Utilization of artificial neural network for the estimation of size and position of metallic particle adhering to spacer in GIS

The defects caused by the presence of metallic particles are the most frequent threats in Gas-Insulated Substation (GIS). These particles can move freely in the GIS enclosure and thereby when they come in the vicinity of spacers they can adhere to the solid spacer surface due to electrostatic forces resulting in the initiation of partial discharges (PDs) affecting the GIS reliability. The magnitude and frequency of these PDs depend among others on the size and position of particles. Thus the knowledge of the size and position of contaminating particles and its PD characteristics is essential for the improvement of the reliability of such equipment. This paper is aimed at the use of the Back-Propagation Artificial Neural Network (BP-ANN) technique for recognizing the PD patterns to estimate the particle size (length) and position on the spacer surface in a simulated GIS. The PD patterns were characterized by a number of statistical operators describing the shapes of distributions of the PD signals acquired from the measurements carried out using IEC 60270 method. In developing the BP-ANN, some parameters were varied to find the most optimal network. The results show that the best-developed ANN in this study is able to recognize various PD patterns in the employed GIS model and is able to estimate particle length and position on the spacer surface at different SF
₆
pressures. The achieved accuracy in detecting the size and location of particle was about 92%. Thus, the proposed method constitutes a helpful tool in improving the reliability of GIS as well as for diagnosis.