Abstract: Gas insulated switchgear（GIS）is a key equipment in power system. Obtaining the discharge type of insulation defects in GIS equipment through partial discharge signal is very important for fault diagnosis and early warning. A method for identifying different types of discharge defects in GIS equipment through experiments was proposed. Firstly,typical defect models in GIS were built and UHF sensors were applied to obtain partial discharge signals. Then,Gabor transform was performed on the threedimensional partial discharge PRPS patterns of GIS and texture and shape feature characteristics of the decomposition patterns were extracted. The hierarchical clustering method was used to cluster the extracted feature quantity. The clustering results show that the feature quantity has a good correlation with the discharge characteristics of defects,and verify the feasibility of identifying typical defects according to the feature quantity extracted from PRPS patterns by Gabor transform. Different kinds of machine learning algorithms were used to identify the discharge types for different typical discharge defects.The results show that the recognition accuracy of the 3 typical discharge types by various machine learning algorithms is high,and the extracted feature characteristics of PRPS patterns after the Gabor transform well reflect the discharge characteristics with high differentiation. The proposed diagnosis method can provide a reliable reference basis for GIS fault warning.