Feature selection for power quality event classification

Power quality events may result in interruption or malfunctioning of electrical equipments that are fed through the voltage line. Detection of such cases by monitoring the voltage waveform remains an important engineering problem. Pattern recognition and classification methods are used for both detection and classification of such events. Although several feature construction and detection/classification methods are reported in the literature, there is no reported research on the comparison of the usefulness of constructed features. This work compares commonly used spectra, wavelet-based and statistical features for their suitability for event classification. As a result of Bhattacharyya analysis and genetic algorithms, the more useful set among a wider set of features is obtained. It is observed that such analysis is not only useful for reducing the feature dimension, but it also improves classification accuracy.