SE-ResNet18 Permeability Prediction for Multi-Mineral Digital Cores
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting shale oil permeability, such as molecular dynamics simulation, pore network modeling, and core analysis, are computationally intensive and costly, and lack accuracy due to the complexity of nanoscale pore structures in shale formations.
Innovation Solution
A deep learning-based method using convolutional neural networks (CNNs) to predict permeability by constructing a three-dimensional digital core, segmenting images, and training an SE-ResNet18 network with multi-mineral digital core data sets to simulate fluid flow and calculate permeability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If molecular dynamics simulation is used to simulate fluid flow in porous media, then the simulation can capture nanoscale pore structure effects, but the computational workload becomes huge and the simulation scale becomes too small for practical application
Solution Approach 1:
The patent creates a digital core model that copies the essential features of the physical core's pore structure at a reduced scale. This digital replica allows for efficient simulation and analysis without requiring computationally expensive molecular dynamics simulations on actual nanoscale structures, thus maintaining accuracy while dramatically improving calculation efficiency
Solution Approach 2:
The patent replaces the physical molecular dynamics simulation system with a computational neural network system. The neural network learns from training data and substitutes the need for repeated molecular dynamics simulations, transforming a computationally intensive mechanical/physical process into an efficient information processing task
2Measurement precision
If direct numerical simulation is used to solve partial differential equations for fluid flow, then the solution precision is higher, but the calculation amount increases due to large matrices and iterative solutions
Solution Approach 1:
The patent performs preliminary training of the neural network using training data that incorporates results from direct numerical simulations. This preliminary action allows the network to learn the complex relationships and patterns, so that subsequent predictions can be made rapidly without repeating the time-consuming iterative solution processes
Solution Approach 2:
The neural network creates a computational copy of the direct numerical simulation capability, learning to predict outcomes without actually solving the partial differential equations each time. This copying approach maintains the precision benefits while eliminating the repetitive calculation time
3Measurement precision
If special core analysis is conducted to obtain detailed rock characteristics, then the permeability measurement is more accurate, but the process becomes time-consuming and expensive
Solution Approach 1:
The patent uses digital core models that copy the essential pore structure and mineralogical characteristics of physical cores. These digital replicas allow for rapid virtual analysis that maintains the accuracy benefits of detailed core analysis while eliminating the time and cost of physical laboratory procedures
Solution Approach 2:
The patent replaces the physical special core analysis system with a computational neural network system. The neural network processes digital core images and predictions, substituting time-consuming laboratory measurements with rapid computational analysis that maintains measurement accuracy
4Adaptability or versatility
If routine core analysis is performed with frequent sampling to capture fine stratification characteristics, then the representation of rock types improves, but the number of samples required and processing time increase
Solution Approach 1:
The patent segments the core analysis into digital image processing and neural network prediction components. This segmentation allows the system to handle fine stratification characteristics through image resolution rather than physical sampling frequency, improving rock type representation while maintaining sampling efficiency
Solution Approach 2:
The digital core model creates a complete virtual copy of the core's stratification and rock type distribution. This digital copy allows for unlimited virtual sampling and analysis without requiring additional physical samples, thus improving representation while maintaining sampling efficiency
Data Source
AI summary
Disclosed a method for predicting permeability of a multi-mineral phase digital core based on deep learning. In the present disclosure, a three-dimensional digital core is constructed and a pore structure is randomly generated; after a plurality of multi-mineral digital core images is acquired by performing image segmentation on the three-dimensional digital core, permeability corresponding to each of the multi-mineral digital core images is acquired by simulation using multi-physics field simulation software and a multi-mineral digital core data set is constructed based on the plurality of multi-mineral digital core images and the permeability corresponding to each of the multi-mineral digital core images; an SE-ResNet18 convolutional neural network is constructed and trained with the multi-mineral digital core data set; and an image of a multi-mineral core to be predicted is input into the trained SE-ResNet18 convolutional neural network to determine the permeability of the multi-mineral core.