Machine Learning Subsurface Data Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current approaches to generating subsurface data as a function of position and time are inefficient, requiring months for measurements and failing to account for uncertainty, and are computationally challenging, especially in complex oil-water-gas systems.
Innovation Solution
The use of machine learning techniques, such as neural networks, to train models that generate subsurface data based on subsurface property and energy value changes over time, incorporating uncertainty from various sources and allowing for immediate application to measured or simulated data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional measurement methods are used to generate subsurface data, then measurement precision may be maintained, but the process requires months for measurements and has low productivity
Solution Approach 1:
The patent replaces traditional mechanical measurement systems with machine learning models (neural networks) that can generate subsurface data computationally. The trained model instantaneously predicts subsurface properties from input data, eliminating the need for lengthy physical measurements while maintaining accuracy through learned patterns from training datasets.
Solution Approach 2:
The patent performs preliminary training of machine learning models using extensive subsurface data before actual application. This pre-training phase captures complex subsurface relationships, enabling the model to rapidly generate accurate predictions without requiring time-consuming measurements during the actual subsurface data generation process.
2Reliability
If complex oil-water-gas systems are analyzed using traditional methods, then comprehensive subsurface understanding may be achieved, but computational requirements become challenging
Solution Approach 1:
The patent creates a computational copy of the complex subsurface system through machine learning models. The trained neural network encapsulates the complex relationships between oil-water-gas systems, replacing the need for direct complex computational analysis while maintaining predictive accuracy through learned patterns from training data.
Solution Approach 2:
The patent transforms the complex multi-phase subsurface problem into a different parameter space where machine learning models can efficiently operate. By changing from direct physical simulation parameters to learned statistical parameters, the system achieves comparable or superior accuracy with reduced computational complexity.
3Reliability
If traditional subsurface modeling is used, then subsurface data can be generated, but uncertainty from various sources is not accounted for
Solution Approach 1:
The patent incorporates uncertainty quantification by using machine learning models that can provide not only predictions but also measures of confidence or uncertainty associated with those predictions. This feedback mechanism allows the system to identify areas where uncertainty is high, enabling targeted data collection or model refinement without compromising overall efficiency.
Data Source
AI summary
Systems and methods are disclosed for generating subsurface data as a function of position and time. Exemplary implementations may include obtaining a first initial subsurface model and a first set of subsurface parameters, obtaining training subsurface property data and a first training subsurface dataset, generating a first conditioned subsurface model, and storing the first conditioned subsurface model.