In this article, an application of regression-based machine learning (ML) approaches to compute resonant frequency at dominant mode ${\mathit{TM}}_{10}$, slot dimensions of square patch, and patch dimensions of compact microstrip antenna (SPCMA) in the frequency band of 0.4856–7.8476 GHz is presented. In the design process, a squared patch microstrip antenna with two identical slots at the opposite side of a radiating edge of the antenna is loaded. The resonant frequencies of three thousand eight hundred and twenty-two SPCMAs have simulated with CST microwave studio 2019 by varying slot size, the thickness of the material, patch length, and dielectric materials is in accordance with specification of VHF, ULF, L, S, and C band applications. A comparison of 20 regression-based machine learning algorithms including artificial neural network is presented, and it is observed that the Gaussian Process Regression(GPR) model predicts physical or electrical parameters more accurately. The proposed GPR model is validated by fabricating and characterizing a prototype of a microstrip antenna. The fabricated antenna performance is very close to the designed antenna and predicted by GPR.

在本文中，基于回归的机器学习（ML）方法在主导模式下计算共振频率的应用 ${\mathit{TM值}}_{10}$，正方形贴片的缝隙尺寸和紧凑型微带天线（SPCMA）在0.4856–7.8476 GHz频带内的贴片尺寸均已展示。在设计过程中，将加载方形贴片微带天线，该天线在天线的辐射边缘的相对侧具有两个相同的缝隙。通过更改缝隙大小，材料的厚度，贴片长度和介电材料符合VHF，ULF，L，S的规格，使用CST微波工作室2019模拟了382个SPCMA的谐振频率，以及C频段应用。比较了包括人工神经网络在内的20种基于回归的机器学习算法，发现高斯过程回归（GPR）模型可以更准确地预测物理或电参数。通过制造和表征微带天线的原型，可以验证提出的GPR模型。制成的天线性能非常接近设计天线，并由GPR预测。