Accurate Modeling of Frequency Selective Surfaces Using Fully-Connected Regression Model With Automated Architecture Determination and Parameter Selection Based on Bayesian Optimization

Surrogate modeling has become an important tool in the design of high-frequency structures. Although full-wave electromagnetic (EM) simulation tools provide accurate account for circuit characteristics and performance, they entail considerable computational expenditures. Replacing EM analysis by fast surrogates provides a way to accelerate procedures. Unfortunately, microwave passives is challenging task due their highly-nonlinear outputs. Frequency selective surfaces (FSSs) constitute representative example with multi-resonant reflection transmission responses that need be represented over broad frequency ranges. Deep neural networks (DNNs) seem promising techniques handling such cases. However, serious practical issue associated employment appropriate selection model parameters, including its architecture. A common practice experience-driven setup, heavily based on trial error, which does not guarantee optimum determination may lead multiple problems as poor generalization or high variance predictive power respect training data set selection. This paper proposes novel framework, referred fully-connected regression (FCRM), where crucial role played Bayesian Optimization (BO), incorporated determine DNN-based both architecture hyperparameter values, fully automated manner. For validation, FCRM applied construct Minkowski Fractal-Based FSS. The efficacy methodology demonstrated through comparisons several benchmark techniques, DNN established using traditional methods well conventional models. numerical results indicate exhibits considerably improved prediction reduced sensitivity sample assignment.