基于神经网络的圆孔阵列垂直入射耦合截面预测

Prediction of Vertical Incident Coupling Cross Section of Circular Aperture Array Based on Neural Network

为了分析电子系统金属屏蔽外壳上孔阵的耦合截面,引入神经网络建模方法。分析了垂直入射条件下一些特征参数对圆孔阵列耦合截面的影响,用全波分析法计算了不同特征参数下的孔阵归一化耦合截面,将获得的6000组样本数据输入神经网络进行训练,获得根据孔单元的电尺寸、行/列数、行/列间距电尺寸及入射波的极化角度6个参数预测孔阵归一化耦合截面的神经网络模型;将样本按比例随机分为训练集和测试集,得出最少约3000组数据就能使神经网络模型达到较高的预测精度。为了进一步验证神经网络模型的普适性和有效性,选取了2组没有在训练集和测试集中出现的特征参数,分别用全波分析法和该神经网络模型进行预测,发现基于神经网络的预测结果和全波分析法的计算结果吻合良好。

In the increasingly complicated external electromagnetic environment,in order to analyze the coupling cross section of the aperture array on the metal shield shell of electronic system,a neural network modeling method is introduced.Firstly,the influence of some characteristic parameters on the coupling section of the circular aperture array under the condition of vertical incidence is analyzed.Then,the full wave analysis method is used to calculate the normalized coupling cross section of the aperture array under different characteristic parameters,and a total of 6000 sets of sample data are obtained.The 6000 sets of sample data are input into the neural network for training,and a neural network model is obtained to predict the normalized coupling cross section of the aperture array according to six parameters including the electrical dimension of the aperture unit,the number of rows/columns,the electrical dimension of the row/column spacing and the polarization angle.In addition,6000 sets of samples obtained are randomly divided into the training set and the test set according to the proportion.And it is concluded that a minimum of about 3000 sets of data can make the neural network model to reach the desired prediction accuracy.In order to further verify the universality and effectiveness of neural network,two groups of characteristic parameters that don’t appear in the training set and the test set are selected,and the prediction results based on the neural network model and the calculation results of full wave analysis method are found to be in good agreement.