文献类型：期刊文献

中文题名：基于卷积神经网络的多孔材料有效扩散系数预测

英文题名：Prediction of Effective Diffusivity of Porous Material Based on Convolutional Neural Network

作者：宋新宽[1];叶光华[1];周静红[1];周兴贵[1]

机构：[1]化学工程联合国家重点实验室,华东理工大学,上海200237

年份：2018

卷号：34

期号：2

起止页码：97

中文期刊名：化学反应工程与工艺

外文期刊名：Chemical Reaction Engineering and Technology

收录：CSTPCD;;Scopus;北大核心:【北大核心2017】；CSCD:【CSCD_E2017_2018】；

基金：国家自然科学基金项目(21676082);国家重点基础研究计划(973)项目(2014CB239702)。

语种：中文

中文关键词：卷积神经网络;多孔材料;有效扩散系数;预测

外文关键词：convolutional neural network;porous material;effective diffusivity;prediction

摘要：提出了一种利用卷积神经网络预测多孔材料内有效扩散系数的方法,其中多孔材料微观结构的训练样本通过计算机随机模拟生成,对应的有效扩散系数通过有限元方法计算,并联用Matlab和Comsol实现,卷积神经网络训练在单块NVIDIA K80 GPU上进行,训练过程中出现的过拟合现象通过Dropout进行缓解。训练后的卷积神经网络对测试集的预测精确度达96.70%。利用这种方法,能够通过多孔材料的显微图片快速和准确计算其有效扩散系数。
In this work, a convolutional neural network was utilized to train and predict the effective diffusivity of porous material. The microstructure of porous material was generated by python script and the effective diffusivity was calculated by finite element method and using Matlab and Comsol. The training of convolutional neural network was carried on a NVIDIA K80 GPU of Google Co-laboratory, and Dropout method was utilized to reduce the over-fitting during the training process. Finally, the prediction accuracy of the trained convolutional neural network on the test samples reaches as high as 96.70%. This method can effectively and precisely predicts the effective diffusivity of porous material by using its graph information such as scanning electron microscope (SEM) pictures.