An Electromagnetic Parameter Retrieval Method Based On Deep Learning

The traditional Nicolson-Ross-Weir(NRW)method has the characteristics of wide frequency band and high accuracy,but it has problems such as half-wave resonance,phase angle jump,and multi-value.Among them,the popular solution to the multi-value problem has disadvantage such as complex calculations and strict conditions,which have many limitations in actual measurement.Therefore,studying a simple and accurate electromagnetic parameter retrieval method has important application value.This article briefly describes the principle of electromagnetic parameter retrieval,as well as the classification and current research status of retrieval methods,and conducts an in-depth analysis of the most commonly used NRW method and its existing problems,and proposes a GRU neural network-based electromagnetic parameter retrieval method.The main research contents and results are as follows:(1)An electromagnetic parameter retrieval method based on deep learning is proposed,which solves the multi-value problem in the traditional NRW method.This is the first time that deep learning has been applied to electromagnetic parameter retrieval.In order to make full use of the sequence information of the parameters,an retrieval model is established using gated recurrent unit(GRU).Compared with the group delay method and the imaginary part compensation method which is the main solution to the multi-value problem of the NRW method,it has the advantages of looser requirements for measurement conditions and higher calculation accuracy.(2)A method to automatically collect data sets through CST simulation tools is proposed.By writing VBA scripts in CST,simulation projects are automatically run in batches,and data sets related to electromagnetic parameters and parameters are collected.The collected data is then preprocessed,key feature values are extracted,and then combined with the corresponding electromagnetic parameters to form a sample data set for training the deep learning model.Train the network model through the processed data set.The experimental results show that the model proposed in this paper can find the internal law between the parameters and their electromagnetic parameters.It can automatically output electromagnetic parameters according to the S-parameters.Compared with the traditional NRW method,the mean square error is reduced by 48%,and there is no problem of multi-value and phase angle jump.It has application value for the study of electromagnetic properties of complex materials.