Abstract
This article proposes WT (Wavelet Transform) and an ANN (Artificial Neural Network) based approach for detection and classification of EPQDs (Electrical Power Quality Disturbances). A modified WT known as ST (Stockwell Transform) is suggested for feature extraction and PNN (Probabilistic Neural Network) for pattern classification.
The ST possesses outstanding time-frequency resolution characteristics and its phase correction techniques determine the phase of the WT to the zero time point. The feature vectors for the input of PNN are extracted using ST technique and these obtained features are discrete, logical, and unaffected to noisy data of distorted signals.
The data of the models required to develop the distorted EPQ (Electrical Power Quality) signals, is obtained within the ranges specified by IEEE 1159-1995 in its literatures.
The features vectors including noisy time varying data during steady state or transient condition and extracted using the ST, are trained through PNN for pattern classification. Their simulation results demonstrate that the proposed methodology is successful and can classify EPQDs even under a noisy environment very efficiently with an average classification accuracy of 96%.