Subsurface Geological Models Using Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing subsurface geological models often fail to accurately represent non-stationary facies and lithologies, leading to geologically unrealistic models that do not adhere to observational data.
Innovation Solution
The use of machine learning and deep learning algorithms, specifically conditioned generative machine learning algorithms (CGMLAs), in conjunction with stratigraphic forward models (SFMs), to develop geologically realistic, non-stationary subsurface geological models that incorporate 3-dimensional facies distribution and stratal relationships.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If stratigraphic forward modeling is used to develop subsurface geological models, then the models are geologically realistic, but they may not always adhere to the collected observational data such as geological data gathered from previously drilled wellbores
Solution Approach 1:
The patent combines stratigraphic forward modeling with machine learning algorithms (specifically conditional generative machine learning algorithms) to create a hybrid approach. The SFM provides the geological framework and realism, while the CGMLA adjusts the model to fit observational data from wellbores, seismic data, and other measurements, thereby resolving the contradiction between geological realism and data adherence.
Solution Approach 2:
The machine learning component receives feedback from observational data (wellbore data, seismic data, etc.) and uses this feedback to conditionally generate geological models that adhere to the observed constraints. This feedback mechanism allows the model to adjust its predictions based on actual measurements, ensuring both geological realism and data consistency.
2Ease of manufacture
If traditional interpolation and extrapolation methods are used to generalize geological properties to a 3-dimensional volume, then the modeling process is simple, but the models fail to accurately represent non-stationary facies and lithologies
Solution Approach 1:
The patent changes the fundamental parameters of the modeling approach by using conditional generative machine learning algorithms that can handle non-stationary processes. Instead of traditional interpolation that assumes stationarity, the CGMLA uses probability density functions and conditional generation to model facies that vary non-stationarily in space and time, significantly improving accuracy while maintaining computational feasibility.
Solution Approach 2:
The patent replaces traditional mechanical interpolation methods with machine learning-based conditional generative models. This substitution allows the system to capture complex non-stationary patterns in geological data that traditional interpolation cannot handle, improving the representation of facies and lithologies while maintaining ease of use through automated processing.
Data Source
AI summary
In general, in one aspect, embodiments relate to a method that includes selecting one or more stratigraphic forward models from a digital analogue library, generating one or more k-layers based at least in part on the one or more selected stratigraphic forward models and one or more generative machine learning models, and predicting thicknesses of one or more geological properties based at least in part on the one or more k-layers.