This paper presents a wavelet analysis technique together with support vector machines (SVM) to discriminate partial discharges (PD) from external disturbances (electromagnetic noise) in a GIS PD measuring system based on magnetic antennas. The technique uses the Cross Wavelet Transform (XWT) to process the PD signals and the external disturbances coming from the magnetic antennas installed in the GIS compartments. The measurements were performed in a high voltage (HV) GIS containing a source of PD and common-mode external disturbances, where the external disturbances were created by an electric dipole radiator placed in the middle of the GIS. The PD were created by connecting a needle to the main conductor in one of the GIS compartments. The cross wavelet transform and its local relative phase were used for feature extraction from the PD and the external noise. The features extracted formed linearly separable clusters of PD and external disturbances. These clusters were automatically classified by a support vector machine (SVM) algorithm. The SVM presented an error rate of 0.33%, correctly classifying 99.66% of the signals. The technique is intended to reduce the PD false positive indications of the common-mode signals created by an electric dipole. The measuring system fundamentals, the XWT foundations, the features extraction, the data analysis, the classification algorithm, and the experimental results are presented.
Robust Condition Assessment of Electrical Equipment with One Class Support Vector Machines Based on the Measurement of Partial Discharges
