Application Of Kriging Surrogate Model And Multi-objective Optimization Algorithm In Antenna Design

As the transceiver of radio,antenna is a very important part in wireless communication system.With the popularization of wireless communication in the world and the development of communication technologies,the performance requirements of antenna become higher and higher,and antenna design has become a very challenging task.With the increasing of the dimension of antenna design variables,the manually design method of parameters is no longer applicable.At the same time,there are high requirements for the performance of the antenna in many aspects.When designing the antenna,multiple performance indexes of the antenna need to be considered at the same time.The traditional single objective optimization method can not complete the current antenna design task.Fast multi-objective antenna optimization design method has become the main research direction in the field of antenna design.Firstly,this paper introduces the principle of Kriging model and its application in antenna optimization,and uses Kriging model combined with particle swarm optimization to optimize the design of X-slot on rectangular patch antenna,which verifies the effectiveness of the method.The experimental results show that after using Kriging surrogate model,the demand of PSO for electromagnetic simulation is greatly reduced,a lot of electromagnetic simulation time is saved,and the optimization efficiency is improved.Then the basic principle of NSGA3 multi-objective optimization algorithm is introduced,and the application method of NSGA3 algorithm based on Kriging model in antenna design is described,and the method is used to optimize the return loss and gain of H-shaped patch antenna.Experimental results show that the NSGA3 algorithm can achieve good design results in return loss and gain at a lower cost.Then a new heuristic sampling method based on iteration is proposed,which can effectively solve the problem that the number of samples is difficult to determine,and make full use of the existing sample location and objective function value information.Experiments show that heuristic sampling has a higher sample utilization rate,and the surrogate model is closer to the real optimal solution after optimization.At the same time,the variance of the statistical results of heuristic sampling is smaller and more stable than the traditional method.The effectiveness of this method is verified during the design of an ultra-wideband printed antenna with a fork patch.Finally,policy gradient algorithm in reinforcement learning is used to construct the sampler.The performance of policy gradient sampling and heuristic sampling is compared through experiments.Experiments show that policy gradient sampling can be used to construct samplers,and has good results.The constructed sampler is used in the design task of a dual-frequency monopole antenna,and a complete application method is proposed.