Abstract

Seismic emissions from wind turbines (WTs) affect the quality of seismological measurements, especially for the investigation of local earthquakes or for the monitoring of geothermal operations. Not all sources of the observed WT emissions are understood, and other related patterns may even be unknown. However, understanding these emissions is crucial to address and mitigate this problem.

This study presents a workflow employing unsupervised machine learning techniques to identify patterns in WT emissions utilizing ground motion data collected during four months in 2022 and 2023 in the Inter-Wind project near the Tegelberg wind farm in southwest Germany.

Aiming to extract known and unknown patterns of seismic WT emissions in a data-driven fashion, the hierarchical clustering algorithm HDBSCAN is applied, allowing a layered investigation of the clustering results. To ensure robust clustering of the dataset, a translation-invariant representation is essential. As the scattering transform proves effective in multiple seismological studies, it is applied before the clustering using the scatseisnet Python package.

The clustering results reveal distinct patterns correlating with wind direction, WT rotation rate, and wind speed, contributing to a better understanding of seismic WT emissions. Thus, the developed workflow is an important step toward the decorrelation of seismic WT emissions from meteorological conditions.