Table 2 Benchmark modeling methods.
From: Two-stage variable-fidelity modeling of antennas with domain confinement
Modeling approach | Domain | Setup |
|---|---|---|
Kriging interpolation | Conventional (parameter space X) | Gaussian correlation function, second-order polynomial used as a trend function |
Radial basis functions (RBF) | Conventional (parameter space X) | Gaussian correlation function, scaling coefficient determined through cross-validation |
Nested kriging81 | Confined domain XS | Thickness parameter T = 0.05 (Antennas I through III), 10 reference designs used in all cases, the numbers of EM simulations required for their identification: • Antenna I (acquisition cost 864 EM simulations) • Antenna II (acquisition cost 930 EM simulations) • Antenna III (acquisition cost 1899 EM simulations) |
Reference-design-free modelling65 | Confined domain XS | Thickness parameter T = 0.05 (Antennas I through III), the following numbers of EM simulation used to identify Nr = 50 accepted observables: • 106 (Antenna I) • 230 (Antenna II) • 192 (Antenna III) |
