Research On Electromagnetic Propagation Method Based On Deep Learning

With the development of deep learning(DL)and the deepening of neural network research,deep learning has been widely used in natural language processing,image target detection,speech recognition and other fields.Over these years,the application of deep learning as a powerful tool in the field of electromagnetics has become more and more important.In particular,when solving some more complex electromagnetics inverse scattering(EMIS)problems and antenna optimization problems,deep neural networks(DNN)and convolutional neural networks(CNN)based on deep learning have broad and attractive application prospects.This paper mainly focuses on the problem of electromagnetic propagation,studies the principles,methods and applications of deep learning.The paper designs four deep neural network structures,which are used to predict the electromagnetic reflection of dielectric sheet with metal underlay and the electromagnetic reflection of the non-uniform dielectric radome.The paper also designed a deep neural network structure,which can be applied to the optimal design of non-uniform dielectric radome.The main contents are as follows:First,the paper introduces the basic knowledge of computational electromagnetics,radome and deep learning,expounds the basic theories of electromagnetic wave propagation and deep neural networks,and introduces the application of deep learning in the field of electromagnetics,which provides a theoretical basis for the research in subsequent chapters.Then,the paper designs two deep neural network structures to solve the electromagnetic reflection of the dielectric sheet with metal underlay under the action of vertically polarized waves and parallel polarized waves.The paper conducted an in-depth analysis of several aspects such as data set,hyperparameter initialization,model training,output results,generalization ability,etc.and simulated and verified the accuracy and effectiveness of the model.In this way,the prediction accuracy of the deep neural network model that has successfully completed the training is over 90%.Next,the paper designs two deep neural network structures to solve the electromagnetic reflection of inhomogeneous dielectric radomes under the two cases of vertically polarized waves and parallel polarized waves.The paper established a deep neural network model,and conducted in-depth analysis on several aspects such as data set,hyperparameter initialization,model training,output results,generalization ability,etc.and verified the accuracy and effectiveness of the model in this paper through numerical simulation.The simulation results show that the prediction accuracy of the deep neural network model that has successfully completed the training is over 90%.Then,the paper talked about the application of deep learning theories and methods to the optimal design of radome.Aiming at the vertically polarized incident wave,the paper studies and designs a deep neural network model applied to the optimization design of a nonuniform dielectric radome.It can accurately predict the optimal thickness of the radome under the conditions of given reflection coefficient,relative permittivity and other design indicators.The optimized design of the deep neural network model can accurately predict the thickness of each layer of the non-uniform dielectric radome,and the accuracy of the prediction results is above 90%.This provides a certain reference value for the design of the radome.Finally,the research work of the full text is summarized,and the further research content is prospected.