Machine Learning-Assisted Resonance Response Prediction With Data Augmentation

In this paper, a machine learning-assisted prediction model with data augmentation (DA-MLAP) is proposed for accurate resonance response prediction. Traditional multi-layer perceptrons (MLPs) normally struggle to accurately predict S-parameters in regions close to resonant frequencies due to limited data availability. To address this limitation, the DA-MLAP incorporates interpolation-based data augmentation to enhance the quality and diversity of the training dataset, particularly in resonance regions. Initially, a conventional MLP is employed to establish baseline predictions for the frequency and reflection coefficient (ISn!). Subsequently, an augmented dataset created through interpolation is used for a second prediction phase with the MLP. The effectiveness of the approach is validated by comparing predictions with actual| S11| values. Experiments are conducted using a probe-fed patch antenna and a cross-shaped slot frequency selective surface (FSS). Results indicate that the DA-MLAP significantly improves the prediction accuracy of resonant responses compared to traditional MLPs, offering a reliable method for applications in resonant filter, antenna, and single-layer frequency selective surface design.