Skip to Main Content
This paper proposes a neural-network (NN)-based approach to nonintrusive harmonic source identification. In this approach, NNs are trained to extract important features from the input current waveform to uniquely identify various types of devices using their distinct harmonic "signatures". Such automated, noninvasive device identification will be critical in future power-quality monitoring and enhancement systems. Several NN-based classification models including multilayer perceptron (MLP), radial basis function (RBF) network, and support vector machines (SVM) with linear, polynomial, and RBF kernels were developed for signature extraction and device identification. These models were trained and tested using spike train data gathered from the Fourier analysis of the input current waveform in the presence of multiple devices. The performance of these models was compared in terms of their accuracy, generalization ability, and noise tolerance limits. The results showed that MLPs and SVM were both able to determine the presence of devices based on their harmonic signatures with high accuracy. MLP was found to be the best signature identification method because of its low computational requirements and ability to extract the information necessary for highly accurate device identification.