Predictive Proxy Model for Reservoir Simulation Convergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current grid-based models for simulating oil, gas, and hydrocarbon reservoirs face challenges such as long simulation times, limitations in parallelization due to co-dependencies of grid cells, and the impracticality of using brute force techniques to find optimal parameters, leading to inefficiencies in predicting fluid flow and reservoir performance.
Innovation Solution
A computer-implemented method that performs an iterative physics-based simulation using a numerical solver, where an initial estimate for each time step is improved using a predictive model derived from statistical and machine learning processes, allowing for faster convergence and reduced computational resources by predicting initial estimates for subsequent time steps based on time step information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If grid-based models are used to represent reservoirs, then the simulation can capture fluid flow and reservoir performance, but the simulation time becomes very long
Solution Approach 1:
The patent applies preliminary action by using a trained proxy model (neural network) to predict the initial estimate for the next time step before the actual physics-based simulation runs. This predictive model is trained in advance on historical simulation data, allowing it to provide accurate initial conditions for subsequent time steps without requiring full physics-based calculations, thereby reducing overall simulation time while maintaining accuracy.
Solution Approach 2:
The patent creates a simplified copy of the complex physics-based simulation in the form of a proxy model (neural network). This proxy model replicates the essential behavior of the reservoir system but operates much faster. By using this copied model to generate initial estimates, the system avoids repeatedly executing the full complex simulation, thus reducing computation time while preserving predictive accuracy.
2Productivity
If parallel processing is used to improve simulation speed, then computation time is reduced, but hardware limitations and grid cell co-dependencies limit further parallelization
Solution Approach 1:
The patent extracts the initial estimate prediction function from the main physics-based simulation loop and implements it as a separate, pre-trained proxy model. This extraction allows the initial condition generation to be performed independently and in parallel without being constrained by the grid cell co-dependencies that limit parallelization of the full simulation. The proxy model can be applied to multiple time steps simultaneously, reducing overall computation time.
3Manufacturing precision
If brute force techniques are used to find optimal parameters, then all possible parameter combinations can be tested, but the number of combinations makes the approach impractical
Solution Approach 1:
The patent replaces the mechanical brute-force search approach with a machine learning-based proxy model. Instead of systematically testing all parameter combinations through direct computation, the trained neural network learns the relationship between parameters and outcomes from historical data. This substitution allows rapid prediction of optimal parameters without exhaustively testing all possibilities, making the optimization process feasible and time-efficient.
Data Source
AI summary
A computer-implemented method of performing an a numerical solving process using a numerical solver for one or more times steps to obtain a numerical solution for each of the one or more time steps, wherein performing the numerical solving process at each time step comprises providing an initial estimate to the numerical solver and applying the numerical solver to a set of equations representative of a real or virtual process or system, wherein the method comprises, for a current time step of the one or more time steps: obtaining time step information for the current time step: predicting an initial estimate for a subsequent time step using the obtained current time step information wherein predicting the initial estimate uses a predictive model characterised by one or more model parameters that are pre-determined using a statistical and/or machine learning derived process; and performing a numerical solving process using the numerical solver for the subsequent time step thereby to obtain a numerical solution for the subsequent time step, wherein performing the numerical solving process comprises providing at least the predicted initial estimate to the numerical solver.