# Multi-source harmonic estimation method for distribution networks based on variational modal decomposition

## TL;DR

This paper introduces a new method to estimate harmonics in power networks using data analysis and image-based machine learning techniques.

## Contribution

A novel multi-source harmonic estimation method using variational mode decomposition and an improved PSRGAN model for distribution networks.

## Key findings

- The method effectively maps power signals into images for improved harmonic estimation.
- Simulation and field tests confirm the method's accuracy in multi-source harmonic scenarios.
- The approach uses easily accessible data, showing strong potential for real-world applications.

## Abstract

To address the limitation of harmonic monitoring on the low-voltage side of distribution networks, this paper proposes a multi-source harmonic estimation method based on variational mode decomposition. The method integrates short-term test data with long-term power data. First, dominant harmonic users are identified through a strategy that combines Fisher optimal segmentation and derivative dynamic time warping. Second, an electrical data transformation approach is designed by combining variational mode decomposition with Gramian angular fields, which maps the power signals of dominant harmonic users and low-voltage side harmonic signals into pseudo-color Gramian power images and grayscale Gramian harmonic images, respectively. Finally, an improved PSRGAN (pix2pix-super-resolution generative adversarial network) model is constructed to train and learn from these images, establishing the mapping relationship between power data and low-voltage side harmonic data of the distribution network, thereby enabling the migration and generation of long-term low-voltage side harmonic monitoring data. Simulation cases and field measurements validate the effectiveness and accuracy of the proposed method in multi-source harmonic scenarios. Moreover, the required data are easily accessible, demonstrating strong potential for engineering applications.