Design Of Antennas Based On Gauss Process Model

At present,most of the problems on antenna are solved by relying on the full wave electromagnetic simulate.However,the use of electromagnetic simulation software to analyze the antenna is complex and spends a lot of time.Therefore,Artificial Neural Network(ANN),Support Vector Machine(SVM)and Gaussian Process(GP)have been proposed to analyze the antenna problem.ANN can achieve parallel processing,self-learning and nonlinear mapping,but its structure is relatively complex and is difficult to be determined.Moreover,its generalization ability is poor.SVM has many unique advantages in solving small sample,nonlinear and other problems,but it is difficult to choose the kernel parameters.Furthermore,predictive output does not have the probability significance.As a machine learning method,GP has developed rapidly in recent decades.It has a good adaptability to solve complex problems such as high dimension,small sample size and nonlinearity.It is also easier to be implemented than SVM and ANN.Its super-parameters can be obtained adaptively and predictive output value also has significance of probability.Therefore,GP model can be used as a rapid alternative to accurate full-wave analysis in antenna design,while accuracy of the model is ensured and time required of antenna design is significantly reduced.On the basis of the existing GP,this paper proposes some GP modeling methods whose modeling effects are more accurate and the time required for optimization is reduced.The main research works are discussed as follows:(1)This paper briefly introduces the basic principles of GP,the implementation of GP modeling,and the evaluation method of GP model,which lays a solid foundation for further research.(2)This paper studies the GP modeling method based on the characteristics of self-updating.Its characteristic is to replace the full wave electromagnetic simulation software using GP model.And the particle swarm optimization algorithm is used to optimize the design of antenna.This scheme can effectively reduce the time required for optimization on the basis of ensuring accurate prediction results.The method is used to optimize the dual frequency microstrip antenna,the antenna can meet the design index on the basis of greatly reducing the optimization time,and the effectiveness of the method is proved.(3)In this paper,a GP modeling method based on kernel function is studied.This paper presents a new kernel function which is defined as the square exponential form,and carries out the numerical simulation based on the polynomial functions fitting.The simulation results show that the proposed kernel function can improve not only the accuracy and efficiency of the model,but also the learning ability and generalization ability of the model.This paper applies the GP based on the kernel function to the design of rectangular dual band microstrip antenna,WLAN dual band monopole antenna,resonance frequency prediction of annular ring microstrip antenna and E-shaped microstrip antenna,and proves the method to be feasible.(4)In this paper,a GP modeling method based on prior knowledge is studied.It is divided into two parts: priori knowledge based on formulas and prior knowledge based on rough grids.In the first part,summed up the relationship between the size of the antenna and the circuit component values and the resonant frequency of the antenna in this paper will be organized into the formula which is as the prior knowledge to establish GP model,optimize the design of the side fed microstrip antenna.In the second part,GP model is established by using the analysis result of the HFSS with coarse mesh as a priori knowledge.The method is applied to the modeling and optimization of conical horn antennas and the optimization results are ideal.(5)In this paper,a method of antenna modeling based on two-stage GP is presented.The method consists of two stages.In the first stage,it studies the mapping between the antenna models of different thicknesses.In the second stage,the practical model of high precision fine model is established,so that the amount of computation required is greatly reduced.Finally,the two-stage Gauss process modeling method is applied to the optimization of the inverted-F antenna and the prediction of resonant frequency of dual frequency PIFA antenna.The mean absolute error,mean square error and mean percentage error are compared by selecting the different proportion of the fine model data to the total training data.The validity and accuracy of the two stage GP antenna modeling method are verified.