Application of Joint Time-Frequency Analysis on PD Signal Based on Improved EEMD and Cohen's Class

The feature extraction and pattern recognition of partial discharge signal are key steps of equipment condition assessment and fault diagnosis. Time-frequency analysis on PD pulse can make up for deficiencies of traditional phase statistical method by extracting more comprehensive and effective information from waveform. Cohen's class distribution is a commonly used time-frequency analysis method except for influence of cross interference terms. This paper presents a method of joint time-frequency analysis on PD pulse signal based on EEMD and Cohen's class. The end effect of EEMD is studied and an extending technology based on SVR-regression fitting method is proposed as well. The exponential attenuation oscillating function added with Gaussian white noise and narrow band interference is used to simulate the high frequency current PD signal of power equipment. The results show that this method can accurately identify the characteristic PD pulse. This method can not only guarantee the time-frequency concentration of effective signal, but also inhibit the influence of IMFs' cross interference terms. Finally, we prove the effectiveness and practicality of this method by applying it on PD signal field measured in substation.