Machine Learning Reservoir Model Switching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current geomechanical simulation methods in the petroleum industry face challenges in accurately modeling subsidence and stress changes in reservoirs, particularly in fractured systems, leading to inefficient fluid flow modeling and potential well equipment damage due to the complexity of permeability changes and stress arching phenomena.
Innovation Solution
A machine learning classifier is used to dynamically reconfigure and switch between different reservoir models, such as single porosity, dual porosity, and dual permeability models, based on recent geomechanical simulation results and historical data, allowing for iterative updates of coupling parameters and improved simulation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a single reservoir model is selected and fully implemented for coupling, then the simulation process is straightforward, but the accuracy and realism of the simulation results may be insufficient when the reservoir conditions change
Solution Approach 1:
The patent implements dynamic model selection where the reservoir model is not fixed but can change based on simulation conditions. The system automatically switches between different reservoir models (single porosity, dual porosity, dual permeability) depending on the geomechanical state, allowing the simulation to adapt to changing conditions while maintaining operational simplicity through automation.
Solution Approach 2:
The patent changes the model selection parameter dynamically based on simulation results. By monitoring geomechanical parameters and using machine learning classifiers to determine when model switching is needed, the system adjusts the reservoir model parameters to match actual reservoir conditions, improving accuracy without manual intervention.
2Reliability
If multiple reservoir models are used to capture complex reservoir behavior, then the simulation accuracy improves, but the device complexity and computational requirements increase
Solution Approach 1:
The system performs self-service through automated machine learning-based model selection. The machine learning classifier automatically analyzes simulation results and determines when to switch between reservoir models, eliminating the need for manual model selection complexity while maintaining high simulation accuracy across different reservoir conditions.
Solution Approach 2:
The patent implements feedback mechanisms where simulation results are continuously monitored and fed back into the model selection process. The machine learning classifier uses this feedback to automatically adjust the reservoir model, creating a closed-loop system that maintains accuracy while managing complexity through automation.
3Productivity
If conventional reservoir simulation methods are used, then the computational time is shorter, but the ability to accurately model stress changes and permeability variations in fractured systems is reduced
Solution Approach 1:
The patent dynamically switches between different reservoir models based on the presence and characteristics of fractures in the system. When fractures are detected through machine learning analysis of geomechanical data, the system transitions to dual porosity or dual permeability models that can accurately capture stress changes and permeability variations, while using simpler models when fractures are absent to maintain computational efficiency.
Data Source
AI summary
An embodiment includes a method for use by at least one machine learning classifier. The method comprises the machine learning classifier obtaining one or more recent results from at least one geomechanical simulation; the machine learning classifier comparing the recent results to stored historical data; and, based on the comparing, the machine learning classifier deciding at least one reservoir model for use by at least one reservoir simulation.