The ultra-high frequency partial discharge detection method is one of the most widely used methods for PD monitoring. The PD signal type can be identified by extracting the time-frequency characteristics of UHF PD signals. We proposed a time-frequency feature extraction method for UHF PD signal based on the improved S-transform. Firstly

the S-transform was improved by introducing a new calculation formula for the frequency-domain window width. Then

the six-dimensional statistical features were computed from the frequency-maximum value graphs

which were calculated based on the S-transform modular matrix. Finally

the SVM algorithm was employed to perform the PD signals recognition. The experimental results show that the proposed method can reflect the distribution characteristics of PD components more accurately in the time-frequency domain due to the consideration of both time-domain and frequency-domain resolution. The proposed method can achieve a high classification accuracy of 97.33% even without denoising processing

which proves its strong noise robusticity. Compared with some commonly used feature extraction methods of PD signal

the proposed method can obtain a higher PD type recognition rate. The results provide references for the recognition of UHF PD signals.