7 - Identification of source harmonics in electrical networks using spatiotemporal approaches

Abstract

The harmonics are the main cause of malfunction of the components in modern distribution systems. Mechanical and electrical failures can affect equipment if they are not properly filtered. As the distribution technologies evolves, the nature of the harmonics observed becomes complex and the identification task cannot be done with conventional methods. Machine learning methods (ML) are taking advantage of data-driven methods for characterizing the harmonics of large data measurement sets. In this chapter are evaluated two spatio-temporal approaches for identification of the harmonic distortion in distribution systems and that can be implemented as part of the ML algorithms due their accuracy in feature extraction. Firstly, the dynamic mode decomposition (DMD) method, is optimized to estimate the behavior of a system and extract the multiple amplitude and frequency components, including harmonics, mode shape, and participation factors simultaneously, through which the measurement matrix is created by appending multiple power signals. Secondly, the Padé approximation theory permits the estimation of amplitudes, frequencies, and phases of harmonics. The presented based-Padé method takes advantage of the singular value decomposition being possible with this, the extraction of features embedded in bulk simultaneous signals. Finally, both techniques are tested on the 14 bus IEEE test system for the assessment of the harmonic pollution. Several harmonic stress conditions are evaluated in both balanced and unbalanced conditions.