Classifying Power Quality Disturbance using Time and Multiresolution Features through Artificial Neural Network

Power quality deterioration is one of the major problems in the area of power system. Usually, deterioration in power quality happens because of various environmental like animal contact, tree contact, vehicle collision, and electrical factors like electrical equipment malfunction or failure. Every source amounts to dropping the level of power quality by distorting the voltage and current waveforms. Some distortions, such as swag, swell, and transient are reflected in the waveform showing peculiar signatures. It is desirable that using suitable computational methods identify the occurrence of such signatures in the voltage/current waveform so that their source can be tracked, rectified, and eventually an uninterrupted and good quality power supply can be given to the consumers. Therefore, this paper proposes machine learning methods for automatically identification of power quality disturbances utilizing voltage waveforms. Artificial neural network (ANN) is used on a set of features extracted from the voltage waveforms of the EPRI power quality dataset. Features based on root mean squared method, Fourier transform, and wavelet packet decomposition are extracted for the identification of the power quality disturbance in voltage waveform which are categorized into five types: normal voltage, voltage sag, swell, sag with transient and oscillatory transient. These features are applied to ANN for classification yields an output accuracy of 93.33%.