An EFPI fber optic ultrasonic sensor can be used for the detection and pattern recognition of partial discharge ultrasonic signals in Gas Insulated Switchgear. Compared with traditional piezoelectric sensors, it has many advantages, such as high sensitivity and strong anti-interference ability. Based on this, four typical PD models of the tip, metal particles, suspension and creeping surface were set up in the GIS cavity flled with 0.4 MPa SF6 gas, and the EFPI sensor was innovatively used to detect the discharge ultrasonic signal and extract the single ultrasonic pulse signal. The waveform features form a feature parameter database, and the probabilistic neural network algorithm and the support vector machine algorithm are used for pattern recognition and comparative analysis, respectively. The ultrasonic signal detected by the EFPI sensor has prominent features. On the basis of extracting the feature parameters, the two pattern recognition algorithms can achieve an average recognition rate of more than 85%, and the recognition efect of the support vector machine is better than that of the probabilistic neural network