Wavelet packet transform, a reliable and fast method to obtain the fundamental components required for active filtering in power plants

Abstract

The paper focuses on the joint analysis (using Wavelet Packet Transform (WPT)/Discrete Wavelet Transform (DWT) and respectively Fast Fourier Transform (FFT)) of voltages and currents acquired from a generator operating in the Power Plant Turceni, Romania, as a major stage in designing the components of an active power filter (APF). When fast methods, providing accurate results are used, fundamental active currents can be calculated and subtracted from the load currents. Afterward the results are to be inverted in order to obtain the reference currents for the APF. Firstly brief information about the algorithms implementing DWT/WPT, respectively FFT are given. Afterward a study of errors associated to the evaluation of the three major RMS (for fundamental frequency, for distortions and total) is presented. The errors were yielded when analyzing signals with known characteristics (sinusoidal and polluted by a single harmonic), by using 2 interpolation methods (IM): spline and respectively shape-preserving piecewise cubic interpolation. Different numbers of calculation points and disjoint ranges of harmonic orders were used. Comparisons were made between results yielded by the same method but using different simulation parameters, respectively FFT versus DWT (WPT). Two sets of data from the power plant were analyzed. The 1-st one had a reduced harmonic content, whilst the 2-nd one presented significant multi-harmonic distortions. The errors in evaluating the total RMS were computed considering a reference calculated with a Riemann sum. For this PQ index, DWT(WPT) records smaller errors in all studied cases. The differences between the values provided by FFT and respectively by DWT(WPT) are very small (<0.21% in absolute values), proving the reliability of both methods. Spline generated better results. For similar memory consumptions, FFT performed very fast, while WPT achieved very good runtimes (2.98×10−2 sec/period for the calculation of the fundamental's RMS). Smaller mean errors were generated by WPT when evaluating the fundamental's RMS when the fundamental signal was slightly deviated from sinusoidal.