基于分类学习器的超材料微带天线高增益性能逆设计

High-Gain Performance Inverse Design of Metamaterial Microstrip Antenna Based on Classification Learning Algorithm

超材料微带天线在通信、雷达等领域有着重要应用。首先,通过仿真获得了852组数据样本,根据与电磁特性的映射关系建立由参数到性能的数据库,得到了852组样本点并验证该样本点的微带天线的增益、E面半功率波瓣宽度和H面半功率波瓣宽度绝对误差均小于2 dB;其次,基于分类学习器和人工神经网络建立样本点的6个结构几何参数与3个性能参数的关系数据模型,得到的训练集、验证集、测试集的均方误差均小于1%;最后,基于人工神经网络数据模型实现了超材料微带天线高增益性能的逆设计,并对逆设计得到的非样本点结果进行天线性能的重分析。结果表明,非样本点的增益性能真实值与验证值的绝对误差不超过1 dB,这也验证了所提设计具有较高的精准性和可行性。

Metamaterial microstrip antennas have important applications in communication,radar and other fields.First,852 sets of data samples are obtained through simulation,and a database from parameters to performance is established according to the mapping relationship with electromagnetic characteristics.The obtained 852 sets of sample points are verified and the absolute errors of gain,E-plane half-power lobe width and H-plane half-power lobe width of the sample points are all less than 2 dB.Then,based on the classification learner and artificial neural network,the relationship data model of 6 structural geometric parameters and 3 performance parameters of the sample points is established,and the mean square error of the obtained training set,verification set and test set is less than 1%.Finally,based on artificial neural network data model,the high gain performance of the metamaterial microstrip antenna is achieved by inverse design,and the antenna performance is reanalyzed for the non-sample point results obtained through the inverse design.The results indicate that the absolute error between the real value and the verified value of the gain performance of non-sample points is less than 1 dB,which also verifies the high accuracy and feasibility of the proposed design.