Gaussian Process Regression Modeling Based on Landmark Isometric Feature Mapping for Antennas

摘要

Efficient modeling method accelerates computer-aided antenna design. In this paper, a novel Gaussian process regression (GPR) modeling based on landmark isometric feature mapping (LISOMAP) for antennas is proposed to improve the accuracy of modeling. In the GPR-LISOMAP method, LISOMAP, a dimension reduction method, is used to reduce the dimension of data for eliminating useless information. GPR is utilized as the modeling method to establish the relationship between antenna design space (multiple inputs) and response space (multiple outputs). The proposed modeling method is demonstrated by a circularly polarized antenna. Numerical results show that the GPR-LISOMAP method improves the accuracy of antenna modeling.