A novel complex self organising neural network for PD pattern recognition

As per the standard IEC 60270, Partial Discharge is defined as localized discharge process in which the distance between two electrodes is only partially bridged (i.e., the insulation between the electrodes is partially bridged). This phenomenon caused by the defects in the dielectric and is therefore gains the importance for the diagnosis of insulation systems. The typical partial discharges are corona or gas discharge, surface discharge, and cavity discharge, treeing channels. Every discharge event deteriorates the material by the energy impact of high-energy electron or accelerated ions, causing chemical transformations of many types. Actual deterioration depends upon the dielectric/insulation material used. Partial Discharge (PD) is insulation systems comprises a large variety of physical phenomena, charge carrier emission from PD pulses are inherently stochastic process in which there can be significant statistical variability. The key variable parameters which are characteristics of the PD pulses and which define the basis of the physical phenomena of PD are the time of occurrence (F), the pulse amplitude (q) and the number of pulses (n). The process being stochastic, the associated effects of memory propagation with of residues from previous PD pulses etc. have been made the classification of such PD patterns in terms of f-q-n even more complex. This article concentrates only on the categorization of Void, Oil Corona and air Corona using an unsupervised neural network. Selforganising Map (SOM) is a neural network developed by Kohonen of Helsinki University of technology and often used to classify the input into different categories. SOM is a clustering algorithm, creates the map of relation among input patterns. The map reduces the representation of original data.