Prediction Of Thermal Conductivity Of Porous Media And Structure Generation Based On Machine Learning

The effective thermal conductivity of porous media is a very important physical property of porous media.Porous media with high thermal conductivity can be widely used in a large number of heat dissipation fields,and porous media with low thermal conductivity can be used as good thermal resistance materials.It is important and meaningful to clarify the invisible relationship between effective thermal conductivity and the structure of porous media.At present,the calculation method of effective thermal conductivity of porous media is mainly based on traditional numerical calculations.At present,the more commonly used numerical calculation method is the lattice Boltzmann method,but the lattice Boltzmann method has high calculation cost and slow calculation speed.The machine learning algorithms have made rapid progress since coming in,and some studies have used machine learning algorithms to predict the effective thermal conductivity of porous media and have achieved certain results.Accurately predicting the effective thermal conductivity of porous media is part of the relationship between thermal conductivity and porous media structure.On the other hand,if a given thermal conductivity and porosity are given,how to design a reasonable porous media structure is worthwhile.Analyze the problem.This article first introduces the basic theory of machine learning.It mainly deduces and analyzes the basic machine learning regression algorithms.It uses the QSGS algorithm and lattice Boltzmann method to generate a large number of data to make a data set.The traditional sequential pattern convolutional neural network is improved,and the multi-feature fusion pyramid structure is introduced,so that the neural network not only pays attention to the global structure of the porous medium,but also the local structure of the porous medium.Then use traditional numerical calculation methods and machine learning regression algorithms to conduct experiments on the produced data set and analyze and compare the experimental results.Finally,a conditional generation adversarial network is introduced to generate a reasonable porous medium structure based on the input thermal conductivity and porosity.The basic theory of the conditional generation adversarial network is introduced.According to the structural characteristics of the porous medium,a local structure loss function is introduced to make conditional generation confrontation The network can effectively learn the hidden mapping relationship between local structure and effective thermal conductivity.Experiments were performed on the dataset using the trained conditional generation adversarial network model,and the effective thermal conductivity of the produced porous media was verified using the lattice Boltzmann method.The research in this paper demonstrates the effectiveness of machine learning in predicting the thermal conductivity of porous media and generating the structure of porous media.The average error of machine learning algorithms in predicting thermal conductivity is 1%,and the average error of generating thermal conductivity in porous media structures is 0.9%.Both are lower than traditional numerical calculation methods,and have low calculation cost and fast calculation speed.They can be used to quickly calculate the thermal conductivity of porous media and obtain a reasonable structure of porous media,providing new ideas and assistance for the research of thermal conductivity of porous media.