Evolutionary algorithm for bandstop Frequency Selective Surface

Abstract

In recent years,there has been an increasing interest on the study of Frequency Selective Surfaces, thanks to their particular electromagnetic properties. One among their broad range of applications is band-stop spatial filter, which is constructed by periodic arrays of usually metallic elements on a dielectric substrate. This paper will present the effectiveness of an Evolutionary Algorithm, namely Meta-PSO, in optimizing multi-dimensional electromagnetic problem. All these concepts are integrated into a case study of dual rectangular ring loop-type frequency selective surface, printed on a FR4 substrate

Neural network training schemes for antenna optimization

“…In [4] and [5], sampling target data and using neural networks are two discrete steps. Regarding this "Regular methodology", the desired outputs are obtained by full-wave analysis from formally chosen geometrical inputs in the region Table I In new proposed "Irregular scheme", the geometrical parameters are arbitrarily selected by optimizer to generate different antenna configurations.…”

“…The dual rectangular ring configuration, shown in Figure 1 not only brings more degree of freedom for designers to adjust but also adds more complexity. This structure was first optimized in [2], became the Electromagnetic (EM) object for the comparison of different optimizers in [3]. In order to lessen the computational effort, a fast and accurate model was first introduced in [4], where this simplified equivalence was embedded and directly managed by global optimizers.…”

“…This structure was first optimized in [2], became the Electromagnetic (EM) object for the comparison of different optimizers in [3]. In order to lessen the computational effort, a fast and accurate model was first introduced in [4], where this simplified equivalence was embedded and directly managed by global optimizers. In [4], ANN systematic pattern was trained by a Gradient Descent Method: Error Backward Propagation.…”

“…In order to lessen the computational effort, a fast and accurate model was first introduced in [4], where this simplified equivalence was embedded and directly managed by global optimizers. In [4], ANN systematic pattern was trained by a Gradient Descent Method: Error Backward Propagation. Afterwards, a faster training by the use of Second-Derivative Method: Levenberg Marquandt with splitted Neural Networks was discussed precisely in [5].…”

“…In [4], ANN systematic pattern was trained by a Gradient Descent Method: Error Backward Propagation. Afterwards, a faster training by the use of Second-Derivative Method: Levenberg Marquandt with splitted Neural Networks was discussed precisely in [5]. In this context, a new solution is presented in attempt of reducing computational time more sensibly.…”