Efficient Unsupervised Partial Discharge Detection using Empirical Mode Decomposition and Clustering

Unattended long term partial discharge activities adversely affect the power system performance. Its early detection and prevention is of utmost importance to reduce costly repairs and catastrophic power outages. Biggest challenge in Partial Discharge (PD) detection is that the power line signals are non stationary, exhibiting transient and abrupt changes and contaminated with noise signals. In this paper we propose an efficient unsupervised method for detection of partial discharges of nonlinear, non-stationary three phase time series signal of medium voltage overhead transmission lines using Empirical Mode Decomposition (EMD) and clustering. This method provides higher accuracy and efficiency in comparison with the existing methods. Classification indices like Precision, Recall and F1 score were calculated for the proposed method and the values far outweigh the values obtained from existing methods.