Seismic Data Quality via Machine Learning Transformation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Seismic data processing in the first stage often compromises on signal quality due to cost and computing time constraints, which affects the quality of subsurface interpretation.

Innovation Solution

A method involving the use of machine learning models, specifically generative adversarial networks (GANs), to transform low-quality seismic data processed using fast-track techniques into high-quality data comparable to full-track processing, within the time frame of fast-track processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If fast-track processing is used to reduce computing time and cost, then processing speed and cost efficiency are improved, but signal quality deteriorates

Engineering Contradiction:
Improveprocessing speedVSAvoidsignal quality
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The system performs fast-track processing first to obtain initial processed data quickly, then uses a trained machine learning model to transform this data into full-track quality results. The preliminary fast-track processing provides a good starting point that the ML model can refine, combining the speed advantage of fast-track with the quality of full-track processing.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

A machine learning model acts as an intermediary between fast-track and full-track processing. The model is trained on pairs of fast-track and full-track processed data to learn the transformation patterns, then applies this learned transformation to convert fast-track results into high-quality full-track equivalent results, bridging the quality gap while maintaining speed advantages.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Manufacturing precision

If full-track processing is used to improve signal quality, then processing quality is improved, but computing time and cost increase

Engineering Contradiction:
Improvesignal qualityVSAvoidturnaround time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The system uses fast-track processing to create a copy or approximation of the full-track processing result, then applies a machine learning model to transform this copy into a high-quality result that matches full-track processing quality. This copying approach allows achieving full-track quality without incurring the full computational cost and time of actual full-track processing.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The machine learning model learns to transform the parameters and characteristics of fast-track processed data into those of full-track processed data. By changing the parameters of the input data (fast-track results) through the learned transformation, the system produces output data with the desired full-track quality parameters while maintaining the speed advantage of fast-track processing.

Inventive Principle:
Principle #35Parameter changes

3Manufacturing precision

If machine learning models are used to transform fast-track data to full-track quality, then signal quality is improved, but model training complexity increases

Engineering Contradiction:
Improvesignal qualityVSAvoidmodel training complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The system performs preliminary training of the machine learning model using paired fast-track and full-track processed data before deployment. This preliminary training action prepares the model in advance to perform the quality transformation, so that during actual operation, the model can quickly transform fast-track results into high-quality outputs without requiring complex real-time computations or iterative training.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4571366A1System and method for generating high-quality seismic data from low-quality seismic data
Publication Date: 2025.06.18 SERVICES PETROLIERS SCHLUMBERGER SA
  • EP4571366A1 patent drawingFigure 1
  • EP4571366A1 patent drawingFigure 2
  • EP4571366A1 patent drawingFigure 3

AI summary

A method for generating high-quality seismic data includes receiving a first dataset. The method also includes selecting a first processing technique based upon the first dataset. The method also includes producing first processed data from the first dataset using the first processing technique. The method also includes selecting a second processing technique based upon the first dataset. The second processing technique is different than the first processing technique. The method also includes producing second processed data from the first dataset using the second processing technique. The method also includes training a machine learning model to transform the first processed data into the second processed data.