Identifying Incompatibilities of the Pentagon Methods for Dissolve Gas Analysis Using Validation Indices

Authors:

Abstract
Dissolved gas analysis (DGA) plays a crucial role in identifying faults in power transformers. To visualize the parts per million (ppm) values of five gases in DGA, various pentagon methods, namely EW pentagon, Duval pentagon, and Mansour Pentagon, are employed. However, these methods exhibit certain limitations, which are mathematically elucidated using multiple indices evaluated in this research. The limitations of pentagon methods are not discussed in the established literatures. The study focuses on several key indices, including the Silhouette Coefficient, Davies-Bouldin Index, Calinski-Harabasz Index, Hubert Index, and reconstruction error, to quantify and assess the performance of the pentagon criteria. By analyzing the changes in these indices after applying the pentagon methods, the effectiveness and suitability of each approach are evaluated. Understanding these changes in the indices provides valuable insights into the performance of the pentagon methods. The Silhouette Coefficient measures the quality of clustering, indicating how well data points are grouped together. The Davies-Bouldin Index assesses cluster separation, determining how distinct clusters are from one another. The Calinski-Harabasz Index measures cluster compactness, evaluating how tightly the data points are grouped within clusters. The Hubert Index combines measures of cluster separation and compactness. Lastly, the reconstruction error quantifies how well the original data can be reconstructed from the identified clusters.
Funding Information
  • IPPP (IIRG001B-2020IISS)
  • MOSTI (TDF06221586)

This publication has 19 references indexed in Scilit: