Non-intrusive harmonic source identification using neural networks

This paper proposes the neural network (NN) based approach for the identification of various harmonic sources present in an electrical installation. In this method the harmonic injecting devices are identified using their distinct `harmonic signatures' extracted from the input current waveform. The complexity increases with increase in the number of loads and their combinations. Such automated non-intrusive device identification helps in monitoring and enhancing power quality. The performance of a neural network to a large extent depends upon the type of architecture used and their learning algorithm. Eight commonly used domestic loads are identified and their harmonic signatures obtained. The data is used to design a Feed Forward neural networks (FF) and Single Neuron Cascade networks (SNC). The performance of these models was compared in terms of their recognition accuracy and network complexity. Both the networks are shown to perform well in terms of accuracy. However CC network has been found to be the most suitable architecture because of its low computational requirements and ease in design.