captured from a population of in-service distribution
circuit breakers and empirical UHF data captured from
laboratory experiments simulating partial discharge defects
typically found in HV transformers. This discovered
knowledge then forms the basis of two separate decision
support systems for the condition assessment/defect
classification of these respective plant items. In this paper
is shown a comparative of competitive learning algorithms
to classify measured PD activities into underlaying
insulation defects or source that generate PD’s using Self
Organizing Maps (SOM). Multidimensional scaling
(MDS) is a nonlinear feature extraction technique [12]. It
aims to represent a multidimensional dataset in two or
three dimensions such that the distance matrix in the
original k-dimensional feature space is preserved as
faithfully as possible in the projected space. The SOM, or
Kohonen Map [13], can also be used for nonlinear feature
extraction. It should be emphasized that the goal here is
not to find an optimal clustering for the data but to get
good insight into the cluster structure of the data for data
mining purposes. Therefore, the clustering method must be
fast, robust, and visually efficien
The charge that a PD generates in a cavity (see Fig. 2) is
called the physical charge and the portion of the cavity
surface that the PD affects is called the discharge area.
Eapplied is the applied electric field and qphysical is the
physical charge [14