Hierarchical Geological Model Conditioning with ML Templates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating geological models of hydrocarbon reservoirs face challenges in modeling large-scale features, managing a large number of parameters, and maintaining geologic realism, particularly due to difficulties in enforcing higher-order patterns and updating highly correlated parameters.
Innovation Solution
A hierarchical conditioning methodology is employed to build and condition geological models by generating template instances based on larger-scale and smaller-scale data, using machine learning techniques such as auto-encoders, generative adversarial networks, and functional form modeling to parameterize templates at different levels, allowing for iterative conditioning and deformation of templates to match geological features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional geological modeling methods are used to model large-scale features, then the model can be generated, but the ability to enforce higher-order patterns and maintain geologic realism deteriorates
Solution Approach 1:
The patent segments the geological modeling process into hierarchical levels (large-scale features, medium-scale features, small-scale features), where each level is modeled separately with appropriate templates and parameters. This segmentation allows enforcement of higher-order patterns at each level without being overwhelmed by the complexity of managing all parameters simultaneously across the entire model.
Solution Approach 2:
The patent transforms the modeling approach by changing parameters from a flat, comprehensive parameter set to a hierarchical parameter structure where parameters are organized by scale and type. Machine learning techniques are used to learn parameter relationships and constraints, automatically managing parameter correlations and reducing the burden of manual parameter management while maintaining geologic realism.
2Manufacturing precision
If a comprehensive parameter set is used to model all geological features, then the model detail is improved, but the difficulty of managing highly correlated parameters increases
Solution Approach 1:
The comprehensive parameter set is segmented into hierarchical groups corresponding to different geological scales and feature types. Each segment is managed independently with its own template and parameter constraints, reducing the complexity of managing correlations across all parameters while maintaining detailed modeling capability within each segment.
Solution Approach 2:
Machine learning models serve as intermediaries between the comprehensive parameter set and the final geological model. These ML models learn the complex correlations between parameters from training data and automatically adjust parameter values to satisfy geological constraints, eliminating the need for manual management of parameter correlations while preserving model detail.
3Reliability
If iterative conditioning is performed to improve model accuracy, then the geological feasibility is improved, but the computational time and processing complexity increase
Solution Approach 1:
The patent performs preliminary conditioning at each hierarchical level before proceeding to the next level. Templates are pre-configured with scale-appropriate parameters and constraints, and machine learning models are pre-trained on geological data to enable faster inference during the conditioning process. This preliminary preparation reduces the computational burden during iterative conditioning while maintaining geological feasibility.
Solution Approach 2:
The patent adds a hierarchical dimension to the conditioning process, transitioning from a single-level iterative conditioning to a multi-level hierarchical conditioning. This dimensional change allows conditioning to proceed systematically from large-scale to small-scale features, reducing the overall computational complexity by breaking down the conditioning task into manageable stages that can be performed more efficiently.
Data Source
AI summary
A hierarchical conditioning methodology for building and conditioning a geological model is disclosed. In particular, the hierarchical conditioning may include separate levels of conditioning of template instances using larger-scale data (such as conditioning using large-scale data and conditioning using medium-scale data) and using smaller-scale data (such as fine-scale data). Further, one or more templates, to be instantiated to generate the geological bodies in the model, may be selected from currently available templates and/or machine-learned templates. For example, the templates may be generated using unsupervised or supervised learning to re-parameterize the functional form parameters, or may be generated using statistical generative modeling.