Partial Discharge Detection Method for Distribution Network Based on Feature Engineering

Partial discharge phenomenon of overhead lines in distribution network is usually caused by the concentration of local electric field inside or on the surface of electrical equipment. According to partial discharge problem, based on the characteristics of the project the use of a machine learning is...

Full description

Saved in:
Bibliographic Details
Published in:Journal of physics. Conference series 2023-03, Vol.2456 (1), p.12048
Main Authors: Liu, Hongxiao, Xiang, Mian Xiang, Zhou, Bingtao, Zhu, Li, Duan, Yaqiong, Zhang, Xiaoyan
Format: Article
Language:English
Subjects:
Accuracy
Algorithms
Back propagation
Classification
Discharge
Electric equipment
Electric fields
Feature recognition
Frequency domain analysis
Machine learning
Physics
Power lines
Propagation
Recall
Support vector machines
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Partial discharge phenomenon of overhead lines in distribution network is usually caused by the concentration of local electric field inside or on the surface of electrical equipment. According to partial discharge problem, based on the characteristics of the project the use of a machine learning is proposed for distribution network overhead line partial discharge detection model, first using the characteristics of engineering to extract the signal characteristics of different sides characterization, then respectively using K neighbor algorithm and back propagation algorithm, support vector machine (SVM) classification algorithm test. Experimental results show that when machine learning algorithm is used to classify time domain characteristic signals based on the feature engineering selected in this paper, k-nearest Neighbor algorithm has better classification and recognition effect than back propagation algorithm and support vector machine algorithm, with accuracy rate of 97.20%, recall rate of 96.30% and F value of 96.73%. In the frequency domain feature recognition and classification, the k-nearest Neighbor algorithm has 98.95% accuracy, 99.42% recall rate and 97.61% F value. Compared with the back propagation algorithm and support vector machine algorithm, the K-nearest Neighbor algorithm has the highest detection accuracy in the frequency domain feature detection of partial discharge on overhead lines of distribution network.
ISSN:1742-6588
1742-6596
DOI:10.1088/1742-6596/2456/1/012048