Inversion for Equivalent Electromagnetic Parameters of Nonuniform Honeycomb Structures Based on BP Neural Network
Wei-Jia He, Yuxin Zhang, Bi‐Yi Wu, Sheng Sun, Ming‐Lin Yang, Xin‐Qing Sheng · IEEE Antennas and Wireless Propagation Letters · 2024
In this letter, we introduce a backpropagation (BP) neural network-based inversion method for deriving the equivalent electromagnetic parameters of cellular microwave absorbing honeycomb structures. The conventional honeycomb structure is first homogenized into homogenous layers using the Hashin–Shtrikman (H–S) variational theory. Then, the sample honeycombs are generated by sampling the H–S unknown variables using prior knowledge of the physical and geometric characteristics of the honeycomb, and the training dataset are generated by computing the scattered field using the finite element-boundary integral-multilevel fast multipole algorithm. A BP neural network is trained using the scattered field from the sample honeycomb structures as the input, while the output is the undetermined variables for describing the equivalent electromagnetic parameters of the layered homogenous sample honeycomb using H–S theory. Numerical examples are presented to demonstrate the accuracy and effectiveness of the proposed BP neural network for predicting equivalent electromagnetic parameter of microwave absorbing honeycomb structures.