期刊
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION
卷 69, 期 4, 页码 2302-2315出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TAP.2021.3051034
关键词
Antennas; Optimization; Computational modeling; Training; Predictive models; Sociology; Base stations; 5G base station antenna; antenna design; complex antenna; computationally expensive optimization; differential evolution; Gaussian process; radial basis function; surrogate model
资金
- MathWorks Development Collaboration Research Grant
In this article, a new method called TR-SADEA is proposed for complex antenna optimization by reducing training time and increasing convergence speed. Experimental results demonstrate significant improvements in training time and iteration numbers compared to a state-of-the-art method, while maintaining high antenna performance.
Surrogate models are widely used in antenna design for optimization efficiency improvement. Currently, the targeted antennas often have a small number of design variables and specifications, and the surrogate model training time is short. However, modern antennas become increasingly complex, which needs much more design variables and specifications, making the training time become a new bottleneck, i.e., in some cases, even longer than electromagnetic (EM) simulation time. Therefore, a new method, called training cost reduced surrogate model-assisted hybrid differential evolution for complex antenna optimization (TR-SADEA), is presented in this article. The key innovations include: 1) a self-adaptive Gaussian process surrogate modeling method with a significantly reduced training time while mostly maintaining the antenna performance prediction accuracy and 2) a new hybrid surrogate model-assisted antenna optimization framework that reduces the training time and increases the convergence speed. An indoor base station antenna with 2G to 5G cellular bands (45 design variables and 12 specifications) and a 5G outdoor base station antenna (23 design variables and 18 specifications) are used to demonstrate TR-SADEA. Experimental results show that more than 90% of the training time and about 20% iterations (simulations and surrogate modeling) are reduced compared to a state-of-the-art method while obtaining high antenna performance.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据