Classification of Power Quality Disturbances in Emerging Power System Using Discrete Wavelet Transform and K-Nearest Neighbor

Abstract

Power quality becomes a highly important issue in the power system operation due to the increasing use of power electronic devices. Waveform distortions are frequently caused by the power quality disturbances such as sag, swell, interruption, harmonics, flicker, transients, and notch. These disturbances cause malfunctions, instabilities, and failure of end user equipment’s [1].

It is important to select appropriate features for the classification of PQDs. In order to extract features from the captured signal, signal processing techniques are used. These features are further fed to the classifier. The feature extraction is done in two steps. To begin, the time domain voltage signal is processed using signal processing techniques. Second, extraction of appropriate feature is done using the processed signal. A well selected feature vector reduces the classifiers load [2]. Features can be directly extracted from the original time domain signals, and from the transformed frequency domain signals. To convert the time domain signal into a frequency domain signal, signal processing techniques such as Fourier transform (FT), short-time Fourier transform (STFT), wavelet transform (WT), and S-transform (ST) are utilized.

The frequency contents in the signal were extracted using FT. The frequency content of the signal can be used to identify some PQDs. However, FT is ineffective for transient signals. This is due to the fact that FT only tells you if a frequency component exists, but not when it appears [2]. In order to get this information, the time-frequency localization techniques are used. This enables you to obtain time-evolved signal components in different frequency ranges. This problem can be solved by using STFT, but it has a fixed window size. The STFT approach is found to be insufficient for evaluating non-stationary signals. Hence there is a requirement of efficient and powerful techniques for analysing non-stationary signals [3-6]. Many researchers have recommended Wavelet transform for the analysis of PQDs to overcome the fixed window width problem of STFT [7]. This approach automatically adapts to give correct time and frequency resolutions. High-frequency signal components have a higher time resolution, whereas the low-frequency signal components have higher frequency resolution with this technique. These characteristics make the WT ideal for analysing power system transients induced by a variety of PQDs.

In Wavelet transform wavelet is used as the basis function, and it scales according to the frequency under consideration. The WT delivers superior results than the FT and STFT because the basis function is a wavelet rather than an exponential function. The WT divides the signal into several frequency levels and displays them as wavelet coefficients. Continuous wavelet transform (CWT) and discrete wavelet transform (DWT) are used based upon type of signal under study. CWT-based decomposition is used for continuous time signals, while DWT-based decomposition is used for discrete time signals [8]. 

This paper proposes a discrete wavelet transform and KNN based approach for the classification of PQDs. The PQDs such as sag, swell and interruption are created by simulating the emerging power system using MATLAB Simulink. These disturbances are further analyse using wavelet transform for feature extraction. The features extracted from DWT are further used for training and testing the KNN classifier.