Ardakani, Mohsen Masoomi et Yang, Jianming.
2025.
« Data-driven feature selection for prediction of wind turbine vibrations ».
In Proceedings of the CSME-CFDSC-CSR 2025 International Congress (Montreal, QC, Canada, May 25-28, 2025)
Coll. « Progress in Canadian Mechanical Engineering », vol. 8.
Prévisualisation |
PDF
292 - Data-driven feature selection for .pdf - Version publiée Licence d'utilisation : Tous les droits réservés aux détenteurs du droit d'auteur. Télécharger (438kB) | Prévisualisation |
Résumé
The growing demand for renewable energy necessitates enhancing wind turbine performance and reliability. Wind turbines' operational efficiency and lifespan depend heavily on two critical vibration parameters: tower and drivetrain vibration. Accurate data-driven models for predicting wind turbine vibrations require optimal feature selection as a fundamental step. A supervisory control and data acquisition (SCADA) system dataset containing 301 features and about 30k samples provide the foundation for this study to systematically evaluate and analyze critical factors that impact tower and drivetrain vibrations. Advanced feature selection methods that combine correlation analysis with mutual information and feature importance from Random Forest and XGBoost models determine the impact of different parameters on turbine vibrations. The number of features in the dataset underwent preprocessing to maintain high-quality input by removing unnecessary and redundant data. Mutual information revealed hidden non-linear relationship patterns between features, and feature importance methods confirmed the crucial role of these parameters. The analysis shows that wind speed is the main contributing factor to tower acceleration measurements, and rotor speed is an essential variable for drivetrain vibrations. Research findings create knowledge that enables the development of predictive maintenance models and improved wind turbine reliability through feature selection methods.
| Type de document: | Compte rendu de conférence |
|---|---|
| Éditeurs: | Éditeurs ORCID Hof, Lucas A. NON SPÉCIFIÉ Di Labbio, Giuseppe NON SPÉCIFIÉ Tahan, Antoine NON SPÉCIFIÉ Sanjosé, Marlène NON SPÉCIFIÉ Lalonde, Sébastien NON SPÉCIFIÉ Demarquette, Nicole R. NON SPÉCIFIÉ |
| Date de dépôt: | 18 déc. 2025 15:08 |
| Dernière modification: | 18 déc. 2025 15:08 |
| URI: | https://espace2.etsmtl.ca/id/eprint/32359 |
Actions (Authentification requise)
 |
Dernière vérification avant le dépôt |