Obtaining patterns for classification of power quality disturbances using biorthogonal wavelets, RMS value and support vector machines

This paper proposes three strategies in order to obtain patterns that allow identification of power quality disturbances. These strategies use Discrete Wavelet Transform (using biorthogonal Wavelet) and RMS value. Disturbances under survey are: low frequency disturbances (such as flicker and harmonics) and high frequency disturbances (such as transient and sags). Due to time-frequency localization properties, Discrete Wavelet Transform permits decomposition of signals in different energy levels, which are used to characterize disturbances that contain information in frequency domain. Four wavelet families were studied and Biorthogonal showed excellent performance. Also, RMS value is used to characterize those disturbances that show big changes in magnitude. The combination of both strategies produces excellent results. Patterns are automatically classified by support vector machines (SVM). Thus, Radial Base Function (RBF) was used as kernel, because RBF requires only two parameters (sigma and C). Cross validation technique and grid search were used in this work. SVM exhibit a good performance as classifier despite similitude between some disturbance patterns.