SODA: A USB-PD Waveform-Based SoC Classification System Using TimeGAN Data Augmentation and Transformer

With the increasing adoption of renewable energy, the concept of the Virtual Grid Hub (VGHub) has been proposed to promote local energy self-sufficiency. Accurate monitoring of the state of charge (SoC) of connected devices is essential for efficient power sharing; however, achieving high estimation accuracy remains challenging due to waveform diversity and measurement limitations. To address these issues, we present the SoC data augmentation (SODA) system, which combines the time-series generative adversarial network (TimeGAN) with a Transformer-based SoC classifier. SODA improves model generalization under limited training data and achieves stable multi-class SoC classification across 3-class, 4-class, 5-class, and 10-class settings for smartphones, tablets, and mobile battery devices. Experimental results demonstrate that SODA effectively supports device-level SoC monitoring in VGHub environments.