Power harmonics analysis for electrical device signature identification using the artificial neural network and support vector machine

Abstract

A novel idea of nonintrusively identifying the electrical devices present in a power system by analyzing the current waveform harmonics using the artificial neural network (ANN) or support vector machine (SVM) was developed. The research has shown that different devices produced distinctly different current harmonic signatures. Various ANN architectures and SVM-based classifiers with different kernels were compared. The ANN and SVM-based classifiers were trained to map phase angles and magnitudes (represented in the complex form) of the current waveform harmonics to the combinations of devices present in the system. The trained ANN and SVM-based models performed with high classification accuracy. Classification of harmonic signatures from multiples of electrically identical devices was also studied. The multilayer perceptron ANN, which was the best classifier due to its high accuracy yet low computational resource requirement, was further optimized by evolving its weights using genetic algorithm.