Source database for Machine-learning-based PD Pattern Recognition in GIS

Published: 30-04-2020| Version 1 | DOI: 10.17632/cz8gwg9d2v.1
治诚 吴


The PD in GIS may be induced by a conductor protrusion, insulator surface contamination, floating electrode defects, voids in the bulk insulation, or free conductive particles. The insulator must pass an X-ray inspection as part of the factory test necessary for epoxy resin vacuum casting process optimization, so void defects inside the insulator to not restrict GIS reliability. Defects caused by on-site contamination or switching actions, however, are likely to cause insulation failure. The main PD type includes corona type (C-type), surface type (S-type), and floating electrode type (F-type). Source data was generated in the simulated defect set in a 550 kV GIS. A needle with varying length and curvature radius was fixed on the busbar to create a conductor protrusion defect [1]. A linear particle was adhered to the convex surface of a single-phase basin-type insulator at a varying distance to the high voltage electrode to create a surface contamination defect [2]. In addition, an aluminum curved plate was placed at a distance of 1 mm from the high voltage electrode to create a floating electrode defect. These three simulated defects respectively induced C-type PD, S-type PD, and F-type PD defects. The PRPD patterns with the applied voltage from PDIV to the PD test voltage (385 kV, according to IEC 62271) were recorded by Omicron MPD600. The SF6 gas pressure was set to the rated working condition of 0.5 MPa. The patterns were organized into the tuple (Φ, Q, N). Experiments on real GIS equipment were used to resolve the equivalence problem of defect models. A total of 418 PRPD patterns were obtained, including 280 cases of C-type PD, 129 cases of S-type PD, and 9 cases of F-type PD. The diversity of training data directly affects the reliability of the PD pattern recognition, thus the authors call on all those who are committed to related research to open source the experimental data. [1] Z. Wu et al, “Effectiveness of on-site dielectric test of GIS equipment,” IEEE Trans. Dielectr. Electr. Insul., vol. 25, no. 4, pp. 1454–1460, Aug. 2019. [2] Z. Wu et al, “Phase-space joint resolved PD characteristics of defects on insulator surface in GIS,” IEEE Trans. Dielectr. Electr. Insul., vol. 27, no. 1, pp. 156–163, Feb. 2020.