Integrated Seismic Data Augmentation With Stochastic Velocity Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing seismic data for exploration seismology is of low quality and lacks variety, leading to poor performance of neural networks in tasks such as seismic denoising, data reconstruction, and lithology prediction, due to the lack of realistic noise, scattering, and fluctuations in traditional synthetic data.
Innovation Solution
Integrated seismic data augmentation is achieved through image-guided interpolation and sequential Gaussian simulation, generating high-quality seismic data that includes realistic scatterings and fluctuations, using well logs and seismic images to create stochastic seismic models that mimic real-world conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional synthetic seismic data is used for training neural networks, then data generation is simple and fast, but the data lacks realistic noise, scattering, and fluctuations resulting in poor model performance
Solution Approach 1:
The patent combines multiple data generation techniques (image-guided interpolation, sequential Gaussian simulation, and forward simulation) into an integrated pipeline. This merging of methods allows the system to generate seismic data that incorporates realistic noise, scattering, and fluctuations while maintaining computational efficiency through parallel processing of multiple simulation steps.
Solution Approach 2:
The patent performs preliminary processing of well log data and seismic images before generating the final augmented seismic data. By pre-processing the input data to extract meaningful features and prepare stochastic models in advance, the system reduces the computational burden during the main data generation phase while ensuring high data quality.
2Adaptability or versatility
If high-quality augmented seismic data with realistic characteristics is generated, then neural network generalization ability improves, but data generation time and computational resources increase
Solution Approach 1:
The patent divides the data generation process into distinct sequential stages: image-guided interpolation of well log data, sequential Gaussian simulation to create stochastic velocity models, and forward simulation to generate final seismic data. This segmentation allows each stage to be optimized independently and enables parallel processing of multiple simulation steps, reducing overall generation time while maintaining data quality.
Solution Approach 2:
The patent employs sequential Gaussian simulation to generate stochastic velocity models by varying statistical parameters (mean, variance, covariance) to match observed seismic characteristics. By adjusting these parameters to represent different geological conditions and noise levels, the system efficiently generates diverse training data that enhances model generalization without requiring exhaustive computational resources.
3Quantity of substance
If existing seismic data is used directly for training, then no data processing is needed, but the data quality is low and lacks variety for effective deep learning
Solution Approach 1:
The patent creates synthetic copies of real seismic data by using image-guided interpolation and sequential Gaussian simulation to generate virtual seismic sections that replicate the statistical properties and geological features of actual seismic data. These synthetic copies include realistic noise patterns, scattering effects, and fluctuations, providing diverse training data without requiring physical field measurements or processing of existing low-quality data.
Data Source
AI summary
A computer implemented method that enables integrated seismic data augmentation is described. The method includes generating a three-dimensional (3D) seismic velocity model using sequential gaussian simulation. The method includes executing forward modeling simulations on the seismic velocity model to generate primaries and multiples of wavelets associated with the seismic velocity model. Additionally, the method includes combining the primaries and the multiples of wavelets with noise to generate augmented seismic data.