Feature parameters extraction of power transformer PD signal based on texture features in TF representation. Issue 4 (1st July 2017)

• Record Type: Journal Article
• Authors: Zhou, Wenjun; Liu, Yushun; Li, Pengfei; Wang, Yong; Tian, Yan
• Is Part Of: IET science, measurement & technology. Volume 11:Issue 4(2017)
• Page Start: 445
• Page End: 452
• Publication Date: 2017-07-01
• Subjects: feature extraction -- time‐frequency analysis -- partial discharge measurement -- principal component analysis -- support vector machines -- power transformers; feature parameter extraction -- texture features -- time–frequency representation -- ultrahigh‐frequency method -- power transformer partial discharge detection -- time‐domain waveform -- grey‐level cooccurrence matrix -- principal component analysis -- support vector machine classifier -- defect type recognition; Measurement -- Periodicals; Electrical engineering -- Periodicals; Electronics -- Periodicals; Nanotechnology -- Periodicals; Electromagnetism -- Periodicals; Medical instruments and apparatus -- Periodicals; 621.3
• DOI: 10.1049/iet-smt.2016.0342
• Languages: English
• ISSNs: 1751-8822

Abstract

Abstract : Ultra‐high‐frequency (UHF) method is an effective approach to power transformer partial discharge (PD) detection. The feature parameters extracted from UHF PD signal can be applied to insulation defect type recognition. In this study, a novel feature parameters extraction method based on texture features in time–frequency (TF) representation is proposed. PD detections of four typical insulation defects were performed on a 110kV oil‐immersed power transformer. Time‐domain waveform and corresponding TF representations or images of UHF PD signals were obtained. About 36 texture features were extracted from the grey‐level co‐occurrence matrix of TF images. The texture features were reduced into six new feature parameters by principal component analysis. These feature parameters were used as input of the support vector machine classifier for defect type recognition. The recognition accuracies of four kinds of typical defects reached 97.67, 97.00 97.67 and 98.33% proving that the proposed extracted feature parameters are suitable for insulation defect type recognition in power transformer.