Home // INTELLI 2014, The Third International Conference on Intelligent Systems and Applications // View article

A Comparative Study of Imputation Methods in Predicting Missing Attribute Values in DGA Datasets

Authors:
Sahri Zahriah
Yusof Rubiyah

Keywords: dissolved gas analysis; missing values; imputation methods; gas ratios method; fault diagnosis

Abstract:
Dissolved Gas Analysis (DGA) is one of the most deployable methods for detecting and predicting incipient faults in power transformers. For predicting faults, DGA uses tools such as Doernenburg, Rogers and IEC methods. The presence of missing values in a DGA dataset may affect the diagnostic performances of these three methods. This study applies the mean, regression, expectation-maximization, multiple imputation, and k-nearest neighbor methods to replace the missing values with estimated values in a DGA dataset. Using the number of unresolved diagnoses, the number of wrong diagnoses, and the number of correct diagnoses as the criteria to evaluate the effects of the imputation methods on the DGA diagnostic methods, this study shows that k-nearest neighbor increases the performances of Doernenburg, Rogers and IEC methods the most on two datasets with actual missing values. Experimental results show that imputing missing values in DGA datasets has increased diagnostic performance of the three ratios methods of DGA

Pages: 67 to 74

Copyright: Copyright (c) IARIA, 2014

Publication date: June 22, 2014

Published in: conference

ISSN: 2308-4065

ISBN: 978-1-61208-352-0

Location: Seville, Spain

Dates: from June 22, 2014 to June 26, 2014