Abstract: Ultra-high frequency (UHF) partial discharge (PD) signals of UHV gas insulated switchgear(GIS) are characterized by small pulses and few pulses, which are often submerged by interference signals. Conventional single-channel PD identification method is difficult to effectively extract UHF GIS occasional PD signal, thus it is urgent to study the multi-channel UHF occasional PD signal recognition algorithm. First, the 1000 kV GIS multi-channel UHF PD signal is measured in the field. Based on its spectrum characteristics, a simulation experiment method of UHV GIS UHF accidental PD signal is proposed by using double-exponential pulse signal injection and real interference source. By changing the amplitude of the injected signal to simulate the UHF PD signal pulse with different signal noise ratio(SNR), 9000 sets of multi-channel UHF PD signal and interference signal data samples are obtained. Secondly, the time-frequency feature clustering method of multi-channel UHF signals is studied. Based on the conventional K-Means method, an identification method of UHV GIS UHF PD based on modified time-frequency clustering boundary is proposed, and the optimal value of the correction coefficient L is proposed. Finally, this method is applied to live detection of a 1000 kV GIS UHF PD site, and the effectiveness of this method is verified.