Features generation by means of currents' physical components for load identification

In this paper the process of extracting a set number of features that uniquely describe a nonlinear load is described. The method is used for classifying and identifying as well as determining the operation state of these harmonic nonlinear loads. The method is based on harmonic spectral decomposition of periodic waveforms and on sorting and calculating the features by means of Currents' Physical Components Theory. Using this theory enables to create features with physical meaning and to enable to calculate the actual various currents and powers of these loads. These features are then used in order to train an Artificial Neural Network which is used to identify and calculate the powers and currents of the identified load. The presented theory is implemented on a custom made simulator which uses simulated waveforms, as well as measured ones attained from power quality monitors.