Efficient power quality identification algorithm based on knowledge distillation and improved GhostNet

Zhenguan Cao

doi:10.1016/j.epsr.2026.113822

Efficient power quality identification algorithm based on knowledge distillation and improved GhostNet

2026-07-16

0

PRE

AI

DOI：10.1016/j.epsr.2026.113822

Original

Original request for help

Save

Abstract

En 中文

• A lightweight power quality disturbance (PQD) identification algorithm is proposed based on knowledge distillation and improved GhostNet. • Gramian Angular Field (GAF) is utilized to transform 1D PQD time-series signals into 2D images for efficient spatial feature mining. • GhostNet is optimized via ECA attention, grouped convolution and channel shuffle, and trained under the guidance of high-robustness DRSN teacher model. • The proposed model reduces FLOPs and parameters by 97.2% and 92.5% respectively with minor accuracy loss under 20 dB noise, adapting to low-computing power scenarios.

Keywords:

Power quality disturbance

Knowledge distillation

Lightweight

GhostNet

Gramian angular field

Deep residual shrinkage network

Efficient power quality identification algorithm based on knowledge distillation and improved GhostNet

Abstract

Journal

Organization