EOR Front Tracking on Coarse Grids Using ML Correlations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing EOR flooding simulations face a dilemma between achieving fast results on coarse grids and maintaining accuracy by capturing fine geological features, with current adaptive grid refinement methods struggling to predict front position and requiring significant computational resources.
Innovation Solution
A machine-learning-based workflow separates front tracking from reservoir simulation, using trained correlations to predict front position and topology on coarse grids, leveraging localized static and dynamic reservoir properties to refine front position and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a fine scale grid is used to capture front gradients accurately, then measurement precision is improved, but computational time increases significantly
Solution Approach 1:
The reservoir grid is segmented into different resolution levels: a coarse grid for overall simulation and a fine grid only where needed near the flooding front. This allows accurate front capture in critical areas while maintaining fast computation on the majority of the domain using coarse grid.
Solution Approach 2:
Different parts of the reservoir are simulated at different grid resolutions. The fine grid is applied locally only in regions where the flooding front is present or expected, while the coarse grid is used in regions far from the front. This local differentiation maintains accuracy where needed and speed elsewhere.
2Productivity
If a coarse grid is used to speed up simulation, then productivity is improved, but measurement precision deteriorates due to increased numerical dispersion error
Solution Approach 1:
The computational domain is divided into coarse and fine grid regions. The coarse grid provides fast simulation for the majority of the reservoir, while the fine grid is segmentatively applied only in front zones where numerical dispersion would otherwise severely degrade accuracy.
Solution Approach 2:
An interface mechanism is introduced between coarse and fine grid regions to transfer front position and property information. This intermediary allows the coarse grid simulation to benefit from fine grid accuracy near the front without requiring the entire grid to be fine.
3Measurement precision
If dynamic grid refinement is implemented to capture front position, then measurement precision is improved, but device complexity increases due to physical property transfer requirements
Solution Approach 1:
The fine grid is pre-configured at expected front locations based on predictive models or historical data, before the actual simulation runs. This preliminary placement avoids the need for complex real-time grid manipulation and physical property transfer during the simulation.
Solution Approach 2:
Instead of dynamically creating and destroying fine grid cells, the fine grid is copied as a static overlay or embedded structure at predetermined locations. This simplifies the implementation by avoiding complex grid generation and property interpolation algorithms.
Data Source
Figure 1A~1B
Figure 2A~2B
Figure 2C~2E
AI summary
The present disclosure provides a workflow for modelling EOR flooding operations performed on a reservoir by separating front tracking from the reservoir simulation process, so that the front's position and topology evolves in parallel with the coarse grid simulation, through modifications using machine-learning-trained correlations.