Extraction of PD fingerprints using correlation learning

Partial discharge (PD) source identification using statistical parameters to describe PD distribution patterns is regarded as an important approach for insulation diagnosis. It has been proved that phase resolved (PRPD) and pulse height resolved (PHPD) PD distribution patterns are related to the characteristics of the PD. Statistical parameters describing distribution shapes of PRPD and PHPD patterns as well as the relationship among them can be employed to identify the type of PD source. An effective way to describe the calculated statistical parameters is to define a new set of axes that can extract the maximum amount of information about an input vector. Hebbian learning utilizes directly the product of its local input and output to drive the weight updates. In this paper, it has been demonstrated that improved Hebbian learning networks are useful for PD data compression as well as feature extraction.