Appropriate Feature Extraction Methods for Detection of Disturbances Associated with Power Quality

Monitoring disturbances in power quality (PQD) has gained importance in recent decades. This interest arises from the fact that disturbances associated with power quality directly affect the equipment connected to the grid, leading to malfunctions or complete equipment failure. For this reason, efforts have been made to optimize the methods for detecting these disturbances occurring in the electrical network.In this study, a Python algorithm is proposed to improve the detection and classification of seven types of simple electrical disturbances (sag, swell, flicker, oscillatory transient, interruption, harmonics, and notch). The algorithm varies the combination of characteristic vectors extracted (energy, mean, standard deviation, skewness, Shannon entropy, RMS, kurtosis, and logEnergy entropy) obtained from the Discrete Wavelet Transform (DWT) through Multiresolution Analysis (MRA) with 6 levels of detail. A database was generated by sampling three thousand five hundred electrical disturbances at 10 kHz. The experiment involved connecting a Beagle Bone Black (BBB) development board to a BK Precision 4064 arbitrary waveform generator.Combinations of two characteristic vectors were extracted from each signal in the database to evaluate them using the Random Forest classifier and determine which one is most suitable for this type of analysis based on their accuracy percentages.