An Active Learning-Based Adaptive Sampling Method for Efficient Far-Field Modeling of Flexible Antennas

Quantifying uncertainties in the far-field radiation of flexible antennas with surrogate models is essential for maintaining the stability of portable communication systems. However, building accurate surrogate models typically requires extensive training datasets, demanding substantial computational resources and time. This letter presents a novel adaptive sampling method based on active learning (AL) to minimize the training dataset for antenna far-field modeling. The proposed AL framework, called improved expected model change maximization (IEMCM), leverages a support vector regression (SVR) model to predict changes in the far-field model (MC) when new candidate samples are added. The SVR model then selects the most informative candidate sample with the highest MC to label and use in the next modeling iteration. To maintain sampling diversity and global coverage, additional distance features are also incorporated. Tests on a textile patch antenna demonstrate that the IEMCM strategy can significantly reduce the required training set by 24.7% and decrease overall modeling time by 17.4% without compromising model accuracy.