BP神经网络在异向介质基本结构分析中的应用

为了减少传统数值分析法由于厚度谐振而引起的结果错误问题,实现异向介质高分析精度与高效率的共存,建立基于反向传播多层前馈型神经网络(BP神经网络)的异向介质电磁特性与介质敏感结构参数之间的神经网络模型,对异向介质的基本结构进行分析。实验结果表明,采用量化共轭梯度法的分析时间为145.535648 s,训练均方误差为0.00020679,所得结果与全波分析相吻合,满足工程要求。有效地克服了再次预测过程中提取参数的不稳定性,为异向介质的分析提供一种快速而准确的方法。

Applications of BP Neural Network in Analyzing Metamaterials Elemental Basic Structure

To reduce the faults caused by thickness resonance in the traditional numerical analysis method, and achieve the coexistence between the high analysis precision and the high efficiency of the Left-Hand Materials(LHMs), the neural network model based on the back propagation multi-layer forward feed neural network is built between electromagnetism characteristic parameters and the dielectric sensitive structural parameters, which is used to analyze the basic Left-Hand Materials structural. The experimental results show that the analysis time is 145.535648 seconds and the training mean square error is 0.00020679 while adopting the scaled conjugate gradient method. The results achieved in the paper are coincident with these ones by the full wave method, satisfying the engineering demand. It effectively overcomes the instability while extracting the parameters in re-forecasting process, and provides a method with rapid and precise for the structure of Left-Handed Materials.