Low‐cost surrogate modeling of antennas using two‐level Gaussian process regression method. (27th April 2021)

• Record Type: Journal Article
• Authors: Zhang, Zhen; Jiang, Fan; Jiao, Yaxi; Cheng, Qingsha S.
• Other Names: Vadalà Valeria guestEditor.; Crupi Giovanni guestEditor.
• Is Part Of: International journal of numerical modelling. Volume 34:Number 5(2021)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-04-27
• Subjects: antennas -- Gaussian process regression -- machine learning -- modeling method; Electric networks -- Mathematical models -- Periodicals; Electronics -- Mathematical models -- Periodicals; 621.3011
• DOI: 10.1002/jnm.2886
• Languages: English
• ISSNs: 0894-3370

Abstract

Abstract: In order to improve the accuracy of the surrogate model for antennas, a novel two‐level Gaussian process regression (GPR) modeling method is proposed in this paper. A heuristic hypercube sampling method is proposed using the K ‐means clustering method to generate the training dataset with high uniformity. Based on the training dataset, the first‐level GPR model is established between the design parameters and the full‐wave electromagnetic (EM) simulation responses. The second‐level GPR model is established using the design parameters and the residuals between the first‐level GPR model and the EM simulation model. The sum of the two surrogate models is the two‐level GPR model. The performance of the proposed modeling method is verified by two antenna examples including an ultra‐wideband antenna and a circularly polarized dielectric antenna. Numerical results show that the proposed two‐level GPR method achieves higher accuracy of antenna models than the conventional methods (GPR method and neural networks) with no additional cost. The overall time saving of the proposed method compared to the conventional methods is more than 50% for the majority of our tests.