies, generating a velocity model based on the measured seismic data, the reconstructed seismic data, and a full waveform inversion (FWI), and when the generated velocity model does not satisfy a preset condition, updating the training velocity model based on the generated velocity model, to obtain updated reconstructed seismic data for determining a subsurface feature.