The Prediction Of Rectangular Microstrip Antenna’s Resonant Frequency Based On BP Neural Network And GA

Microstrip antenna has been widely used in many fields such as wireless communication because it is a thin profile light weight and easy conformal antenna.Rectangular microstrip antenna is the most basic form of microstrip antenna.Resonant frequency is an important technical indicator of microstrip antenna.Get its resonant frequency by using known microstrip antenna structure parameters quickly is a hots pot of modern antenna design method.Because the neural network have many advantages such as adaptive ability and parallel processing ability,it has been widely used in many engineering fields.In the present study, we use a high-frequency electromagnetic simulation software HFSS get a lot of rectangular microstrip antenna samples at first which be used as training samples in subsequent experiments.And then,we set the parameter of BP neural network legitimately and build mathematical model which can reflect the nonlinear relationship between rectangular microstrip patch antenna structure parameters and the resonant frequency of the mathematical with the MATLAB software.When we enter the size of the antenna,we can quickly calculate the corresponding resonant frequency of the microstrip antenna.But BP algorithm is a gradient based method, its network training is influenced by the initial value and the number of sample and easily trapped in a minimum trap, which leads to the failure of network training.In order to solve the above problems, this research has introduced genetic algorithm(GA) based on the optimization group with the global search ability.By combined using the BP and GA algorithm,we solve the problem of using BP algorithm alone in fast calculation of microstrip resonant frequency well.When we using the high frequency electromagnetic simulation software HFSS, BP algorithm and GA algorithm in combination for the fast calculation of the resonant frequency of microstrip antenna, its general process includes:(1) Using HFSS to get training samples antenna based on minute adjusting the size of the microstrip antenna;(2)Establishing the 4-45-1 neural network which can be used to predict microstrip antenna’s resonance frequency;(3) Using GA to optimize the parameters of the model and get the initial weights and thresholds of BP neural network.(4)Giving the best initial weights and thresholds to BP neural network and predict the resonant frequency of microstrip antenna.The studies show: the network which has been optimized by GA can get higher prediction accuracy of of the antenna resonant frequency when we only have limited samples.In the final part of the paper, we use a small amount of samples to establish a rectangular microstrip antenna resonance frequency prediction model which based on BP neural network optimized by GA and calculate the resonant frequency of the rectangular microstrip antenna in a certain size.Finally, we write a query software, when we enter the resonant frequency of the antenna into the software,we can immediately get its corresponding multiple antenna sizes.Verified by using simulation software HFSS,the query speed of the software is fast and the accuracy meets the engineering design requirements of microstrip antenna, so it has a certain application value.