Physics-Driven Deep Learning Inversion for Reservoir Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for reservoir monitoring, particularly in interwell spaces, are insufficient for comprehensive mapping of fluid distribution due to uncertainties in fluid flow prediction and insufficient well patterns, leading to inadequate optimization of oil production operations.
Innovation Solution
A hybrid approach combining physics-driven and machine learning/deep learning inversions is employed to generate models that map and monitor reservoir depletion, using a feedback loop to converge to a true model distribution through iterative processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physics-driven inversion methods are used for reservoir monitoring, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent combines physics-driven inversion with deep learning inversion into a hybrid coupled approach. The physics-driven inversion component provides accurate reservoir monitoring by integrating physical laws (fluid flow equations) with measured data, while the deep learning component accelerates computation. This merging resolves the contradiction by maintaining high measurement precision through physics constraints while reducing device complexity through intelligent algorithmic integration.
Solution Approach 2:
The patent introduces a coupling operator as an intermediary between the physics-driven inversion and deep learning inversion. This coupling operator facilitates information exchange and coordinate transformation between the two inversion methods, enabling them to work together synergistically. The intermediary resolves the contradiction by allowing the complex physics-based method to achieve high precision while the deep learning method provides computational efficiency.
2Productivity
If deep learning inversion is used for reservoir monitoring, then productivity is improved, but measurement precision deteriorates
Solution Approach 1:
The patent merges deep learning inversion with physics-driven inversion in a hybrid coupled framework. The deep learning component provides fast computation and high productivity by learning patterns from training data, while the physics-driven component ensures measurement precision by enforcing physical constraints. This combination resolves the contradiction by allowing both high productivity and high precision to coexist through complementary strengths.
Solution Approach 2:
The patent implements a feedback mechanism where the physics-driven inversion results are used to guide and correct the deep learning inversion outcomes. The coupling operator continuously adjusts the deep learning predictions based on physics-based constraints, ensuring that the fast deep learning method does not sacrifice accuracy. This feedback loop resolves the contradiction by maintaining precision while benefiting from the speed of deep learning.
3Manufacturing precision
If standard regularized joint inversion is used for multiple model parameters, then manufacturing precision is improved, but device complexity increases
Solution Approach 1:
The patent combines standard regularized joint inversion with deep learning inversion into a hybrid coupled approach for simultaneously inverting multiple model parameters. The regularized joint inversion ensures manufacturing precision by applying appropriate regularization constraints for each parameter type, while the deep learning component streamlines the computational process. This merging resolves the contradiction by maintaining high inversion accuracy for multiple parameters while reducing overall process complexity through intelligent algorithm integration.
Data Source
AI summary
A method for a physics-driven deep learning-based inversion coupled to fluid flow simulators may include obtaining measured data for a subsurface region, obtaining prior subsurface data for the subsurface region, and obtaining a physics-driven standard regularized joint inversion for at least two model parameters. The method may further include obtaining a case-based deep learning inversion characterized by a contracting path and an expansive path. The method may further include forming the physics-driven deep learning inversion with the physics-driven standard regularized joint inversion, the case-based deep learning inversion, and a coupling operator based on a penalty function. The method may further include forming a feedback loop between the physics-driven standard regularized joint inversion and the case-based deep learning inversion for re-training the case-based deep learning inversion. The method may further include generating an inversion solution for reservoir monitoring.