An Efficient Multi-Performance Prediction Approach of Reconfigurable Fabry-Pérot Antennas Designed with Generalized Regression Neural Networks

This paper presents a novel generalized regression model (GRM) for efficient multi-performance prediction and optimization of reconfigurable Fabry-Pérot (FP) antennas. The GRM integrates a forward process, where three parallel generalized regression neural networks (GRNNs) predict S-parameters, gain, and radiation patterns, and an inverse process, which derives optimal activation states of reconfigurable units to meet performance constraints $(\mathrm{e}.\mathrm{g}.,\ S11\leq-10d{B},\ \vert G_{x}\vert \leq\theta)$. Optimized via multi-objective particle swarm optimization (MOPSO), the GRM demonstrates high accuracy and robustness. Validated on a liquid-based FP antenna, the model reduces optimization time by 72% compared to full-wave electromagnetic (EM) simulations, offering a scalable solution for reconfigurable antenna design in wireless communication systems.