Partial Discharge Patterns Recognition of GIS with Denoising-stacked Autoencoder Networks

Partial discharge (PD) is the main characterization of gas insulated switchgear (GIS) insulation defects, which will further aggravate equipment aging. Therefore, monitoring the PD of GIS equipment is of great significance to detect insulation defects and avoid GIS equipment failure to ensure safe a...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2020 5th Asia Conference on Power and Electrical Engineering (ACPEE) S. 1815 - 1818
Hauptverfasser: Zhao, Yiming, Yan, Jing, Wang, Yanxin, Liu, Tingliang, Jiang, Junjie
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 01.06.2020
Schlagworte:
Classification algorithms
denoising-stacked autoencoder
Feature extraction
gas insulated switchgear
partial discharge
Partial discharges
Pattern recognition
Stacking
Support vector machines
Training
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Partial discharge (PD) is the main characterization of gas insulated switchgear (GIS) insulation defects, which will further aggravate equipment aging. Therefore, monitoring the PD of GIS equipment is of great significance to detect insulation defects and avoid GIS equipment failure to ensure safe and reliable operation of the grid. However, the traditional partial discharge pattern recognition mostly relies on artificial feature engineering, and the appropriateness of feature selection directly affects the recognition result. This paper proposes a pattern recognition classifier that directly and automatically selects and classifies fault features by denoising-stacked autoencoder. Automatic feature extraction effectively reduces the dependence of traditional pattern recognition classification algorithms based on expert systems and excessive human intervention. The results show that it not only inherits the advantages of the generalization ability of the denoising autoencoder model, but also has the advantages of easy stacking, faster convergence and higher accuracy.
DOI:10.1109/ACPEE48638.2020.9136370