基于多层感知器神经网络的波导匹配负载设计

Design of waveguide matched load based on multilayer perceptron neural network

讨论了多层感知器神经网络 (MLPNN)在矩形波导终端匹配短负载设计中的应用。网络学习过程采用反向传播算法 (BP) ,并对训练和测试用样本进行随机化 ,训练过程中加入动量项 ,网络结构可进行自动调节。对样本进行了线性定标 ,用定标后的样本训练神经网络 ,建立系统模型 ,通过优化神经网络相应参数成功实现了矩形波导H面T型结构的终端短小匹配负载的结构设计。

Design of rectangular waveguide terminal matched load based on multilayer perceptron neural network (MLPNN) is presented. Improved back propagation (BP) method is used to train network, training and testing sample are sorted random to make them effective, moment term is added to make the network stable, and the structure of MLPNN can be adjusted automatically. At the same time, input and output sample are scaled linearly to make the MLPNN general and easy to train, network parameters that keep system information are optimized, and H-plane T-kind shorter terminal matched load of rectangular waveguide is designed successfully.