Methodology based on higher-order statistics and genetic algorithms for the classification of power quality disturbances

The present work proposes a methodology for the detection and classification of some transient power quality disturbances. It uses a genetic algorithm for estimating the amplitude, frequency, and phase of the fundamental component in an optimum way to suppress it from an electric signal. By using this pre-processing strategy, it is possible to remove a single frequency component instead of removing a whole frequency band as other methodologies do. Once the fundamental component is suppressed, it is possible to perform a better identification of an anomalous condition in a signal because its high energy does not hide the presence of a disturbance. After the suppression, the proposed methodology computes three higher-order statistics (variance, kurtosis, and sixth-order cumulant) from the resulting signal and uses them as inputs for a fuzzy-based classifier. Higher-order statistics are selected because they are insensible to the presence of Gaussian noise adding robustness to the proposed methodology. Experimentation is performed using both synthetic and real signals. Real signals come from a photovoltaic generation plant and a hospital facility, both located in Spain. Results prove that the proposed methodology allows enhancing the results delivered by other methodologies up to 30%.