Reservoir Productivity Mapping Using Random Forest Parameter Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for characterizing subsurface reservoirs often overfit and fail to capture complex interaction structures in noisy reservoir data, leading to inadequate characterization of rock properties and production performance.
Innovation Solution
A method and system utilizing random forest machine learning to generate refined production parameter maps, applying a parameter model to identify critical production parameters, and generating visual representations of estimated reservoir productivity as a function of position, incorporating user-defined well designs to enhance production predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing methods for characterizing subsurface reservoirs are used, then the process is simpler, but the methods overfit and fail to capture complex interaction structures in noisy reservoir data
Solution Approach 1:
The patent introduces random forest machine learning models as an intermediary between raw reservoir data and productivity predictions. These models serve as a mediator that can handle noisy data and complex interactions without overfitting, bridging the gap between simple characterization methods and the need for accurate, complex relationship capture.
Solution Approach 2:
The patent transforms the approach by changing from traditional deterministic parameter analysis to probabilistic machine learning parameter estimation. The random forest models estimate productivity parameters as probability distributions rather than fixed values, allowing the system to capture uncertainty and complex interactions while maintaining computational tractability.
2Measurement precision
If traditional parameter analysis methods are used, then the computational process is faster, but they provide inadequate characterization of rock properties and production performance
Solution Approach 1:
The patent performs preliminary action by training random forest models on historical reservoir data before actual productivity prediction. This pre-computation phase captures complex relationships and interaction structures in advance, so that when actual predictions are needed, the models can provide rapid, precise estimates without repeating the full computational analysis.
Solution Approach 2:
The patent creates simplified copies of complex reservoir systems through random forest surrogate models. These surrogate models replicate the behavior of full numerical reservoir simulators but run much faster, allowing precise characterization of rock properties and production performance without the computational burden of repeated full-physics simulations.
3Ease of operation
If simple linear regression techniques are used, then the analysis is easier to perform, but they cannot capture complex interaction structures in reservoir data
Solution Approach 1:
The patent substitutes traditional mechanical linear regression techniques with machine learning-based random forest models. This replacement maintains ease of operation through automated model training and prediction while dramatically improving the ability to capture complex non-linear interactions in reservoir data, as random forests naturally handle feature interactions without requiring explicit specification.
Data Source
Figure 1A
Figure 1B
Figure 2
AI summary
Systems and methods for estimating reservoir productivity as a function of position in a subsurface volume of interest are disclosed. Exemplary implementations may: obtain subsurface data and well data corresponding to a subsurface volume of interest; obtain a parameter model; use the subsurface data and the well data to generate multiple production parameter maps; apply the parameter model to the multiple production parameter maps to generate refined production parameter values; generate multiple refined production parameter graphs; display the multiple refined production parameter graphs; generate one or more user input options; receive a defined well design and the one or more user input options selected by a user to generate limited production parameter values; generate a representation of estimated reservoir productivity as a function of position in the subsurface volume of interest using the defined well design and visual effects; and display the representation.