Self-organizing feature map based unsupervised technique for detection of partial discharge sources inside electrical substations. (December 2019)

• Record Type: Journal Article
• Authors: Mishra, Dipak Kumar; Dhara, Sourav; Koley, Chiranjib; Roy, Nirmal Kumar; Chakravorti, Sivaji
• Is Part Of: Measurement. Volume 147(2019)
• Publication Date: 2019-12
• Subjects: Partial discharge -- Continuous Wavelet Transform (CWT) -- Self Organizing Feature Map (SOFM) -- Unsupervised learning -- Time difference of arrivals (TDOA) -- Ultra-High-Frequency (UHF); Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2019.07.046
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: An online and non-contact method for condition monitoring of power system utility. Detection of multiple PD sources in electrical substation using UHF based sensor. Time-frequency domain based features are useful in the detection of PD sources. SOFM based unsupervised learning algorithm for identification of PD sources. Abstract: Detection of partial discharges (PD) that occur due to weakness or defect in electrical insulations has been proven to be most effective tool for condition monitoring of power system equipment. In any power system utility such as electrical substation, there are many electrical components distributed over some specified region. Therefore monitoring of entire substation by mounting few Ultra High Frequency (UHF) based PD sensors around the periphery of the substation could be economical and convenient in comparison with existing component specific PD sensors. In this paper, a novel process for detection and localization PD sources inside electrical substation is proposed, with the help of four UHF sensors. Since the emitted PD pulses in the UHF frequency band are highly non-stationary signature, therefore Continuous Wavelet Transform (CWT) has been applied for extraction of time-frequency domain based specific signature that can help to identify PD sources. Field based experimental investigation shows that presence of multiple PD sources and external pulsating noises are very common. Therefore Self-Organizing Feature Map (SOFM) based unsupervised classifier has been applied for automated detection of actual number of PD sources. Laboratory and Field based experimental investigation showed promising results in applicability of the proposed scheme.