Features Selection for Partial Discharge and Interference Recognition of HV Cables based on Random Forest Method

Thumbnail Image

Files

CIRED 2019 - 1390.pdf (1.43 MB)

Paper number

1390

Working Group Number

Conference name

CIRED 2019

Conference date

3-6 June 2019

Conference location

Madrid, Spain

Peer-reviewed

Yes

Short title

Convener

Authors

Wang, Ganjun, Zhongshan Power Supply Bureau of the Guangdong Power Grid Corporation- China Southern Power Grid Co.- Ltd., China
Li, Jingshu, Huazhong University of Science and Technology, China
Hu, Yufeng, China Southern Power Grid Co. Ltd., China
Peng, Xiaosheng, Huazhong University of Science and Technology, China
Wu, Yijiang, Zhongshan Power Supply Bureau of the Guangdong Power Grid Corporation China Southern Power Grid Co. Ltd., China
Chen, Yuzhu, Huazhong University of Science and Technology, China

Abstract

Optimal feature selection will contribute to the pattern recognition efficiency and accuracy of partial discharge (PD) signals by removing irrelevant and redundant PD features. Meanwhile, the effective new PD features, selected from a large amount of candidate features, could be applied to the PD based condition monitoring and diagnostic. A novel Random Forest (RF) algorithm based PD feature selection method is presented in the paper. Firstly, the flowchart of RF-based optimal PD feature selection for HV cables and the feature importance measurement of RF method were introduced. Secondly, based on the 5 types of artificial defects of ethylene-propylene (EPR) cables, PD data acquisition and feature extraction were carried out. Subsequently, the RF-based feature selection was applied to the laboratory PD data and was evaluated by the visualization analysis of several top features. The results show that the features which could describe the "fast pulses" and "slow pulses" are effective for PD and interference identification. The wavelet combination features are evaluated to be effective for the recognition of different types of PD signals. RF method is an effective way for PD feature selection, which could potentially be applied to feature selection of other HV apparatus.

Table of content

Keywords

Publisher

AIM

Date

2019-06-03

Published in

CIRED 2019 Conference

Permanent link to this record

https://cired-repository.org/handle/20.500.12455/395
http://dx.doi.org/10.34890/621

ISSN

2032-9644

ISBN

978-2-9602415-0-1