Cluster analysis on signals from XLPE cable partial discharge detection

XLPE cables have had a large number of applications in the power grid, with good insulation properties, reliable power supply and easy to manufacture and install, however partial discharges caused by insulation defects would have serious consequences, such as insulation breakdown. Therefore, the cable partial discharge detection is of great significance. Since the partial discharge signal is weak and complex, it can easily be interfered by background noise or electromagnetic interference from outside, mainly from the radio signal propagation, a variety of spark discharge near the test site, discharge welding and high-voltage equipment, over-voltage pulse, corona discharge caused by poor contact, inductive discharge caused by a grounded metal object. These signals are random, whose unpredictable nature often have a great interference on partial discharge detection. In this paper, high-frequency current method is used for partial discharge detection. High-frequency current sensors are coupled to a large number of interfering signals. Cluster analysis starting from the distance matrix, selects the magnitude, rise time, fall time and frequency as variable, gets the data matrix and differentiation matrix of four-dimension and calculates using the consolidation method. Clustering results show that the cluster analysis method is effective to separate the signal from each other. In this paper, signals coupled by high-frequency current sensor can be attributed to three categories, namely, partial discharge signals, noise interference signals and corona discharge interference signals. Further study finds that the waveform, magnitude and rise time of above three signals each have significant differences. Basing on the results of the clustering method, frequency domain analysis of partial discharge signal, noise signal and corona discharge interference shows different spectral distribution, and this could be used to distinguish these signals more effectively.