Comprehensive Data Investigation of Harmonic Measurements from Wind Parks for Assessing Mechanisms and Long-Term Variations

This study aims to conduct a comprehensive measurement analysis for a wind park located in

Brazil, employing a two-stage methodology in one year of data. First, it evaluates time-series parameters related to harmonic distortion and wind park operation at the PCC, applying statistical metrics and comparative assessments. Second, it implements unsupervised deep learning and clustering techniques to automatically extract features from parameters associated to the content of voltage and current spectra. The second step involves the use of deep autoencoder for dimension reduction and k-medoids algorithm for clustering the extracted features. These methods help identify patterns by grouping the data and extracting insights related to power variations. The results aim to determine which harmonic components are most affected by the wind park operation under different conditions and periods, and to investigate which components are primarily influenced by variations in background distortion.

Contributions of this work include enhanced visualization and improved assessment of harmonic interaction dynamics, revealing hidden features in large-scale measurement datasets.