Application of Clustering and Dimensionality Reduction Methods for Finding Patterns on Supraharmonics Data

Supraharmonics (waveform distortion between 2 and 150 kHz) proliferate in electrical installations due to the increasing use of power electronics converters and power-line communication. Due to the wide range that the supraharmonics cover and the high frequency resolution needed to measure them, a considerable amount of data is acquired. The analysis is usually done manually by experts. More efficient methods for data mapping and analysis are needed. Machine learning methods are explored in this paper for the analysis of supraharmonics data.

I. Introduction

Supraharmonics are waveform distortion in the range from 2 up to 150 kHz. They are emitted unintentionally by power electronics converters embedded in electrical devices. The rapid advancement of technology in the field of power electronics, encouraged by the need of meeting global goals of electrification and efficient use of electricity, has led to more emissions (in amplitude and frequency) in the supraharmonic range being injected and being present in public grids and industrial installations [1]. Knowledge about methods to rapidly analyze supraharmonics are not generalized and standardized in the industrial community and, when facing supraharmonics-related problems, mapping and characterization of emission based on measurements are needed.