Partial discharge recognition based on SF6 decomposition products and support vector machine

Partial discharge (PD) is usually aroused before the failure of gas-insulated switchgear (GIS) caused by insulation defects, which results in the decomposition of SF₆ used as insulation gas. Concentrations of decomposition products of SF₆ under different kinds of PDs are disparate. Thus, SF₆ decomposition products can be used for PD recognition. In this study, a gas chamber and four defect models were designed to simulate four kinds of typical PDs in GIS. SF₆ decomposition experiments were conducted under the four kinds of PDs. Four kinds of decomposition products, that is, SO₂F₂, SOF₂, CO₂ and CF₄, were selected as feature components. Their concentrations were detected under each experiment. Three concentration ratios, that is, c(SO₂F₂)/c(SOF₂), c(CF₄)/c(CO₂) and c(CO₂+CF₄)/c(SOF₂+SO₂F₂), were proposed as feature parameters for PD recognition. Their physical significances were also analysed. Then, a support vector machine (SVM) was employed as classifier to recognise the four kinds of PDs. The parameters of the SVM were optimised using particle swarm optimisation algorithm. Results show that the recognition method based on SF₆ decomposition products and SVM performs well in PD recognition.