Beyond traditional wind farm noise characterisation using transfer learning.

Area of Science:

• Acoustics
• Artificial Intelligence
• Environmental Science

Background:

• Wind farm noise (WFN) assessment traditionally relies on spectral analysis and overall noise descriptors.
• Existing methods may not fully capture the complex spatial and temporal characteristics of WFN.

Purpose of the Study:

• To propose and evaluate a novel approach for characterizing and assessing wind farm noise.
• To leverage deep learning for extracting advanced acoustic features for WFN analysis.

Main Methods:

• Extraction of acoustic features between 125 and 7500 Hz using a pretrained deep learning model, termed deep acoustic features.
• Analysis of measured data from various locations to link deep acoustic features with noise characteristics.

Main Results:

• Deep acoustic features correlate with meaningful characteristics of wind farm noise.
• This approach provides an improved spatial and temporal representation of WFN.
• Demonstrated superiority over traditional spectral analysis and overall noise descriptors.

Conclusions:

• The proposed deep acoustic feature approach shows promise for WFN assessment.
• This method could form the basis for a future, enhanced framework for wind farm noise evaluation.