Research and Implementation of Localization of Multiple Local Discharge Sources in Switchgear Based on Ultrasound

At present, most of the switchgear partial discharge detection means are offline detection and cannot monitor multiple partial discharge sources online at the same time. Based on this, this paper investigates the application of ultrasonic technology in localized discharge fault localization in high-voltage switchgear, removes the background noise of localized discharge in switchgear by using soft and hard filtering; proposes a generalized cubic correlation algorithm on the basis of TODA, improves the accuracy of the time difference acquisition in the case of low signal-to-noise ratio; determines the number of multiple localized discharging power sources by using the single-channel signal blind source separation technique and singularity spectral analysis; and determines the number of multiple localized discharging power sources by using independent component analysis to separate them. As well as for the problem that TDOA cannot be directly applied to the localization of multiple partial discharge sources, independent component analysis is used to separate the mixed signals, and the disordered coordinate selection method is proposed to determine the coordinates of multiple partial discharge sources. The experimental results show that (1) the noise reduction method is able to remove the excess interference while preserving the localized discharge signals; (2) the improved generalized cubic inter-correlation algorithm is more resistant to interference and has less error than other time delay estimation algorithms. The localization error is reduced by 60 mm~68 mm compared to the basic correlation algorithm, 41 mm~47 mm compared to the twice correlation algorithm, and 17 mm~20 mm compared to the three times correlation algorithm, which is a big improvement compared to the pre-improved algorithm. (3) It is able to locate the multiple localized power sources, and the accuracy of the number of localized power sources reaches 88%.