Título
Context-informed conditional anomaly detection approach for wave power plants: The case of air turbinesOtras instituciones
IkerbasqueEnte Vasco de la Energía (EVE)
Biscay Marine Energy Platform
Versión
Postprint
Derechos
© 2022 ElsevierAcceso
Acceso embargadoVersión del editor
https://doi.org/10.1016/j.oceaneng.2022.111196Publicado en
Ocean Engineering Vol. 253. Artículo 111196. June, 2022Editor
ElsevierPalabras clave
Marine Renewable Energy monitoring
anomaly detection
Prognostics and health management
Turbine ... [+]
anomaly detection
Prognostics and health management
Turbine ... [+]
Marine Renewable Energy monitoring
anomaly detection
Prognostics and health management
Turbine
Power curve and monitoring [-]
anomaly detection
Prognostics and health management
Turbine
Power curve and monitoring [-]
Resumen
The reliability and energy production of wave power plants (WPPs) depend on sea-state conditions, operation efficiency and degradation of its constituent assets. Air turbines are key assets for the ef ... [+]
The reliability and energy production of wave power plants (WPPs) depend on sea-state conditions, operation efficiency and degradation of its constituent assets. Air turbines are key assets for the efficient and reliable operation of WPPs and ensuring their correct operation leads to enhance the efficiency of WPPs. However, the lack of run-to-failure data and scarce fault records hampers the development of predictive condition monitoring solutions. In this context, focusing on unsupervised health monitoring methods, this paper presents an air turbine conditional anomaly detection (CAD) approach with a practical case study tested and validated on the Mutriku wave power plant. In contrast to anomaly detection models, which model the health-state without taking into account the influence of the operating context, the proposed CAD approach learns the expected air turbine operation conditioned on specific sea-states information modelled through wave energy flux concepts. This is achieved through an ensemble of Gaussian Mixture models and the expectation–maximization algorithm. Results show that, the integration of sea-states in the anomaly detection learning process improves the discrimination capability of the CAD model compared with the anomaly detection model without sea-state information, reducing false positive events and improving the accuracy of the CAD model. [-]
Sponsorship
Gobierno Vasco-Eusko JaurlaritzaID Proyecto
info:eu-repo/grantAgreement/GV/Elkartek 2021/KK-2021-00021/CAPV/Modelización del comportamiento térmico de los transformadores para aplicaciones fotovoltaicas/TRASMOIIColecciones
- Artículos - Ingeniería [639]