Partial discharge pattern analysis using support vector machine to estimate size and position of metallic particle adhering to spacer in GIS

•Particle initiated PD patterns for different particle sizes and positions along the length of the spacer in GIS were acquired.•PD patterns can be classified using linear SVM techniques utilizing PD fingerprints obtained experimentally.•Phase resolved partial discharge patterns depend on particle size and its location in GIS.•Statistical operators of different distributions of phase resolved partial discharge patterns provide useful information.•Support vector machine can be successfully used to discriminate between different particles defects.

Gas insulated substations (GISs) are widely used in electrical power transmission and distribution systems. The presence of free and fixed metallic particles can initiate partial discharges (PDs) in GIS which can become a serious defect and reduce the reliability of GIS. The particle initiated PD characteristics depend on the particle size and position. Therefore, the PD characteristics can be used for the estimation of particle size and position on the spacer surface. Knowledge about the particle size and position are the important steps for the reliability improvement of the GIS equipments. This paper investigates the PD characteristics for fixed particle adhering to cylindrical shaped spacer in a simulated GIS. Length of cylindrical particles, their position on the spacer surface and gas pressure is varied to study the PD characteristics that are represented by PD fingerprints. Then these data are used for particle size and position identification. For this purpose, the use of linear support vector machine has been proposed in this paper to classify particle position and size based on the PD fingerprints data acquired at different SF6 gas pressures. It is shown that the proposed method was able to successfully estimate particles size and position at different gas pressures with an accuracy of 94%. Thus in order to improve the reliability of GIS, this approach could be considered as a potential method for particle position and size estimation in GIS.