Submit

Login Login

Register Register

Home / Journals / CMC / Online First / doi:10.32604/cmc.2026.085939
Journal Menu
Special Issues
Table of Content

All issues

Online First

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Open Access

ARTICLE

An Improved Res-Swin Fusion Network for Partial Discharge Pattern Recognition in Power Cables

Zihao Jia1, Jingrui Zhang1,2,*
1 Department of Instrumental and Electrical Engineering, Xiamen University, Xiamen, China
2 Shenzhen Research Institute of Xiamen University, Xiamen University, Shenzhen, China
* Corresponding Author: Jingrui Zhang. Email: email

Computers, Materials & Continua https://doi.org/10.32604/cmc.2026.085939

Received 21 May 2026; Accepted 10 August 2026; Published online 10 September 2026

Abstract

Accurate partial discharge (PD) pattern recognition is essential for assessing power cable insulation, yet it remains challenging due to complex discharge features and limited data availability in practical engineering. To address these issues, this paper first establishes a high-voltage experimental platform with four carefully designed physical defect models to construct a highly authentic Phase-Resolved Partial Discharge (PRPD) dataset. Subsequently, an improved Res-Swin fusion framework is proposed. It integrates a ResNet branch to decouple fine-grained local discharge textures and a Swin Transformer branch to model global phase-amplitude topologies. To alleviate overfitting in data-scarce scenarios, a multi-stage parameter freezing strategy is implemented. Furthermore, an Adaptive Chaotic Lévy-flight Sand Cat Swarm Optimization (ACLSCSO) algorithm is introduced to globally optimize branch-wise initial learning rates for the heterogeneous network. Extensive cross-validation experiments demonstrate that the proposed method achieves an average accuracy of 97.54%. The freezing strategy significantly reduces training time by 11.5%, while the synergistic optimization resolves the confusion between morphologically similar patterns, reducing the misclassification rate of “No Partial Discharge” as “Void Discharge” from 6.2% to 2.1%. Finally, attention visualizations show that the network’s focal regions are broadly consistent with characteristic PRPD regions, providing qualitative interpretability evidence for insulation condition assessment.

Keywords

Deep learning; partial discharge pattern recognition; PRPD; Res-Swin; swarm optimization

  • 98

    View

  • 18

    Download

  • 0

    Like

Related articles
Copyright© 2026 Tech Science Press
© 1997-2026 TSP (Henderson, USA) unless otherwise stated

Share Link