改进人工神经网络算法及其在E面分支波导耦合器优化设计中的应用  

Improved Artificial Neural Network Used in Design and Optimization of E-Plane Branch-Waveguide Couplers

作　　者：许殿[1] 史小卫[1] 

机构地区：[1]西安电子科技大学理学院,西安710071

出　　处：《微波学报》2005年第4期16-19,共4页Journal of Microwaves

基　　金：国防预研基金资助项目(11001030103)

摘　　要：将混合遗传算法用于人工神经网络,训练出全局最优的权值和偏差,解决了反向传播网络收敛于局部极值的问题。运用该方法训练出E面分支波导耦合器的输入输出人工神经网络模型,并以此仿真并优化其他结构的耦合器。相对于精确电磁场数值计算,前者在保证有较高仿真精度的前提下,大大提高了仿真速度。ANN combined HGA makes the weight and the deviation converge to the global solutions, which escapes local optima of back-propagation network. The network models of E-plane braneh-waveguide couplers are trained by this method and it is simulated for some couplers of different dimensions. Based on this ANN, an excellent construction coupler is achieved. Comparing to numerical computation technology for electromagnetic field, the results show that HGA-BP is accurate and efficiency.

关 键 词：人工神经网络 反向传播网络 混合遗传算法 E面分支波导耦合器 人工神经网络算法 波导耦合器 优化设计 E面 人工神经网络模型 应用 仿真精度 

分 类 号：TN622.2[电子电信—电路与系统]