Abstract

Dissolved gas analysis (DGA) is an essential chemical diagnostic test that can provide information about different types of faults occurring inside a power transformer. However, the accuracy of fault prediction is dependent on the appropriate selection of features. Inappropriate and redundant feature selection increases the computational time and degrades classification performance. Considering the aforesaid issue, this study provides a novel feature selection method termed recursive feature elimination (RFE) for accurate fault classification of power transformers based on DGA data. 31 features were extracted from a total of 1506 DGA data by computing several gas ratios such as the Doernenberg ratio, IEC ratio, Rogers ratio, non-code ratio, etc. Instead of using all of the features, the RFE method was utilized in this study to choose the best features. After utilizing RFE to identify the best gas ratio features, a weighted k-nearest neighbor (KNN) classifier was employed to classify faults. Using the RFE method, it was discovered that the total detection accuracy was 96.84% for eight and 96.57% for twelve features, respectively. The suggested approach may be utilized to diagnose incipient faults in power transformers reliably.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call