基于BP神经网络的圆形孔缝耦合截面预测

在对电磁波经孔缝传输/泄露的分析中，孔缝耦合截面的获取十分重要。针对现有公式无法准确获取谐振频段圆形孔缝耦合截面的问题，将BP神经网络应用于圆孔耦合截面的快速获取，该模型适用于电尺寸(半径波长比)在0.08, 3]之间的圆形孔缝。在不同入射角度和极化角度的入射波辐照下，用全波分析软件计算无限大理想导体平板上不同电尺寸圆孔的耦合截面，用圆孔的耦合截面除以其几何面积得到圆孔的归一化耦合截面。利用这些数据训练神经网络，建立了一个以圆形孔缝的电尺寸、入射波的入射角度和极化角度为输入参数，孔缝的归一化耦合截面为输出参数的BP神经网络模型。通过与全波分析的对比可知，该模型能够快速准确地预测任意入射角与极化角平面波辐照下电尺寸在0.08, 3]之间的圆形孔缝的归一化耦合截面。

Prediction of coupling section of circular aperture based on BP neural network

It is important to obtain the coupling section(CS) of aperture for analyzing aperture penetration/leakage of electromagnetic wave. However, there are no ready-made formulas for calculating CS of circular apertures in the resonance region. Therefore, a backpropagation(BP) neural network is applied to the fast acquisition of CS of circular apertures with electrical dimensions (ratio of radius and wavelength) of 0.08, 3]. Full-wave analysis software is used to calculate the CSs of circular apertures with different electrical dimensions on an infinitely large, perfectly conducting plate illuminated by plane-waves at different incident angles and polarization angles. By dividing the CS by the geometric area of the aperture, the normalized CS is obtained. And then, the full-wave analysis data are used as training data for a BP neural network model, including aperture electrical dimension, incident angle and polarization angle as input and normalized CS as output. Test results suggest that this model can quickly and accurately predict the normalized CSs of circular aperture with electrical dimensions 0.08, 3] under a plane-wave illumination at any incident angle and polarization angle.