A Deep Learning-Based Method for Harmonic Spectrum Analysis

This paper proposes a novel method for harmonic spectrum analysis based on 2D CNN(Two-Dimensional Convolutional Neural Network) to overcome the limitations of conventional techniques. The increasing use of power conversion devices with switching behavior and nonlinear load characteristics has become a major source of harmonics. For effective harmonic signal analysis, various spectrum analysis methods such as FT(Fourier Transform), FFT(Fast Fourier Transform), and WT(Wavelet Transform) are known. However, existing methods rely heavily on numerical signal data and parameter settings, which limit their accuracy and applicability when sufficient data are not available. To address these limitations, this paper introduces a deep learning-based method that can effectively estimate harmonic components and amplitudes without requiring numerical signal data and parameter settings. The performance of the proposed method was validated through various case studies using PSCAD/EMTDC.