【作者】 陈晓添； 李金鹏； 彭盛超； 李皓； 肖培； 李高升；

【Author】 Xiaotian Chen;Jinpeng Li;Shengchao Peng;Hao Li;Pei Xiao;Gaosheng Li;College of Electrical and Information Engineering, Hunan University;

【机构】 湖南大学电气与信息工程学院；

【摘要】 转移阻抗是屏蔽蔽线缆的屏蔽效能的一种表征方式,转移阻抗越小则屏蔽效能越强。本文基于全连接神经网络模型,以屏蔽线缆的芯线半径、绝缘层半径、外层绝缘层厚度、编织屏蔽层的单根金属丝直径、编织束数、编织束数中金属丝根数、编织密度这些参数作为模型的训练输入,以屏蔽线缆的转移阻抗曲线作为训练输出,输入层有7个神经元,隐藏层有10个神经元,输出层有101个神经元,采用均方误差评价模型的性能,使用贝叶斯正则化进行训练,使用求解最小梯度值优化迭代隐藏层神经元的权重参数。结果表明:训练完成的神经网络能够根据屏蔽线缆的参数预测其对应的转移阻抗曲线,模型预测结果与商业软件CST Cable Studio的仿真结果进行对比得到的均方误差在10%以内。更多还原

【Abstract】 Transfer impedance is a characterization of the shielding effectiveness of shielded cables, and the smaller the transfer impedance, the stronger the shielding effectiveness. This article is based on a fully connected neural network model, using parameters such as the core wire radius, insulation layer radius, outer insulation layer thickness, single metal wire diameter of the braided shielding layer, number of braided bundles, number of metal wires in the braided bundles,and weaving density of the shielded cable as training inputs. The transfer impedance curve of the shielded cable is used as training output. With 7 neurons in the input layer, 10 neurons in the hidden layer and 101 neurons in the output layer, the performance of the model is evaluated using mean square error. Bayesian regularization is used for training, and the weight parameters of the hidden layer neurons are optimized by solving the minimum gradient value. The results show that the trained neural network can predict its corresponding transfer impedance curve based on the parameters of the shielded cable, and the mean square error obtained by comparing the model prediction results with the simulation results of the commercial software CST Cable Studio is within 10%.更多还原