| Autor: |
Gloaguen, Jean-Rémy, Ecotiere, David, Gauvreau, Benoit, Finez, Arthur, Petit, Arthur, Lebourdat, Colin |
| Přispěvatelé: |
Cadic, Ifsttar |
| Jazyk: |
angličtina |
| Rok vydání: |
2021 |
| Předmět: |
|
| Popis: |
The sound emergence is the main regulatory estimator for wind turbine noise in France. This criterion aims to limit their noise impact on local residents and is highly dependent on the variation of residual noise over time. Therefore, initially defined curtailment plans can sometimes become inadequate, in which case they cannot easily be updated without leading to significant production losses. Machine learning techniques allow today to consider the continuous estimation by measurements of the sound contribution of wind turbine noise in the ambient noise and thus its noise emergence, without needing to stop the wind farm. This operation makes it possible not only to regularly adapt these reduction plans, thus optimizing electricity production, but also limiting the possible noise annoyance for local residents. For this purpose, semi-supervised Non-negative Matrix Factorization method is considered, enhanced by a temporal regularity constraint. This approach combines a wind turbine dictionary designed on a learning basis and a free dictionary that allows the adaptation of the method to the variability of residual noise. Tests conducted on simulated measurements reveal satisfactory performances with mean estimation errors lower than 2 dBA for wind noise emergences lower than 5 dBA. Finally, the presence of these two types of dictionaries makes it possible to estimate the wind noise emergence according to one or the other depending on the predominance of the estimated wind turbine noise. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
