Ensemble-based surrogate modeling of microwave antennas using XGBoost algorithm

Kalayci H., Ayten U. E., Mahouti P.

INTERNATIONAL JOURNAL OF NUMERICAL MODELLING-ELECTRONIC NETWORKS DEVICES AND FIELDS, cilt.35, sa.2, 2022 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 35 Sayı: 2
  • Basım Tarihi: 2022
  • Doi Numarası: 10.1002/jnm.2950
  • Dergi Adı: INTERNATIONAL JOURNAL OF NUMERICAL MODELLING-ELECTRONIC NETWORKS DEVICES AND FIELDS
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aerospace Database, Applied Science & Technology Source, Communication Abstracts, Compendex, INSPEC, Metadex, zbMATH, DIALNET, Civil Engineering Abstracts
  • Anahtar Kelimeler: boosting, ensemble, surrogate modeling, XGBoosting, GRADIENT-SEARCH, FRACTAL ANTENNA, NEURAL-NETWORKS, OPTIMIZATION, DESIGN, CIRCUITS, SENSITIVITY, CLASSIFIER
  • Yıldız Teknik Üniversitesi Adresli: Evet

Özet

With respect to the ever-increasing performance needs in communication technologies, the need for accurate and computational efficient design optimization methods for high-end microwave designs are also increased. Many studies had been proposed for the last decades for creating numerical modeling methods for having high accurate, stable, and computation efficient solutions suitable to be used in the design optimization process. Ensemble learning is a technique that the models are strategically created and combined to solve a specific computer intelligence challenge and primarily employed to boost the efficiency of a model or to lower the risk of a weak learner collection. Herein, XGBoosting-based ensemble learning had been used for having surrogate models for three different microwave designs. In the first and second study cases, two microwave designs from the literature are taken into consideration for testing the performance of the proposed model with existing methods. Furthermore, a novel antenna design had been studied as a third study case with sparse training samples, to test the performance of the proposed modeling technique. As a result, the proposed method had achieved a remarkable performance for all the mentioned study cases both based on its own performance measures and its comparison with the counterpart algorithms.