Neural Network Proxy Model for Reservoir Simulation Speed
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional reservoir simulator models are slow to execute due to the need to solve hundreds to millions of individual equations simultaneously, taking hours to days to complete, which hinders their use in fast-cycle decision support in the oil and gas industry.
Innovation Solution
A proxy model based on a neural network is generated by selecting decision variables, calculating operational base cases, determining values for each time step, and minimizing the number of cases to reduce the run time for the reservoir simulator model, allowing for faster execution while maintaining accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a full physics numerical model (reservoir simulator model) is used to predict future production performance, then prediction accuracy is improved, but execution time increases significantly (hours to days)
Solution Approach 1:
The patent creates a proxy model that is a simplified copy of the full physics numerical model. The proxy model uses neural networks trained on data from the full model to replicate its predictive capabilities while executing much faster, thus resolving the contradiction between accuracy and execution time.
Solution Approach 2:
The patent replaces the mechanical computational system of solving hundreds to millions of equations simultaneously with a neural network-based proxy model. This substitution maintains prediction accuracy while dramatically reducing execution time from hours/days to minutes or seconds.
2Reliability
If the number of equations solved simultaneously is increased to improve model comprehensiveness, then model accuracy is improved, but computational complexity and execution time worsen
Solution Approach 1:
Instead of solving the complete complex system of equations in real-time, the patent creates a proxy model that copies the essential behavior of the full model through neural networks trained on comprehensive data, thereby maintaining reliability while reducing computational complexity.
Solution Approach 2:
The patent performs preliminary computations by training the neural network proxy model on data from the full physics model before actual use. This preliminary action captures the comprehensive model behavior in advance, allowing fast execution without solving all equations during operational use.
3Reliability
If the number of cases generated for model construction is increased to improve proxy model accuracy, then prediction reliability is improved, but training time and computational resources increase
Solution Approach 1:
The patent generates a minimum number of cases that is sufficient to train an accurate proxy model without generating excessive cases. This partial action approach achieves the necessary training accuracy while minimizing training time and computational resources, resolving the contradiction between reliability and training time.
Data Source
AI summary
Systems and methods for reducing run time for a reservoir simulator model using a proxy model based on a neural network.