Prediction of active gate drivers' performance using GAN-augmented data

Jin, Tian, Kumar, Manish, Feng, Zhengyang Jenny, Ming, Wenlong ORCID: https://orcid.org/0000-0003-1780-7292 and Wang, Sheng ORCID: https://orcid.org/0000-0002-2258-2633 2025. Prediction of active gate drivers' performance using GAN-augmented data. Presented at: Energy Conversion Congress & Expo Europe (ECCE Europe), Birmingham, United Kingdom, 1-4 September 2025. Proceedings of the Energy Conversion Congress & Expo Europe. 2025 Energy Conversion Congress & Expo Europe (ECCE Europe). IEEE, pp. 1-6. 10.1109/ecce-europe62795.2025.11238950

Full text not available from this repository.

Abstract

Wide-bandgap (WBG) power converters are increasingly employed in renewable energy and electric vehicle systems. The performance of WBG power devices can be improved by deploying active gate driver (AGD) parameters to optimise their switching during operation, resulting in reduced energy loss and current overshoots. However, tuning of AGD control parameters often requires extensive experimental testing and simulation studies, which is both time-consuming and costly. Therefore, this study presents a data-driven prediction approach that leverages Generative Adversarial Networks (GANs) to generate high-fidelity synthetic training data to reduce the need for experimental work and required engineering time significantly. The approach is structured in three stages. First, conduct a limited number of characterisation tests on WBG power devices under varied operating conditions to collect empirical data. Second, a Wasserstein GAN with Gradient Penalty (WGAN-GP) is utilised to produce a set of synthetic control parameters of AGD, thereby augmenting the original empirical dataset. Finally, an artificial neural network (ANN) model is developed and trained using original and synthetic data to correlate AGD parameters and their performance metrics regarding energy loss and current overshoot. The results show that training with WGAN-GPgenerated data improves the AGD's performance prediction model accuracy and generalisation, achieving a higher fitness than training on the original dataset alone.

Item Type:	Conference or Workshop Item (Paper)
Date Type:	Published Online
Status:	Published
Schools:	Schools > Engineering
Publisher:	IEEE
ISBN:	9798331567538
Last Modified:	11 Dec 2025 10:45
URI:	https://orca.cardiff.ac.uk/id/eprint/183125

Actions (repository staff only)

Edit Item