Mother Wavelet Performance Evaluation for Noise Removal in Partial Discharge Signals

Abstract

This article aims to study the pattern of partial discharge (PD) signals occurring on the insulators of high-voltage systems. A mother wavelet comparison and wavelet decomposition are presented to detect and locate PD signals by dividing them into three processes: 1) Signal test generation, employing RC and RLC impedance circuits whereby the output voltage pulses in the RC impedance circuit are expressed as damped exponential pulses (DEPs) and those of the RLC as damped oscillatory pulses (DOPs). White Gaussian noise (AWGN) is then added, which mimics the effect of many random processes in measurement systems. The concept involves applying noise to the original signal and removing it with wavelet transform using various wavelet templates such as Daubechies, Coiflet, Symlet, and biorthogonal to separate the signal components. The experiment results are then compared, and a performance evaluation performed using mean square error (MSE) as in the first two signals. 2) DEPs and DOPs are added with a sine wave to simulate a virtual measurement from a measuring instrument according to the superposition principle using a band-pass filter with the frequency range specified by the two elements to determine the frequency of the resulting PD. 3). The results of the PD signal experiments developed in the laboratory are also evaluated for efficiency by subjective measurement by a PD signal specialist. Therefore, partial discharge denoizing is evaluated using the mother wavelet as a preprocessing step for feature extraction and classification.