Deep Learning Seismic Inversion With Physics-Guided Refinement

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing seismic inversion workflows are complicated, require significant computational resources, and heavily rely on human supervision, often resulting in lower resolution and inaccurate synthetic seismic data.

Innovation Solution

A deep learning workflow utilizing multi-task convolutional neural networks (CNNs) for large-scale structural model construction, followed by physics-guided refinement, to automate geophysical property estimation with minimal human intervention.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional seismic inversion workflows are used, then computational resources and human supervision are required, but the resolution and accuracy of synthetic seismic data are reduced

Engineering Contradiction:
Improveaccuracy of geophysical property estimationVSAvoidcomplexity of inversion workflow
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces traditional mechanical/computational seismic inversion systems with a neural network-based system. The neural network is trained on seismic data and well logs to directly predict geophysical properties, eliminating the need for complex iterative inversion algorithms and reducing computational resource requirements while improving accuracy and resolution.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If traditional seismic inversion workflows are used, then significant computational resources are required, but the resolution of synthetic seismic data is reduced

Engineering Contradiction:
Improveresolution of synthetic seismic dataVSAvoidcomputational resources
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent applies preliminary action by training the neural network in advance on large datasets of seismic data and well logs. This pre-training phase captures the complex relationships between seismic characteristics and geophysical properties, allowing the network to make accurate predictions during actual inversion operations without requiring significant computational resources at that stage.

Inventive Principle:
Principle #10Preliminary action

3Extent of automation

If traditional seismic inversion workflows are used, then human supervision is required, but the accuracy of geophysical property estimation is reduced

Engineering Contradiction:
Improveautomation of inversion processVSAvoidaccuracy of geophysical property estimation
Core Design Contradiction:
Extent of automationVSMeasurement precision

Solution Approach 1:

The patent implements self-service by enabling the neural network to automatically perform seismic inversion without human intervention. The network learns from training data and independently predicts geophysical properties, eliminating the need for human supervision while maintaining or improving accuracy through its ability to process and interpret complex seismic patterns.

Inventive Principle:
Principle #25Self-service

4Extent of automation

If neural networks are used for geophysical property estimation, then automation is improved, but training data requirements increase

Engineering Contradiction:
Improveautomation of property estimationVSAvoidtraining data requirements
Core Design Contradiction:
Extent of automationVSQuantity of substance

Solution Approach 1:

The patent merges multiple data sources including seismic data, well logs, and synthetic seismic data into a unified training dataset. By combining these different types of data, the neural network can learn from diverse patterns and relationships, improving its predictive capability while efficiently utilizing the available training data.

Inventive Principle:
Principle #5Merging (Combining)

Data Source

PatentEP4617734A1A deep learning workflow for seismic inversion
Publication Date: 2025.09.17 SERVICES PETROLIERS SCHLUMBERGER SA
  • EP4617734A1 patent drawingFigure 1A
  • EP4617734A1 patent drawingFigure 1B
  • EP4617734A1 patent drawingFigure 1C~1D

AI summary

A method for converting seismic data into geophysical property models. The method includes receiving seismic data from a site, wherein the seismic data includes seismic stacks and well logs received from the site. A large-scale structural model (LSSM) is then constructed from the received seismic data. The method includes estimating initial geophysical properties related to the site using a first neural network trained by the seismic data and the constructed LSSM. Data may then be extracted from the received data and used to train a second neural network. The method also includes revising the initial geophysical properties using a second trained neural network. The revised geophysical properties, which may include p-velocity, density, or s-velocity, may be displayed on a screen. The method also includes performing a site action based on the revised geophysical properties.