High-Precision and Robust DNN Model for Predicting Quality Factor of WPT-Oriented Slotted Ground Resonators

Abstract

Machine learning (ML) has emerged as an effective approach for optimizing circuit design and bringing a paradigm shift in the development of wireless power transfer (WPT) systems. Being the main building blocks of near-field WPT, the slotted ground plane (SGP) resonators with a high quality factor (Q) enhance power transfer efficiency. However, it is pertinent to note that the resonator size, slot shape, and location result in distinct Q outcomes. Therefore, this work delves into the use of ML for predicting Q of SGP resonators. It can be predicted through a deep learning approach, owing to its capacity to learn from the implicit associations between input and output data. Hence, a deep neural network (DNN) model was designed using 20006 data files generated by electromagnetic (EM) simulations. DNN demonstrated its effectiveness, achieving an accuracy of 99.26%, thereby outperforming other benchmark ML models. Furthermore, the model proved its robustness in predicting Q of variously sized resonators and showed 98.3% accuracy. Subsequently, this enables anticipating the Q metric of scaled resonators without the need for exhaustive EM simulations. The predicted Q values were supported through experimental measurements. Finally, the SGP resonators were aptly employed to exhibit the near-field WPT system.

Keywords

• Average comparative error (ACE)
• deep neural network (DNN)
• machine learning (ML)
• magnetic resonant coupling (MRC)
• resonator
• slotted ground plane (SGP)