Machine-Learning-Assisted Two-Step Antenna Modelling Method

An efficient machine-learning-assisted two-step antenna modelling method (TSAMM) is proposed based on multioutput Gaussian process regression (MOGPR) methods. Variable-fidelity electromagnetic (EM) models are calculated with their responses trained to build two separate MOGPR models. By using asymmetric and symmetric MOGPR methods, mappings between the same and the different responses of the EM models with variable-fidelity are learned. Using the training set with sufficient coarse data and only a few fine data, fine responses can be accurately predicted. The frequency responses of the reflection coefficient and the gain of a quad-resonance substrate integrated waveguide cavity backed slot antenna are modelled to validate the superior of the proposed method.