As an early signal of insulation degradation on GIS

partial discharge（PD）can lead to further insulationdegradation

so studying PD severity assessment has important theoretical significance for the maintenance of GISequipments. Combined with the monotonic correlation between the statistical characteristics of PD signal and its se-verity

this paper extracted nine statistical characteristic parameters that could characterize the severity of PD

divid-ed the severity of PD into three stages: light discharge

medium discharge and heavy discharge

and then establisheda severity assessment algorithm for PD in the total process based on K-means cluster analysis. The PD test of metalprotrusion defects was as an example

the severities of the obtained PD signals in the total process were evaluated

and the standard numerical intervals of each characteristic parameter under different PD severity were obtained. Theresults showed that the PD severity assessment results were well consistent with the trends of each characteristic pa-rameter

and the feasibility and effectiveness of PD severity assessment algorithm proposed by this paper was verified.