Entropy application in partial discharge analysis with non-intrusive measurement

Partial discharge (PD) occurs when insulation deterioration happens in electrical apparatus. It is often detected in order to evaluate the state of insulation. For metal-clad equipments, external sensors which are easy to install and interruption-free on operations are preferred. However, their performances are compromised by heavy noise. Although time-frequency (TF) spectrum provides much information to discriminate PDs and noises, automatic selection remains a tough issue in field application. Entropy, a measure of disorder, is applied in this paper to extract PD pulses automatically. This entropy-based algorithm is implemented and examined by two field-collected datasets. Practical results show that true PDs can be identified and extracted effectively