Partial discharge recognition through an analysis of SF6 decomposition products part 2: feature extraction and decision tree-based pattern recognition

The decomposition characteristics of the SF₆ under the different kinds of partial discharges (PD) should be understood first when recognizing PD by analyzing SF₆ decomposition products in gas insulated switchgear (GIS). Moreover, the characteristic quantities used for recognition must be found. In this paper, the concentration and concentration ratio of SF6 decomposition products were each selected as characteristic quantities. Fuzzy c-means clustering algorithm was adopted to assess the performance of the two types of characteristic quantities, which was based on the data of SF₆ decomposition products under the four kinds of PD in Part 1. Concentration ratio had better performance than concentration as a characteristic quantity in PD recognition. The concentration ratio method for PD recognition was established based on the decision tree theory, in which the three concentration ratios, namely c(SOF₂)/c(SO₂F₂), c(CF₄)/c(CO₂), and c(SOF₂ +SO₂F₂)/c(CO₂+CF₄), were used as characteristic quantities. The physical significance of the three concentration ratios was also analyzed. Finally, the concentration ratio method was applied to test the performance of PD recognition. The method has a good performance and can successfully recognize different kinds of PD.