Lightweight method for remaining useful life prediction of instrument transformers based on quantization-aware training

Authors

• Wei Zhang, State Grid Jingzhou Power Supply Company ， Jingzhou 434000 ， China
• Jiaona Mao, Substation Maintenance Branch Company ， State Grid Jingzhou Power Supply Company ， Jingzhou 434000 ， China
• Zhuolin Bao, College of Electrical and Information Engineering ， Hunan University ， Changsha 410082 ， China
• Zhuo Long, School of Electrical & Information Engineering ， Changsha University of Science & Technology ， Changsha 410114 ， China
• Xiaofei Zhang, College of Electrical and Information Engineering ， Hunan University ， Changsha 410082 ， ChinaFollow

Keywords

instrument transformer, remaining useful life prediction, quantization-aware training

Abstract

Instrument transformers are key measurement devices in power systems，and their performance directly affects the safe and stable operation of power grids.As deep learning is widely applied to device state prediction，the high computational costs and storage requirements brought by expanding model scales severely restrict their deployment on edge devices.To this end，an adaptive quantization-aware training （QAT ） method，named minimizing quantization error-aware training （MinQuant ），is proposed.Firstly，an adaptive quantizer that automatically adjusts quantization parameters according to data distribution characteristics is designed.Secondly，a quantization error minimization algorithm is proposed to explicitly optimize quantization error during training and achieve collaborative optimization of model performance and quantization precision.Finally，the proposed method is validated on a voltage transformer secondary voltage dataset provided by a power grid company.The results show that the proposed method achieves almost no precision loss under 8-bit quantization and maintains high predic tion precision compared to existing methods in low-bit quantization scenarios such as 4-bit，providing an effective solution for lightweight instrument transformer prediction models.

DOI

10.19781/j.issn.1673-9140.2026.04.012

First Page

136

Last Page

144

Recommended Citation

Zhang, Wei; Mao, Jiaona; Bao, Zhuolin; Long, Zhuo; and Zhang, Xiaofei (2026) "Lightweight method for remaining useful life prediction of instrument transformers based on quantization-aware training," Journal of Electric Power Science and Technology: Vol. 41: Iss. 4, Article 12.
DOI: 10.19781/j.issn.1673-9140.2026.04.012
Available at: https://jepst.researchcommons.org/journal/vol41/iss4/12