Denoising highly distorted small currents in an environment with variable levels of noise

The paper is focused on finding appropriate and systematic procedures of denoising highly distorted small currents with significant “signal to noise” ratios and on evaluating the denoising effects over the power quality (PQ) indices. Real currents were acquired from 2 consumers placed in a small industrial hall with highly variable power consumptions. Denoising procedures relying on wavelet based decomposition trees were considered. First evaluations were made for determining parameters like: number of levels in the tree, method of thrashing and type of level noise estimation. Best results were obtained with a tree having 4 levels, considering a white Gaussian noise and performing “per each tree level” estimation of the noise power. With this tree, acquired signals were denoised considering 4 of the most commonly used wavelet mothers, consequently obtaining sets of 4 different noise signals per acquired signal. An average noise signal was used to get an estimated noise powers. For each acquired signal, sets of 10 Gaussian noises with powers identical to the above mentioned estimations were randomly generated and used to determine in an original (statistical) way the best wavelet mother. FFT based programs were used to evaluate PQ indices of the acquired signals before and after denoising. In the denoised signals all total RMS values were found to be smaller, whilst harmonic spectra for orders with weights > 1% were less affected in the low harmonic range (orders<70), certifying the nature of noise and the denoising procedure. The total harmonic distortions were slightly diminished by denoising.