By Topic

Recognition of partial discharges using an Ensemble of Neural Networks

Sign In

Cookies must be enabled to login.After enabling cookies , please use refresh or reload or ctrl+f5 on the browser for the login options.

Formats Non-Member Member
$31 $13
Learn how you can qualify for the best price for this item!
Become an IEEE Member or Subscribe to
IEEE Xplore for exclusive pricing!
close button

puzzle piece

IEEE membership options for an individual and IEEE Xplore subscriptions for an organization offer the most affordable access to essential journal articles, conference papers, standards, eBooks, and eLearning courses.

Learn more about:

IEEE membership

IEEE Xplore subscriptions

4 Author(s)
Mas'ud, A.A. ; Sch. of Eng. & Built Environ., Glasgow Caledonian Univ., Glasgow, UK ; Stewart, B.G. ; McMeekin, S.G. ; Nesbitt, A.

This paper introduces an improved method for classifying Partial Discharge (PD) patterns using Ensemble Neural Network (ENN) learning. The method is based on training several Neural Network (NN) models and combining their predictions. In this paper it is applied to the recognition of PD from artificially created poly-ethylene-terephthalate (PET) voids and in particular the ability of the ENN to categorise statistical Φ-q-n patterns for two different void sizes over 50 and 250 power cycles. The training data for the ENN comprises statistical parameters obtained from voids of 0.6mm and 1mm diameter. Measurements were made on three separately manufactured void samples for both these diameters. Similarities between the different PD measurements and different cycle captures is investigated using both a Single Neural Network (SNN) and the ENN. For each set of 3 void samples, each NN was trained and tested from the data of one PD void defect. Each NN was then tested using data from two other void geometries in order to determine the recognition abilities of both the ENN and SNN. The results show that the ENN always produces higher recognition efficiency for unseen data when compared to the SNN. It is also shown that ENN produces similar recognition predictions for PD patterns captured using either 50 or 250 power cycles while the SNN shows more sensitivity to the number of power cycles captured.

Published in:

Electrical Insulation and Dielectric Phenomena (CEIDP), 2011 Annual Report Conference on

Date of Conference:

16-19 Oct. 2011