Machine learning-based surrogate modelling of reflectarray unit cell in a 4-d parallelotope-shaped domain

Abstract

A novel strategy to define a high-dimensionality parallelotope-shaped domain is proposed to train surrogate models of reflectarray unit cells. The concept is based on the definition of a region or rectangle of stability where sharp resonances are avoided as much as possible. A 4-D parallelotope is defined around the rectangle of stability and by controlling its size, it is possible to avoid new resonances that otherwise would appear as a consequence of increasing a rectangular domain dimensionality. This methodology is applied to generate support vector regression based models of a multi-resonant unit cell. Results show a high degree of agreement between the obtained surrogate models and simulations using a tool based on the method of moments with local periodicity that was, in turn, used to generate the training samples. Results also prove that the proposed method performs better than lower dimensionality methods for wideband optimization.