Optimization Design Of Antenna Based On Deep Gaussian Process

In recent years,electromagnetic simulation software combined with global optimization algorithm is a mainstream method used in antenna design.But due to the long simulation time of the simulation software,this method is very inefficient.Therefore,finding alternative simulation software models for optimization has become a research direction recently,such as artificial neural networks(ANN),support vector machines(SVM),Gaussian Process(GP)and other modeling method.This thesis introduces a network model based on Convolutional Neural Networks(CNN)in Deep Learning(DL)combined with Particle Swarm Optimization(PSO)and GP.This network replaces the fully connected layer of the traditional CNN with the GP model.While retaining the advantages of the CNN and the GP,the convolutional layer and pooling layer in the CNN are used to reduce the dimensionality of the input parameters.While the GP is used to predict the fit,using PSO algorithm optimizes network structure parameters.The modeling method proposed in this thesis can compress the dimensions of the problem,thereby reducing the demand for samples,and effectively improving the modeling efficiency while ensuring the accuracy of the modeling.The modeling method is applied to the design of antennas,and a good optimization design effect is obtained.The deep GP network model given in this thesis can replace the electromagnetic software in the optimization process,and reduce the time required for optimization while ensuring the design accuracy.The main tasks are as follows:(1)Introduce the basic principles of CNN,GP and PSO,and explain the method of calling electromagnetic simulation software HFSS by using Matlab.(2)The structural parameters of the CNN are studied,the PSO algorithm is used instead of the traditional back propagation algorithm(BP)to optimize the CNN,and a PSO-CNN network model is established.The time-consuming full-wave electromagnetic simulation is replaced by the PSO-CNN network model,and the effectiveness of the method is proved by the optimized design of the fragment antenna.(3)Combining GP and CNN,research and establish a deep GP network model(Deep GP Network,DGPN),and use PSO algorithm to optimize the global network.The optimization design of the multi-band microstrip antenna and MPA fed by SIW proves the effectiveness and stability of the method.(4)The CNN is used in the kernel function parameter optimization of the GP model.So the deep GP(DGP)model is studied and established.The UWB monopole antenna with E shape slot and CRLH-TL antenna are used for optimal design to prove the feasibility of the model.