Extreme Learning Machine Based On Gravity Search Algorithm For Electromagnetic Problem Optimization

With the rapid development of information technology,the mobile internet gradually replaces the traditional computer network and becomes the mainstream of communication system.As the core component,the microwave component plays an increasingly important role in the communication system.When designing microwave component,professional software such as HFSS and IE3 D can be used to carry out full-wave electromagnetic simulation and combine with global optimization algorithm.Since electromagnetic simulation requires a lot of grid calculation,it will consume a lot of time.Therefore,an alternative model can be used to reduce the calculation cost.This thesis uses extreme learning machine as an alternative model,and introduces an improved gravitation search algorithm to solve the shortcomings of extreme learning machine.The main works are as follows:(1)In view of the classic gravitational search algorithm search ability is not strong,the particle update strategy is introduced.In each iteration all the particles are rearranged according to the fitness value,30% worst particles are updated,and taking the better particles before and after the update and the remaining 70% particles form a new group to avoid the possibility of falling into local optimum.The classical gravitation search algorithm and the improved gravitation search algorithm are used to deal with numerical simulation and linear array synthesis problems at the same time.It can be found that the improved gravitation search algorithm can achieve better optimization results,which proves the effectiveness of the improved algorithm.(2)For the instability problem of the classic extreme learning machine due to the random input weights and bias of hidden layer,the improved gravitational search algorithm is introduced to optimize them,putting input weights and bias of hidden layer as the particle to optimize,the test error of extreme learning machine function as fitness value,called the GSA-ELM algorithm.The GSA-ELM method is used to model the resonant frequencies of rectangular and E-shaped microstrip antennas,and the results are compared with the existing results,which verify the effectiveness of the proposed method.(3)Due to the difficulty of determining the regularization coefficient and the hyper parameter of kernel function of extreme learning machine,the improved gravitation search algorithm is used to determine the optimal values,putting the regularization coefficient and the hyper parameter of kernel function as the particle to optimize,the test error of kernel extreme learning machine as fitness value,called the GSA-KELM algorithm.By the use of GSA-KELM method on modeling the response of the microstrip filter and the rectangular monopole antenna,the improved gravitation search algorithm is used to optimize the kernel extreme learning machine to get the optimal dimension parameters.Comparing the results of other ways to get the best dimension parameters of microstrip filter and rectangular monopole antenna,it can be found that them designed by this method are more consistent with the requirements of design indexes,which proves the effectiveness of this method.