Super Resolution Model for Seismic Visualization Data Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The oil & gas industry faces challenges in efficiently rendering and storing seismic visualizations due to their voluminous nature, which is network and memory intensive, and requires high resolution for effective interpretation.
Innovation Solution
A super resolution machine learning model is used to reconstruct high resolution seismic data from low resolution data, reducing storage and communication costs, and enabling efficient generation of seismic visualizations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high resolution seismic data is used to generate visualizations, then interpretation quality is improved, but storage and network costs increase
Solution Approach 1:
The patent uses a trained neural network model to generate synthetic high-resolution seismic data copies from low-resolution input data. The model learns the relationship between low and high resolution data during training, then applies this learned transformation to generate high-resolution visualizations without requiring actual high-resolution source data, thereby reducing storage and network transmission requirements while maintaining interpretation quality
Solution Approach 2:
The patent transforms the resolution parameter of seismic data using a trained neural network. The model takes low-resolution seismic data as input and outputs high-resolution seismic data by learning the parameter transformation patterns from training pairs, effectively changing the resolution parameter without proportionally increasing data volume
2Measurement precision
If high resolution seismic visualizations are generated, then interpretation quality is improved, but computational resources increase
Solution Approach 1:
The patent performs preliminary training of the neural network model using paired high and low resolution seismic data before deployment. This preliminary action captures the transformation patterns during the training phase, so that during actual visualization generation, the model can quickly apply the learned transformations without requiring intensive computational resources for real-time high-resolution data processing
Solution Approach 2:
The patent replaces traditional mechanical or algorithmic upscaling methods with a trained neural network model. Instead of using computationally intensive interpolation or super-resolution algorithms during visualization generation, the system uses the pre-trained model to perform the transformation, significantly reducing the computational power required at runtime while maintaining high visualization quality
3Quantity of substance
If low resolution seismic data is used, then storage and network costs are reduced, but interpretation quality deteriorates
Solution Approach 1:
The patent creates synthetic high-resolution copies of low-resolution seismic data using a trained neural network. The model generates detailed features and structures that appear in high-resolution data by learning from training pairs, effectively copying the appearance and interpretability of high-resolution data while working with compact low-resolution storage
Solution Approach 2:
The patent creates an asymmetric relationship between storage resolution and display resolution. The system stores and transmits low-resolution data (small size) but generates high-resolution visualizations (large effective size) through the neural network model, breaking the traditional symmetric requirement where storage resolution must match display resolution
Data Source
AI summary
A method, apparatus, and program product utilize a super resolution machine learning model to reconstruct high resolution seismic data from low resolution seismic data in connection with generating seismic visualizations, e.g., to reduce storage and/or communication costs associated with generating seismic visualizations.