Abstract
Currently, one of the most common methods of assessing the state of high voltage electrical equipment is to measure the activity of the partial discharges (PD) that may occur in it. Usually, most commercial measurement systems show the PD activity as a representation of pulses superimposed on a diagram of the network signal. These plots are called Phase- Resolved Partial Discharge patterns (PRPD), and are used to classify PD sources (corona, internal and surface). However, in common scenarios found in industrial environments, the identification of the type of source is practically impossible with the PRPD patterns, due to the presence of multiple PD sources and electrical noise which can create complex PRPD patterns even for an expert in the field. This challenge can be easily addressed, with the prior application of source separation techniques. In this paper, we propose the application of two meta-heuristic approaches, in order to automatize and improve the performance of the separation technique called spectral power clustering technique (SPCT), which is currently applied in the separation of PD sources and noise.