Reservoir Modeling With Diffusion Upscaling From Coarse Simulations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing subsurface modeling techniques face challenges in generating high-resolution simulation outputs efficiently, requiring resource-intensive numerical methods and failing to capture critical geological features due to computational constraints and limitations in grid uniformity and physics-based constraints.
Innovation Solution
A physics-informed machine learning pipeline using a denoising diffusion probabilistic model (DDPM) trained with coarse-grid simulation results and residuals to generate high-resolution reservoir property fields, reducing computational expense and ensuring physical consistency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full fine-grid simulations are performed to achieve high-resolution reservoir models, then modeling accuracy and detail are improved, but computational resource consumption and runtime increase significantly
Solution Approach 1:
The patent uses machine learning models to create a virtual copy that learns from coarse-grid simulations and generates high-resolution reservoir property fields without performing computationally intensive fine-grid simulations. The ML model copies the essential patterns and relationships from low-resolution data to produce high-resolution outputs, eliminating the need for expensive full fine-grid simulations while maintaining accuracy.
Solution Approach 2:
The patent replaces the mechanical numerical simulation system with a machine learning-based system. Instead of using physics-based numerical solvers that require fine-grid discretization and iterative computation, the system uses trained ML models that can generate high-resolution results from coarse-grid inputs, substituting computational mechanics with data-driven prediction.
2Productivity
If coarse-grid simulations are used to reduce computational cost, then resource utilization is improved, but the ability to capture critical geological features deteriorates
Solution Approach 1:
The patent introduces machine learning models as an intermediary between coarse-grid simulations and high-resolution reservoir models. The ML model acts as a mediator that processes coarse-grid simulation results and transforms them into high-resolution property fields, preserving critical geological features while avoiding the computational burden of direct fine-grid simulation.
Solution Approach 2:
The patent changes the fundamental parameters of the simulation approach by using machine learning to map from coarse-grid parameters to high-resolution outputs. Instead of adjusting grid resolution in traditional numerical methods, the system uses ML to transform predictions across resolution levels, maintaining geological feature accuracy while changing the computational parameters.
3Reliability
If traditional numerical methods are used for subsurface modeling, then physics-based constraints are maintained, but runtime performance and real-time capability deteriorate
Solution Approach 1:
The patent performs preliminary training of machine learning models using coarse-grid simulation data and known high-resolution ground truth. During this preliminary phase, the model learns the physics-based relationships and constraints from the training data. Once trained, the model can rapidly generate high-resolution predictions for new scenarios without requiring time-consuming real-time physics-based simulations, thus maintaining reliability while dramatically reducing runtime.
Data Source
AI summary
A method for modeling a subsurface reservoir includes receiving coarse-grid simulation results for the subsurface reservoir, the coarse-grid simulation results being based on a coarse-grid simulation of the subsurface reservoir and having a low resolution that is less than a high resolution. The method also includes generating a reservoir model using a subsurface simulation machine learning model. The subsurface simulation machine learning model is trained to denoise input noise samples using coarse-grid simulation samples as conditioning data to predict high-resolution target reservoir property fields at the high resolution. The high-resolution target reservoir property fields indicate a predicted structure and one or more predicted flow properties for target subsurface reservoirs. The method further includes providing the reservoir model for operating a wellbore based on the reservoir model.