Analysis And Optimization Of Antenna Performance Based On Machine Learning

As a kind of converter,the antenna can realize conversation between the guided travelling wave in the transmission line and the electromagnetic wave propagating in free space.It plays an important role in wireless communication,radio and television,radar navigation and so on.It has the trend of miniaturization,muiti-band operation and reconfiguration,and the antenna structure also becomes complex and the degree of design freedom increases for the application requirements.Antenna performance analysis is usually based on HFSS,CST and FEKO software.Therefore,the analysis and optimization of antenna performance with multiple design degrees of freedom usually takes a long time and occupies a lot of computing resources.The aided optimization design method based on machine learning can roughly simulate the preliminarily established antenna model based on HFSS and CST to obtain data samples and then use the data samples to train the machine learning model.The trained model can carry out fine simulation within the range of optimization parameters,and combine with the optimization algorithm to realize the rapid analysis and optimization of antenna performance.Based on the machine learning method,this paper deeply studies its application in antenna performance analysis and optimization design.The main work is as follows:Firstly,a planar monopole antenna is designed,and a large number of training samples are obtained by CST.Then,the machine learning method based on the combination of generalized regression neural network(GRNN)and principal component analysis(PCA)is used to quickly analyze the performance of the antenna.This method is also compared with other machine learning methods to verify the effectiveness in antenna performance analysis.In addition,ensemble learning is combined with back propagation neural network(BPNN)and PCA to predict the return loss(S11)and gain of the designed antenna in the whole frequency band.Secondly,the deep learning method is applied to solve electromagnetic problems.Because the electrostatic field problem in two-dimensional finite region is similar to the inverse design problem of planar reflective meta-surface element and antenna array.Based on the convolution neural network(CNN)to solve the two dimensional Poisson’s equation,this paper creatively uses the conditional generative adversarial network(CGAN)to solve this equation in the finite region.Nine secenes are proposed and the training samples are obtained based on the finite difference method(FEM)and used for training CGAN,and then the trained model is used to quickly solve the equation.The prediction performance of CGAN is compared with CNN,and the prediction results show that CGAN can replace FEM to quickly and accurately solve the two-dimensional Poisson’s equation.Finally,variational auto-encoder(VAE)is combined with BPNN to predict the antenna performance in the whole frequency band,and the performance is optimized based on genetic algorithm(GA).Taking monopole antenna as an example,the optimized results indicate that this algorithm can accurately and quickly predict the antenna performance and obtain the optimal structural parameters.Furthermore,this proposed machine learning method is combined with Pareto algorithm to realize the multi-objective optimization of antenna performance.This paper presents a framework for applying machine learning method to antenna performance analysis and optimization design,which can be used as a general solution to this kind of problem.