Method of GIL partial discharge localization based on natural neighbour interpolation and ECOC‐MLP‐SVM using optical simulation technology

• Published in: High voltage Vol. 6; no. 3; pp. 514 - 524
• Main Authors: Zang, Yiming, Qian, Yong, Wang, Hui, Xu, Antian, Sheng, Gehao, Jiang, Xiuchen
• Format: Journal Article
• Language: English
• Published: Beijing John Wiley & Sons, Inc 01.06.2021; Wiley
• Subjects: Accuracy; Algorithms; Aluminum; Discharge; Efficiency; Fingerprints; gas insulated switchgear; gas insulated transmission lines; Insulation; insulation testing; Interpolation; Localization; Localization method; Machine learning; partial discharge measurement; partial discharges; Radiation; Sensors; Simulation; Simulation models; Software; Support vector machines; Transmission lines
• ISSN: 2397-7264; 2096-9813; 2397-7264
• DOI: 10.1049/hve2.12071

Partial discharge (PD) is one of the main causes rendering the deterioration of the insulation state in gas‐insulated transmission line (GIL). Accurate and timely localization of the PD source is essential to ensure the safe and stable operation of the GIL. At present, optical PD detection technology shows advantages in terms of high sensitivity and strong anti‐interference performance. Therefore, an optical PD localization method based on optical simulation fingerprint database is proposed. The introduction of simulation conquers the difficulty of obtaining a PD fingerprint database in field experiments. This method constructs a simulation fingerprint database by performing PD simulation in a GIL simulation model of the same size as the actual GIL. The natural neighbour interpolation algorithm is applied to expand the simulation fingerprint database to cover all locations in the GIL. The two‐level localization method proposed is to match the PD fingerprint to be tested with the expanded fingerprint database, which can reduce the amount of calculation while maintaining the localization accuracy. The experimental results show that the average localization error of this method is only 9.7 mm, and the localization time is reduced by about 11 times compared with the normal one‐level localization method.