ANNs based on subtractive cluster feature for classifying power quality

This paper concerns classification techniques of various events causing unacceptable power quality (PQ). Because of the overall complexity associated with the power quality analysis and diagnosis is a complex one for many reasons, a computerized system analysis is becoming vital for the realization of effective and efficient power quality diagnosis systems. In this paper a novel feature extraction method based on center clustering is obtained which is used as input to artificial neural network (ANN) for classifying power quality disturbances is presented as one of the tools in support of computerized power quality diagnosis. To demonstrate the potential of this approach and provide comparison with other techniques, extensive simulations of different types of poor power quality phenomena to be made. This is followed by applying a wavelets-based analysis to identify the corresponding PQ problems. Several feature extraction methods are proposed to reduce the amount of processed data. The proposed center clustering method leads to dramatic performance improvements when compared to other techniques assessed. The extracted features are used to train three different artificial neural networks (ANNs) and their performance is also assessed. It is concluded that a center clustering-based ANN offers the best performance.