Recurrent Neural Network-Based Charging Model of Supercapacitor for Far-Field Wireless Power Transfer

Supercapacitors, featured for their high-power storage density, swift response, environmental sustainability, and extended lifespan, play a crucial role in energy storage applications. Despite their widespread use, the challenge of self-discharge, particularly in low-power scenarios like fa-rfield wireless power transfer (WPT) systems, remains significant. Existing models based on parameters like maximum leakage current and equivalent series resistance (ESR) encounter limitations during voltage-variable charging, exacerbated by calculation errors. This paper proposes a novel recurrent neural network (RNN)-based charging model for predicting the charging state of supercapacitors. Trained through constant current experiments, the proposed model demonstrates effectiveness in predicting the voltage of the supercapacitor during charging period and it is validated through an experiment adopting a 915MHz voltage doubled rectifier. The presented model contributes valuable insights for optimizing far-field WPT system performance operating at microwave frequency and beyond.