Graph Networks for Geological Reasoning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning approaches for hydrocarbon identification in geophysical prospecting are inefficient due to their inability to effectively reason about the compositional nature of geoscientific data, requiring abundant data and computing resources, and struggling with small data sets, ambiguous interpretations, and generalization across different geological settings.
Innovation Solution
The use of graph networks for geological reasoning, which involves obtaining subsurface data, extracting structured representations using a knowledge model that includes geoscience rules or ontology, and performing tasks like question answering, decision-making, and probability assessment to analyze hydrocarbon systems more efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If end-to-end machine learning approaches are used for structural interpretation, then automation is improved, but the ability to reason about compositional relationships and generalize to new settings deteriorates
Solution Approach 1:
The patent segments the structural interpretation task into distinct compositional components (horizons, geobodies, faults) that can be independently identified and then reasoned about in relationship to each other. This allows the system to maintain automation while incorporating geological reasoning about how these components relate, improving generalization to new settings.
Solution Approach 2:
The patent introduces an intermediary reasoning layer between data processing and final interpretation that applies geological knowledge about compositional relationships. This intermediary component enables the system to generalize better by reasoning about relationships among structural elements rather than relying solely on end-to-end pattern matching.
2Adaptability or versatility
If deep learning approaches with minimal a priori assumptions are used, then adaptability is improved, but reasoning about compositional relationships and building intuition deteriorates
Solution Approach 1:
The patent performs preliminary structuring of geological data into compositional elements before applying machine learning. By pre-organizing data into horizons, geobodies, and faults with defined relationships, the system preserves compositional reasoning capability while still allowing flexible learning about specific geological settings.
3Productivity
If conventional machine learning approaches are used, then computational efficiency is improved, but the ability to handle ambiguous interpretations and small data sets deteriorates
Solution Approach 1:
The patent incorporates feedback mechanisms where the system evaluates multiple possible interpretations of ambiguous geological features and refines its conclusions based on consistency with known compositional relationships. This allows handling of ambiguous interpretations while maintaining computational efficiency through structured reasoning.
Data Source
AI summary
A method and apparatus for performing geological reasoning, A method includes: obtaining subsurface data for a subsurface region; obtaining a knowledge model; extracting a structured representation from the subsurface data using the knowledge model; and performing geological reasoning with a graph network based on the knowledge model and the structured representation. A method includes performing geological reasoning with a knowledge model that includes a set of geoscience rules or a geoscience ontology. A method includes performing geological reasoning with a structured representation that includes a graph. A method includes performing geological reasoning by one or more of the following: question answering; decision making; assigning ranking; and assessing probability.