Recognition of the Partial Discharge Pattern of the Electrode Voids by Neural Network Theory

Abstract

It has become more important than ever to observe the partial discharge phenomena to detect insulation deteriorations, because the applied voltages have become higher. So far, we can mesure the quantity of electric charge (q) and occurrence frequency of partial discharge (n) at the same time. However recently it has become possible to mesure not only these two factors but also the phase (φ) at the same time.
Each electrode model has the specific φ-q distribution pattern. Therefore, in partial discharge diagnosis it is very important to recognize these patterns.
This paper proposes to apply a neural network theory, specifically the backpropagation method, for identifying electrode types (Tree, IEC (b) and Cigre Method I) and estimating the shape of the cylindrical voids of electrodes.
The simulation result have confirmed the validity of applfying the back propagation method to the pattern recognition of the electrodes.