A NOVEL FEATURE SELECTION METHOD FOR THE DETECTION AND CLASSIFICATION OF POWER QUALITY DISTURBANCES

摘要

We propose an automated recognition method using entropy and correlation characteristics to recognize single and combined Power Quality Disturbances (PQD). We determine the linear and non-linear dependencies among power quality disturbance signals to extract the most relevant features and information. Since the provision of a perfect knowledge about different PQD signals is required for recognition, we try to extract the perfect and near perfect features for every disturbance. We then: i) evaluate the joint occurrence probabilities and propose a new method of 'joint mutual information maximization', ii) evaluate the cross-correlation properties and propose a second stage feature selection method. Multi-class Support Vector Machine algorithm is formulated for classification. Extensive simulations are carried out by incorporating different PQD signals like Sag, Swell, Flicker, Transient, etc. to check the performance of the proposed method. The performance is evaluated through a performance metrics for Accuracy, Detection rate and False Alarm Rate under different environments. The results show that the proposed method can effectively recognize single and combined PQ disturbances.