In this study, instantaneous envelope, phase and frequency series are obtained by Hilbert transform for Power Quality (PQ-Power Quality) disturbances signals. Rms, Thd, energy, entropy and statistical properties are applied to these series. With the wrapper feature selection approach, a set of features is obtained that has a small number of feature subset and a high performance from 36 features. Genetic Algorithm (GA) is used as a search algorithm and the classifier algorithm is K nearest neighborhood (KNN). Support Vector Machines (SVM) for selected features are also used in the classification step. The learning algorithm is obtained as KNN, the model performance that classifies PQ classes with 99.07%. The number of feature sets is 8. In addition, performance under noisy data is also tested to show that the generated model has a generalized structure.