A New Multi-Path Hybrid Classifier for Transformer Oil Fault Diagnostic

Abstract

This work aims to provide advances in diagnosis algorithms using intelligent techniques and represents an application in fault detection and classification in oil-immersed power transformers. The paper proposes a new methodology of classification using hybrid algorithms to describe an improved DGA diagnostic tool based on combining different classifiers and several input vectors. A total of six classes of electrical and thermal faults are labeled. For each fault, binary classifications are first conducted using two classifiers trained and evaluated using nine different input vectors. For this, a dataset of 501 samples is used, and the best pairs (classifier, input vector) are selected for each given binary classification. From these pairs, different hybrid classifiers are proposed. Each classifier reaches its outcome through an independent pathway, and these classifiers together form the proposed multi-path hybrid classifier. The final decision of this classifier is obtained from the decisions made at the output of each path. This application brings a global accuracy rate of up to 95% for the transformer oil diagnosis, demonstrating the proposed technique’s effectiveness in the classification field. The proposed model and other conventional algorithms are compared using a small independent database of twenty elements.

Keywords

Authors

1. 1National Higher School of Technology and Engineering Annaba