Power Quality Disturbance Detection and Source Prediction Using Advanced Signal Processing Techniques

Abstract

Signal processing techniques have been widely used for analyzing power signals for the purpose of automatic power quality (PQ) disturbance recognition. Among the different signal processing techniques used in extracting features of disturbances from a large number of power signals, the most widely used techniques are the fast Fourier transform (FFT) and the windowed Fourier transform which comprises of the short time Fourier transform (STFT) and the wavelet transform (Moussa et al., 2004). The FFT is ideal for calculating magnitudes of the steady-state sinusoidal signals but it does not have the capability of coping with sharp changes and discontinuities in the signals. Thus, it cannot accurately detect the end of sustained events such as voltage sag, swell, transient and interruption. Although the modified version of the Fourier transform referred to as the STFT can resolve some of the drawbacks of the FFT, it still has some technical problems. In the STFT technique, its resolution is greatly dependent on the width of the window function in which if the window is of finite length, the technique covers only a portion of the signal, thus causing poor frequency resolution. On the other hand, if the length of the window in the STFT is infinite so as to obtain a perfect frequency resolution, then all the time information will be lost. Due to this reason, researchers have switched to wavelet transform from the STFT (Karami et al., 2000). Some of the well-known wavelet transforms are the continuous wavelet mechanism transform (CWT) and a modification of the CWT which is known as the S-transform. Although CWT based multiresolution analysis monitors the regions of interest closely which is short windows at high frequencies and longer windows at low frequencies, its accuracy is susceptible to noise and if a particular frequency of interest has not been extracted due to octave filter bands, there is a chance of misclassification. To overcome this problem, the Stransform based multiresolution analysis using a variable window (Stockwell, 1996) offers significant advantage with a superior time-frequency localization property and yields amplitude and phase spectrum of the PQ event signals in the presence of noise. The S-transform is based on a moving and scalable localizing Gaussian window and has characteristics superior to the CWT. It is fully convertible from the time domain to the twodimensional frequency translation domain and to the familiar Fourier frequency domain. 1