南京大学学报（自然科学版）2025，Vol.61Issue(5)：805-815,11.DOI:10.13232/j.cnki.jnju.2025.05.009
基于机器学习的宽带异常折射声学超表面
Machine learning-assisted broadband anomaly refractive acoustic metasurface
摘要
Abstract
Anomalous refraction,key to wave acoustics,can break through the limitations of the conventioned laws of refraction and enables unconventional control of sound wave propagation.Recently,acoustic metasurfaces have shown significant progress in precise sound manipulation,but traditional designs face limitations like narrow bandwidth and strong dispersion,restricting applications in acoustic cloaking and noise reduction.This paper presents a machine learning-assisted broadband anomaly refractive acoustic metasurface comprising 16 subwavelength units with hybrid multiple resonances.Each unit achieves broadband high transmission(>98%)and strong linear phase fitting(>97%),enabling constant refraction angles(Δθ<2.6°)across 1000～4000 Hz.The design employs Gaussian Bayesian optimization with adaptive entropy search portfolio to efficiently explore the high-dimensional(16×6)parameter space,determining optimal parameter configurations within 60 iterations.Numerical simulations validate this approach,offering new possibilities for broadband acoustic manipulation,directional transmission,and cloaking technology.关键词
高斯贝叶斯优化/宽带操控/异常声折射/声学超表面/熵搜索策略Key words
Gaussian Bayesian optimization/broadband control/acoustic anomaly refractive/acoustic metasurface/entropy search portfolio分类
数理科学引用本文复制引用
余玉萍,陈安,杨京,梁彬,程建春..基于机器学习的宽带异常折射声学超表面[J].南京大学学报（自然科学版）,2025,61(5):805-815,11.基金项目
国家自然科学基金(12174190) （12174190）
