Research On Modeling Method Of Microwave Devices Based On HNN

When analyzing the output characteristics of microwave devices using traditional microwave device modeling methods,multiple physical domains in the actual environment must be considered,and the devices must be simulated in multiple physical fields.Microwave device simulation takes a long time and requires high computer performance due to the multiple physical domains involved.As a result,the use of neural networks in microwave device modeling to speed up calculations and lower computational costs is critical for rapid microwave device modeling.In this thesis,we employ a multilayer perceptron neural network to model microwave devices after obtaining the appropriate sample data through multiphysics modeling and simulation of microwave devices.Gradient disappearance may exist if the MLP neural network structure uses only sigmoid as the activation function.To establish the MLP multiphysics model,a two-stage training algorithm with a hybrid of sigmoid and ReLU activation functions is used to adjust the structure of the MLP neural network.In order to improve the accuracy and efficiency of the model,based on MLP multi-physical model and radial basis function network,a microwave device modeling method based on hybrid neural network is introduced,and the HNN multi-physical model is established by using this method.Finally,three microwave devices are modelled,simulated,and analyzed for verification:Ion-sensitive field-effect transistor,tunable evanescent mode cavity filter,and four-pole waveguide filter.The results show that the modeling method of microwave devices based on MLP can save 98%-99%of computational memory and 99%of computational time compared with the finite element method.The microwave device modeling method based on HNN is more suitable for dealing with complex microwave device modeling problems.Compared with the microwave device modeling method based on MLP,it can save about 12%-40%of computational memory and 10%-30%of computational time,and the trained HNN multiphysics model can achieve higher accuracy.Therefore,this method has great advantages in modeling accuracy and calculation time.The modeling method of microwave devices based on HNN is studied by combining MLP neural network and RBF neural network.The simulation findings show that training an HNN multiphysics model with good prediction results only requires a small number of training samples and training parameters.The trained model is more computationally efficient and can save a significant lot of time and memory,providing a theoretical foundation for microwave device modeling research.