Self-Guided Simulation-Driven Optimization for Expensive Filtering Patch Antenna Design and a Benchmark
提出自引导仿真驱动优化框架，通过概率代理模型和自适应罚函数处理二元等式与不等式约束，在300次电磁仿真内实现滤波贴片天线的通带性能与结构可行性，并发布基准测试集。
Filtering patch antenna design optimization involves multiple design requirements, including passband flatness, radiation null existence, and stopband attenuation. The corresponding optimization problem can be rigorously formulated as an expensive constrained optimization problem subject to both binary equality and inequality constraints. The major challenge lies in the high computational cost of electromagnetic simulations, which makes it difficult to identify solutions that simultaneously satisfy all requirements within a limited number of evaluations. To tackle this challenge, we propose a self-guided simulation-driven optimization (SGSO) framework that adaptively guides the search based on simulation feedback. First, we develop probabilistic surrogate models with a relaxation strategy to effectively handle binary equality constraints. Second, we design an adaptive penalty function that incorporates inequality constraints into the objective and dynamically adjusts penalty weights based on recent simulation outcomes, thereby accelerating convergence toward feasible and high-quality designs. Finally, we collect a comprehensive benchmark suite, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">FilterPatchAntenna-Bench</i>, for filtering patch antenna design evaluation. Extensive experiments demonstrate that SGSO consistently achieves superior passband performance and structural feasibility within only 300 electromagnetic simulations. The benchmark and source codes are available at https://github.com/HandingWangXDGroup/SGSO.