In this paper, a new pulse separation technique based on fuzzy logic algorithm is proposed. An equivalent Time-Frequency diagram (TF map) is obtained from the feature extraction and clustering of both PD pulse signal and ambient noise signal. Since different kinds of PD types derived from different PD sources, they will form specific clusters in the TF map. It provides a possibility to classify the millivolt voltage PD pulse signals from the ambient noise signals, enhances the anti-interference ability of the PD detection system, classifies partial discharge patterns and locates PD sources. This paper provides a new idea for the accurate extraction and processing of the PD pulse signal of the 12kV switchgear, which is also helpful to improve the detection sensitivity and locate PD sources.