Hierarchical cluster analysis of broadband measured partial discharges as part of a modular structured monitoring system for transformers

In this article investigations on hierarchical cluster analysis methods concerning their ability to evaluate broadband measured partial discharges (PD) are presented. Therefore different cluster algorithms were tested with various data measured in the laboratory on a special modified distribution transformer and with measurements from a power transformer on-site. A comparison of the results obtained by different cluster analyses shows that these methods are suitable for a location of PD sources inside transformers as well as for a separation of these partial discharges from pulse shaped noises, e.g. corona impulses. In order to achieve a higher efficiency and performance in evaluating PD signals cluster algorithms are combined with additional pattern recognition methods in a modular structured monitoring system for transformers.