Automatic Power Quality Disturbances Detection and Classification Based on Discrete Wavelet Transform and Support Vector Machines

In this paper some patterns based on discrete Wavelet transform are studied for detection and identification of both, low frequency disturbances, like flicker and harmonics, and high frequency disturbances, such as transient and sags. Daubichies4 Wavelet function is used as a base function to detect and identify due to its frequency response and time localization information properties. Based on these patterns, power quality disturbances are automatically classified by support vector machines (SVM). Thus, Radial Base Function (RBF) was used as a kernel, because RBF requires only two parameters (σ and C) and cross validation technique and grid search were used in this work. SVM exhibit a good performance as classifier (90 percent of success for most disturbances) in spite of similitude between some disturbance patterns. INTRODUCTION Electromagnetic disturbances cause big economic losses for industry and residential users. Because of this, monitoring of power quality (PQ) disturbances of electrical energy is fundamental to offer solutions to industrial and to electrical areas. Wavelet Transform (WT) processing technique has been proposed for power quality monitoring given its time-frequency multiresolution analysis property. WT properties, like limited effective time duration, band pass spectrum, waveform similar to disturbance and orthogonality, allow locating information in time and frequency domains. Thus, it is possible to obtain high correlation when PQ disturbances occur and decompose these events into different components without energy aliasing. There are several studies [1]-[3] where WT is used for detecting and identifying disturbances with Wavelet function Daubichies 4. Likewise, neuronal networks have been used to classify different disturbances from its WT. [4] shows a method of PQ disturbances detection and classification based on heuristic rules. However, there are no references about using other techniques of classification like Bayesian or support vector machines (SVM) for PQ disturbances. In this article, mathematical concepts of Discrete Wavelet Transform (DWT) are described. The properties that make DWT effective are also discussed. Then, strategies for PQ disturbances detection and identification by using DWT are studied. Strategies used for automatic classification of these disturbances are also presented. Finally, results of simulation and conclusions of this investigation are shown. DISCRETE WAVELET TRANSFORM Fourier Transform (FT) only allows the study of a fixed interval of a transient disturbance, but it is not possible to know its location. Then, a dynamic scheme is necessary where, in the same coordinates system, the width of time and frequency windows can be varied simultaneously preserving resolution in both domains (time and frequency). This characteristic is reached by means of the timefrequency multiresolution analysis that WT makes. The Continuous Wavelet Transform (CWT) is defined in (1),[6]: ) ( ), ( ) ( 1 ) ( , ) , ( t t x dt a b t t x a x W b a a b ψ ψ ψ = ⎟ ⎠ ⎞ ⎜ ⎝ ⎛ − = ∫ ∞