An image-oriented recognition method for PD development stage in oil-pressboard insulation with needle-plate model under AC-DC mixed voltage

Recognition of the partial discharge (PD) development stages in oil-pressboard insulation under AC-DC mixed voltage are the bases for analysis of the insulation faults in converter transformers. In this paper, behavior of different characteristic parameters of the PD activity recorded in the oil-pressboard insulation with needle-plate electrode under 1:1 mixed AC-DC voltage is investigated. Based on the hierarchical cluster analysis with six feature parameters including maximum amplitude of apparent charge in pC (G
_max
), average amplitude of apparent charge in pC (Q
_ave
), PD repetition rate (N), maximum PD interval (rf
_max
), pulse equivalent time (r
_eq
) and pulse equivalent frequency (F
_eq
), the PD development process can be divided into five stages: initial stage (IS), transition stage (TS), developing stage (DS), steady stage (SS), pre-breakdown stage (PS). The possible physical effects of the identified PD development stages are discussed and the radar maps with the six feature parameters are analyzed. The radar maps have different shapes in different PD development stages. The barycenter-coordinate ratio and depth-width ratio of the radar map are extracted as the new features for recognition of PD development stage. Based on least square support vector machine with particle swarm optimization (PSO), the recognizer for PD development stages is designed.