Voltage Waveform Pattern Selection for Power Quality Event Classification

Selection of useful and appropriate identifiers plays an important role in most detection and classification problems including the analysis of voltage waveform for power quality (PQ). In this case, the identifiers are extracted from the acquired voltage waveform. Using such identifiers, several classification algorithms may be applied in order to classify the event type. Statistical, spectral, or directly time domain data can be used as signature identifiers of the voltage waveform. In this work, a method is proposed to select a better set of feature vector elements that are more suitable for classification of PQ events, among a larger set of feature vector elements obtained from numerous methods. For this purpose, a novel covariance based common vector approach (CVA) is proposed. This approach enables a successful PQ event classification, and, at the same time, provides critical information about which of the identifiers within the parameter vector space are more efficient in the classification process. By retaining the more efficient parameters and discarding the rather useless ones, the obtained feature set is small in dimension and efficient in classification.