Ensemble Machine Learning for Lower-Uncertainty Subsurface Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current reservoir modeling methods produce characterization models with uncertainty that affect the accuracy of follow-on reservoir simulation operations.
Innovation Solution
Employ ensemble machine learning prediction using multiple machine learning models, including decision trees, to determine the best and most likely property distribution characteristics of subsurface geological formations, incorporating data from external models to supplement training data and reduce user input.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional reservoir modeling methods are used, then the process is simpler and requires less computational resources, but the characterization models produce higher uncertainty that affects accuracy
Solution Approach 1:
An ensemble of machine learning models serves as an intermediary between traditional reservoir modeling methods and final predictions. Multiple individual models (decision trees, random forests, support vector machines, neural networks) work together to reduce uncertainty and improve prediction reliability while managing computational complexity through structured ensemble architecture
Solution Approach 2:
Multiple machine learning models with different algorithms and training approaches are merged into a single ensemble system. Each model contributes its strengths, and their combined predictions through voting or averaging mechanisms produce more reliable reservoir characterization results than any single model could achieve alone
2Measurement precision
If more training data is collected from additional wells, then model accuracy improves, but the cost and time required for data collection increases
Solution Approach 1:
Machine learning models are trained in advance on available well data before actual reservoir modeling is needed. The ensemble models perform preliminary learning from training datasets, enabling rapid predictions during production planning without requiring additional time-consuming data collection or processing at the decision-making stage
Solution Approach 2:
The machine learning models create virtual copies of the reservoir properties by learning patterns from existing well data. Instead of physically drilling additional wells to gather more data, the models generate synthetic predictions that replicate what additional measurements would reveal, significantly reducing the need for expensive and time-consuming physical data collection
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Method and systems are provided that create one or more models of a subsurface geological formation (such as a reservoir characterization model of a hydrocarbon reservoir or a model of some other subsurface geological formation). The method and systems are configured to extend a machine learning ensemble (such as an ensemble tree-based machine learning model such as a random forest learning model) to use or embed data derived from one or more secondary models as part of the training operations of the machine learning ensemble and online use of the trained machine learning ensemble. Such data can provide information that supplements the information contained in the training data/input data.