Performance of Clustering Techniques of Multiple Partial Discharge Sources in High Voltage Transformer Windings

Abstract

There are numerous of clustering techniques that have been exploited by researchers in many applications such in medical application, image processing application as well as in high voltage application. Clustering technique is an unsupervised learning algorithm used to identify group structure in a set of data that contain different characteristics. Nowadays, within the latest HV insulation system, there are more than one dielectric media, which contribute to multiple source of partial discharge (PD). Therefore, data identification for PD is significantly vital to discover the kinds of faults that inducing discharges in a HV insulation system. Nevertheless, it is critical that the methodology used for further investigation such as phase-resolved partial discharge (PRPD) analysis is capable of producing a sufficient separation between the clustered data. An experiment was performed to generate a pair of PD sources simultaneously within a winding of the HV transformer. The PD pulses were collected from two measuring points measured by two wideband radio frequency current transformers (RFCTs) at the bushing tap-point to earth (BT) and the neutral to earth-point (NE).The performance oft-Distributed Stochastic Neighbour Embedding (t-SNE), Principle Component Analysis (PCA) and time-frequency mapping based on sparsity roughness at distinguishing multiple PD sources is determined and presented.