Scholarlink

<p>A novel generalized regression model (GRM) is proposed to characterize the relationship between the activation states of reconfigurable units or their combinations on a reconfigurable partially reflective surface (PRS) and the multi-objective electromagnetic performance metrics of antennas. The GRM framework integrates a forward prediction process and an inverse design process. In the forward process, three parallel generalized regression neural network (GRNN) subnetworks are utilized to model critical electromagnetic performance metrics, including the return loss (S-parameter), gain, and radiation patterns. To improve prediction accuracy and robustness, each GRNN subnetwork is optimized with multiobjective particle swarm optimization (MOPSO) and refined using a recursive correction method. The inverse process integrates the forward predictions to construct an enumeration-based inverse mapping that identifies reconfigurable unit configurations satisfying predefined performance constraints. The efficacy of the model is validated through cross-validation on a liquid-based reconfigurable Fabry–Perot (FP) antenna, demonstrating its ability to significantly accelerate the performance optimization of reconfigurable antenna systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integrated GRM-based efficient multi-performance prediction method for reconfigurable Fabry–Perot antennas

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月7日

查看全文
  • Yuxuan Huang,
  • Zhiming Liu,
  • Duanqi Wang,
  • Zhixin Lei,
  • Jiali Yan,
  • Huilin Zhou,
  • Yuhao Wang

摘要

A novel generalized regression model (GRM) is proposed to characterize the relationship between the activation states of reconfigurable units or their combinations on a reconfigurable partially reflective surface (PRS) and the multi-objective electromagnetic performance metrics of antennas. The GRM framework integrates a forward prediction process and an inverse design process. In the forward process, three parallel generalized regression neural network (GRNN) subnetworks are utilized to model critical electromagnetic performance metrics, including the return loss (S-parameter), gain, and radiation patterns. To improve prediction accuracy and robustness, each GRNN subnetwork is optimized with multiobjective particle swarm optimization (MOPSO) and refined using a recursive correction method. The inverse process integrates the forward predictions to construct an enumeration-based inverse mapping that identifies reconfigurable unit configurations satisfying predefined performance constraints. The efficacy of the model is validated through cross-validation on a liquid-based reconfigurable Fabry–Perot (FP) antenna, demonstrating its ability to significantly accelerate the performance optimization of reconfigurable antenna systems.