基于循环神经网络的行波管大信号输出特性预测  

Non-linear Output Characteristic Prediction of TWTs Based on Recurrent Neural Network Model

作　　者：李卓芸 殷海荣[1] 李年康 贾栋栋 沈璋 岳玲娜[1] 徐进[1] 赵国庆[1] 王文祥[1] 魏彦玉[1] LI Zhuo-yun;YIN Hai-rong;LI Nian-kang;JIA Dong-dong;SHEN Zhang;YUE Ling-na;XU Jin;ZHAO Guo-qing;WANG Wen-xiang;WEI Yan-yu(School of Electronic Science and Engineering,University of Electronic Science and Technology of China,Chengdu 610054,China)

机构地区：[1]电子科技大学电子科学与工程学院,四川成都610054

出　　处：《真空电子技术》2022年第2期63-66,共4页Vacuum Electronics

基　　金：国家自然科学基金项目(批准号:61771117)。

摘　　要：深度学习是机器学习领域的一个研究方向,已应用于多种人工智能技术的研究。本文基于行波管大信号理论建立了循环神经网络训练模型,将深度学习用于行波管非线性特性的预测,并对行波管的输入参数进行了核函数变换,增强了模型对非线性特征的传递能力。经过训练得到一个适用于8~18 GHz螺旋线行波管的模型,该模型可以预测行波管的输出功率。Deep learning is a subset of machine learning,and it has been applied to many artificial intelligence technologies.A training model of recurrent neural network based on large signal theory of traveling wave tubes(TWTs)is established,with which deep learning is used to predict the nonlinear characteristics of TWTs,and kernel function transformations for the input parameters of TWT is constructed to enhance the nonlinear features transfer ability of the model.After training,a model for 8~18 GHz helix TWTs is obtained,which can predict the output power of TWTs.

关 键 词：行波管 深度学习 循环神经网络 

分 类 号：TN124.2[电子电信—物理电子学]