Antenna Optimal Design And Material Electromagnetic Parameters Prediction Based On Machine Learning

As an independent basic subject,electromagnetic theory has a wealth of theoretical branches,and its research methods are constantly developing and expanding.Some issues including the optimal design of antennas and the influence of electromagnetic parameters of dielectric materials on the transmission characteristics of electromagnetic waves have been hotspots in the electromagnetic field.Because machine learning possesses great advantages and potential in solving nonlinear problems,the use of machine learning algorithms to predict the nonlinear mapping relationship in electromagnetic problems has broad application prospects.This paper first studies the application of machine learning in antenna optimization design.A large number of rectangular microstrip patch antenna resonance frequency samples are obtained by changing the length and width of the rectangular microstrip patch antenna sample,the dielectric constant and thickness of the dielectric substrate via simulation.Back Propagation(BP)neural network is then used to train the data so as to establish the relationship between antenna structure parameters and resonance frequency.In order to solve the problem that the BP algorithm is greatly affected by initial value and the number of samples,the genetic algorithm(GA)is introduced and can achieve higher prediction accuracy of the antenna resonance frequency.On this basis,we extend antenna patch to any shape,and the resonance frequency and bandwidth within the given size range is accurately predicted by GA-BP algorithm.Also,by establishing the relationship between the antenna structure parameters and performance parameters,the antenna size parameter combination that meets the requirements can be obtained quickly and accurately under the premise of given performance parameters.In addition,this paper further researches the application of machine learning theory in the inversion of electromagnetic parameters.Firstly,the reflection coefficient and refraction coefficient of the dielectric plate with known relative permittivity and permeability are solved by analytical method,and simulation solution is replaced by the accurate analytical solution,which can save the time required to obtain samples by simulation.By introducing neural network to train the large amount of samples,the prediction of the relative permittivity of unknown materials under discrete values is realized.Due to the frequency deviation of the dielectric constant and permeability of the actual material,this paper introduces the Lorentz model and ferrite material respectively,analyzes the mechanism of the change in the dielectric constant and permeability with the change of frequency,and uses Random Forest(RF)algorithm to realize the accurate inversion of material permittivity and ferrite permeability under actual conditions.The algorithm is then further verified in the artificial electromagnetic structure,and the accurate inversion of the electromagnetic parameter spectrum is realized.This paper systematically studies the resonant frequency and bandwidth prediction of the microstrip patch antenna and the influence of electromagnetic parameters on the electromagnetic wave reflection coefficient by combining machine learning theory with electromagnetic problems,and realizes inversion of the dielectric constant spectrum line of the material under the Lorentz model and the magnetic permeability spectrum lines of ferrite materials.The obtained research results have important theoretical and practical significance for the auxiliary optimization of machine learning in antenna design,parameters prediction,electromagnetic parameters inversion and the expansion of machine learning in other electromagnetic related fields.