Research On Electromagnetic Simulation Technology Of Metamaterials Based On Deep Learning

Metamaterials have been widely used in radar,antenna,sensor and other microwave devices due to their special electromagnetic properties.In the traditional design and simulation of metamaterials,it is often necessary to use HFSS,CST and other electromagnetic simulation software for complex modeling and a large number of numerical calculations.How to quickly and accurately analyze metamaterials has become a major problem in the field of metamaterials research.In recent years,some scholars have proposed the method of using deep learning to construct neural network to calculate the characteristic parameters and design the structure of metamaterials,which provides another idea for metamaterials design and simulation.Using metamaterials deep learning design is a more general way,don't need to consider the modeling process of metamaterials,also do not need the various parameters of the electromagnetic calculation,only need to learn different structure and the corresponding characteristics of metamaterials characteristics calculation and structure design,the metamaterials research has the very strong practical significance.In this paper,the properties of metamaterials and the traditional electromagnetic numerical method are introduced,and the idea of studying metamaterials structure and properties based on deep learning is studied.For arrays of periodic metamaterials structure,its abstract for the "0","1" in the form of matrix,the use of Matlab simulation method with CST microwave studio on the modeling and simulation,model structure and its corresponding S parameters as sample data sets,and then applying the idea of deep learning build convolution neural networks,analyzes the convolution layer depth,pooling method,activation function and the optimizer to choose affect the performance of the neural network,using the GPU acceleration method of part of the sample data set is obtained by simulation training,so as to build a metamaterial structure and the mapping relationship between the characteristic parameters,Finally,the established mapping relation is used to verify the training set in the sample data set,so as to achieve the effect of the calculation of metamaterials' characteristic parameters and lay a foundation for the reverse design of the structure through metamaterials'characteristics.