Comparative Performance of Blind Source Separation Techniques for Partial Discharge Detection in Electrical Substations

The weakness or defect in electrical insulations used in High Voltage operations always leads to generation of partial discharges (PD). Partial discharges (PD) are a common by product of insulation defects in high-voltage systems. Detecting PD is a crucial aspect of power system condition monitoring. The nature of PD signals varies depending on their source, and large substations often have multiple PD sources. Ultra-high frequency (UHF) sensors offer a cost-effective and safe method for PD detection. Multiple sensors can be mounted around a substation, capturing a mixed PD signal. The Separating individual PD signals from this mix is challenging. The Techniques like the Gaussian mixture model (GMM) and Self-Organizing Feature Map (SOFM) have shown promise in this task. GMM uses time and frequency domain features, while SOFM employs continuous wavelet transform (CWT) time-frequency features. This study compares the effectiveness of these techniques for PD detection and localization using both laboratory and field experiments.