Classification of power quality disturbances in distribution systems using two Mathematical Morpological Filters and multiclass support vector machine

This paper presents a pattern recognition-based scheme for detection and classification of Power Quality Disturbances (PQDs) in distribution systems. The main idea behind the proposed scheme is the use of spatial extracted features from the proposed Mathematical Morpological Filters (MMFs) that can be gained from PQD signals. In this study first, the PQD signals were analyzed using MMF1 for noise reduction. Second, the exit of MMF1 entered into MMF2 to depress the steady state components of these signals and to increase the transient components of signals and again to reduce the noise. Third, the spatial features such as summation, standard deviation, maximum, mean and rms of the MMF2 magnitudes were calculated. Finally, to distinguish the PQD signals from each other, a multi-class support vector machine (MSVM) classifier based on one against one approach was trained based on the selected features. To investigate the power of the proposed scheme for PQD signals classification, the results of studies on the various PQD signals including 6 single types namely pure sine waveform, sag, swell, interruption, harmonics and oscillatory transients were presented. These results show that the proposed method reliably detect and classify the clean and noisy PQD signals. The rate of detection and classification of these signals is 100 percent for clean test data and 95.71 percent for noisy test data of 30 db. The time of proposed method within … ms is considerable.