Electromagnetic Metamaterial Design Based On Deep Learning And Study Of Absorption Characteristics

In recent years,the continuous development of electromagnetic metamaterials has had a profound impact on the field of electromagnetics,realizing functions that natural materials cannot achieve.At present,the electromagnetic metamaterial structure of the electromagnetic metamaterial structure adopts electromagnetic simulation method,but the method has high hardware requirements and requires complex numerical calculations,while the design of electromagnetic metamaterials by deep learning algorithms only needs to learn the characteristic information of the structure to quickly and accurately complete the design of the structure,effectively making up for the shortcomings of the electromagnetic simulation method.According to the optical response of electromagnetic metamaterials,a neural network model is constructed to effectively predict the thickness parameters of multilayer thin film electromagnetic metamaterial structures to meet the needs of structural design and to study the absorption characteristics of structures.In this thesis,a residual neural network model based on AdaBelief optimization algorithm is constructed by taking the multilayer alternating thin film electromagnetic metamaterial structure of graphene and silicon nitride medium constructed as the research object,which effectively predicts the thickness parameters of each layer of the structure by inputting the absorption spectrum of the structure.The absorption spectrum samples of different incidence angles of the structure were obtained by the feature matrix method,and the perfect absorption characteristics of the large incidence angle absorption spectrum in the ultraviolet to near-infrared wavelengths were analyzed.By comparing the prediction results of the network model with other optimization algorithms,the model can achieve higher accuracy,and the relative spectral error of the test sample reaches a minimum of 0.03 within 200 epoches,which meets the goal of structural design,but through comparative analysis,the network model has the defect of slow convergence speed.Aiming at the problems of the above network models,a network model based on adaptive batch normalization algorithm is constructed,which is suitable for predicting the structural parameters of various isotropic and anisotropic electromagnetic metamaterials.A multi-layer alternating thin-film electromagnetic metamaterial structure with anisotropic material black phosphorus as the main body was constructed,and the absorption spectrum samples with different incidence angles of the structure were obtained by the feature matrix method,and the perfect absorption characteristics of the mid-infrared band were analyzed.The performance of the network model was evaluated,and the relative spectral error of the test sample reached a minimum of 0.02 within 150 epoches.Experimental results show that the network model based on the adaptive batch normalization algorithm accurately predicts the thickness parameters of multilayer thin film electromagnetic metamaterial structures with faster convergence speed and stronger generalization performance,and meets the design requirements of electromagnetic metamaterial structures.