Research On Numerical Simulation Technology Of Machine Learning Enhancement Of Single-Phase Flow In Porous Media

Porous media are widely used in the production scenarios of oil extraction,which contain complex flow characteristics among various fluids.The study of fluid flow laws existing in reservoir rocks is the basis for increasing oil production and oil recovery,and it is also one of the important research topics in the petroleum industry.Simulating the seepage of porous media also has important practical significance for groundwater protection.Traditional numerical simulation research methods require a lot of computing resources.The requirement of computational resources for the numerical simulation of single-phase flow in porous media greatly limits the application scope and accuracy of this technology in practice.The accuracy of numerical simulation has a strong correlation with simulation scale.Therefore,in view of the connection between scale changes in numerical simulation,this paper proposes a numerical simulation enhancement method based on deep learning,which is based on image super-resolution technology.From solving the numerical simulation of the grid scale difference as a breakthrough,the up-scaling coarse-grid model simulation with faster simulation speed but low precision is used as input,and the connection between it and the high-precision grid data after the scale down-scaling is established.,Predict output by input,improve the running speed and accuracy of the overall numerical simulation.This method describes the correlation between different grid scales in reservoir numerical simulation from the perspective of artificial intelligence.At the same time,it has a certain general applicability and can be extended to other numerical simulation fields.In this paper,the test set of this method is established through the numerical simulation results of single-phase flow based on the finite volume method.After the data set is preprocessed,the coarse grid data with less computational demand is used as input,and it is enhanced to higher Accurate simulation data.Through theoretical research and analysis and test comparisons of actual data sets,it is proved from two perspectives of theory and data testing that this method can enhance the results of numerical simulation algorithms from the perspectives of computing speed and simulation accuracy,and can be applied in many fields in the future.The main tasks are as follows:(1)In the first part of the work,a numerical simulation process of the pressure of single-phase flow in porous media is realized.Aiming at the spatial distribution difference of porous media permeability,the pressure spatial changes under different address conditions are simulated.This part of the work is the basis of the follow-up work and provides theoretical and data support for the follow-up work.Then we visualized the simulation results.On the one hand,we can explain the realization process of numerical simulation more clearly,and on the other hand,we can better match with the enhanced model.(2)In the second part of the work,we conduct research and analysis on the enhancement model.This article mainly chooses SRCNN and SRGAN two models.The selection of these two models directly represents the two types of convolutional neural networks and generative adversarial networks.Super resolution model.In this part of the study,we adjust the model parameters to perform data enhancement tests on the data set established in the first part.(3)In the third part of the work,the accuracy of the second part of the model and algorithm is mainly analyzed.The analysis results show that this method can significantly improve the accuracy of the numerical simulation results of the coarse grid,and then realize the enhancement of the single-phase flow numerical simulation Features.In addition,this part also tests and analyzes the sources of errors and the conditions of use of the model.