Hybrid Deep Physics Neural Network for Reservoir Simulation

Solution Overview

Problem

Simulating complex physical environments for reservoir engineering and hydraulic fracturing consumes significant time and computational resources, and repeated simulations are often necessary to improve accuracy, leading to further resource consumption.

Innovation Solution

A physical characterization model is trained based on simulated states of a modeled environment, allowing for the prediction of physical characteristics without the need for repeated simulations, using techniques like machine learning and artificial intelligence to map spatial properties temporally across multiple simulated states.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional physics-based simulation methods are used to model complex physical environments, then accuracy of physical characteristic predictions is improved, but computational time and resource consumption increase significantly

Engineering Contradiction:

Improveaccuracy of physical characteristic predictionsVSAvoidcomputational time and resource consumption

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary actions by training the neural network model in advance using simulated states generated from physics-based simulations. Once trained, the model can predict physical characteristics rapidly without requiring repeated full simulations, thus resolving the contradiction between prediction accuracy and computational time.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system creates a computational copy (neural network model) that learns from simulated states and reproduces the behavior of complex physics simulations. This copy can make predictions much faster than running the original physics simulations repeatedly, maintaining accuracy while reducing computational burden.

Inventive Principle:

Principle #26Copying

2Reliability

If repeated simulations are performed to increase model prediction accuracy, then prediction reliability is improved, but computational resource consumption increases

Engineering Contradiction:

Improvemodel prediction accuracyVSAvoidcomputational resource consumption

Core Design Contradiction:

ReliabilityVSUse of energy by moving object

Solution Approach 1:

The system performs preliminary actions by training the neural network model in advance using simulated states generated from physics-based simulations. Once trained, the model can predict physical characteristics rapidly without requiring repeated full simulations, thus resolving the contradiction between prediction accuracy and computational time.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system creates a computational copy (neural network model) that learns from simulated states and reproduces the behavior of complex physics simulations. This copy can make predictions much faster than running the original physics simulations repeatedly, maintaining accuracy while reducing computational burden.

Inventive Principle:

Principle #26Copying

3Measurement precision

If high-resolution grid simulations are used to capture complex physical environment details, then measurement precision is improved, but computational complexity increases

Engineering Contradiction:

Improvedetail resolution of physical characteristicsVSAvoidgrid complexity and simulation complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The neural network model serves as a simplified computational copy that captures the essential behavior of complex grid-based simulations. Instead of running high-resolution simulations repeatedly, the pre-trained model can predict physical characteristics with comparable accuracy but much lower computational complexity.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The system changes the parameters of the computational model from explicit grid-based physics simulations to a neural network representation. This parameter transformation allows the system to maintain prediction accuracy while significantly reducing the computational complexity associated with high-resolution grid simulations.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20220275714A1A hybrid deep physics neural network for physics based simulations

Publication Date: 2022.09.01 LANDMARK GRAPHICS CORP

AI summary

Aspects of the subject technology relate to systems and methods for predicting physical characteristics of a physical environment using a physical characterization model trained based on simulated states of a modeled physical environment. A physical characterization model can be generated based on a plurality of simulated states of a modeled physical environment. Specifically, the physical characterization model can be trained by mapping simulated spatial properties of the modeled physical environment temporally across the plurality of simulated states of the modeled physical environment. Further, input state data describing one or more input states of a physical environment can be received. One or more physical characteristics of the physical environment can be predicted by applying the physical characterization model to the one or more input states of the physical environment.