Reinforcement Learning Surrogate Model for Reservoir Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional physics-based methods for reservoir production optimization are limited by calculation accuracy and time consumption due to the complexity of reservoir geology and multiphase flow uncertainties, requiring iterative calculations of tens of thousands of grids.
Innovation Solution
A reinforcement learning-based decision optimization method for oilfield production systems, which collects dynamic production data to establish a data cube, trains a machine learning model to create a surrogate model for predicting oil production, and uses an evaluation function and constraint models to search for optimal production schemes, quantifying connectivity between injection and production wells.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional physics-based numerical reservoir simulator is used for production optimization, then calculation accuracy can be maintained through complex geological models and multiphase flow mechanism, but time consumption increases significantly due to iterative calculation of tens of thousands of grids
Solution Approach 1:
The patent creates a data-driven surrogate model that copies the input-output relationships of the complex physics-based reservoir simulator without replicating its internal computational structure. This surrogate model is trained on historical data from the numerical simulator and can predict production outcomes rapidly, eliminating the need for repeated iterative calculations while preserving prediction accuracy for optimization purposes
Solution Approach 2:
The patent replaces the physics-based mechanical calculation system (numerical reservoir simulator solving partial differential equations) with a data-driven statistical system (machine learning model). This substitution uses patterns learned from training data to predict reservoir behavior, replacing the computationally intensive physical simulation with a lightweight predictive model that achieves similar accuracy much faster
2Reliability
If traditional physics-based methods are used for reservoir simulation, then comprehensive physical laws can be incorporated, but device complexity increases due to complex geological models and multiphase flow mechanism
Solution Approach 1:
The surrogate model copies only the essential input-output relationships needed for optimization decisions, rather than replicating the full complexity of geological models and multiphase flow mechanisms. This selective copying captures the dominant patterns in production data while eliminating unnecessary computational complexity
Solution Approach 2:
The patent transforms the problem from solving complex physical equations with many parameters to fitting a statistical model with optimized hyperparameters. The machine learning model learns effective parameter relationships from data, replacing the need to explicitly model complex physical processes while achieving comparable predictive reliability
Data Source
AI summary
The present disclosure provides a reinforcement learning-based decision optimization method of an oilfield production system, including: collecting dynamic production data of an oilfield production site to establish a data cube for reservoir production optimization; training a preset machine learning model based on the data cube to obtain a reinforcement learning-based reservoir injection-production system surrogate model configured to predict oil production according to the dynamic production data available on site; constructing an evaluation function for production optimization of a gas injection reservoir; establishing, during a process of production optimization, an enforced constraint model based on input parameters and a boundary constraint condition; and with the constraint model and the boundary constraint condition as constraints, the reservoir injection-production system surrogate model as a basis, and the evaluation function as an optimization direction, searching reservoir production optimization schemes for an optimal production scheme.