Performance evaluation of 2D vs 4D surrogate models of reflectarray unit cells based on support vector regression

Abstract

Surrogate models of reflectarray unit cells are usually generated employing a number of input variables such as geometrical features of the cell, frequency and angles of incidence. Here we show how surrogate models based on support vector regression can be improved by removing their dependence on the angle of incidence. This is done by grouping the reflectarray elements under a relatively small set of incidence angles. Thus, instead of generating models with the angles of incidence as input variables, models are obtained per angle of incidence pair, reducing dimensionality to improve their performance without significant impact on the reflectarray analysis accuracy.