Predicción de fallas en los aerogeneradores de la Central Eólica Villonaco, aplicando la técnica de aprendizaje profundo basado en transformadores.
Wind energy represents a renewable and sustainable alternative that can contribute to reducing pollution generated by the use of fossil fuels. The Villonaco wind farm (CEV) supplies the Ecuadorian electrical system, and a stoppage caused by a failure in its wind turbines can be costly and involve lo...
Wedi'i Gadw mewn:
| Prif Awdur: | |
|---|---|
| Fformat: | bachelorThesis |
| Iaith: | spa |
| Cyhoeddwyd: |
2024
|
| Pynciau: | |
| Mynediad Ar-lein: | https://dspace.unl.edu.ec/jspui/handle/123456789/29693 |
| Tagiau: |
Ychwanegu Tag
Dim Tagiau, Byddwch y cyntaf i dagio'r cofnod hwn!
|
Ychwanegu Tag | Crynodeb: | Wind energy represents a renewable and sustainable alternative that can contribute to reducing pollution generated by the use of fossil fuels. The Villonaco wind farm (CEV) supplies the Ecuadorian electrical system, and a stoppage caused by a failure in its wind turbines can be costly and involve long downtime. Therefore, it is important to predict failures in the different components. This research focuses on the converter, especially on the IGBT module, because this component has had the most replacements during the operation time of the CEV. Deep learning based on transformer models is a new and powerful technique that is ideal for working with large datasets. Therefore, it will be used to predict failures in the IGBT module in the 11 wind turbines of the CEV. Initially, there are 1.63 GB of data corresponding to the operation of the wind turbines, as well as the SCADA (Supervisory Control and Data Acquisition) alarm file and the Operation and Maintenance (O&M) plan carried out in the period 2014-2021. These data are preprocessed and variable selection methods are applied; in addition, a data filtering process is performed to ensure that the data correspond only to the “normal” operation of the wind turbine. Then, two methodologies based on Transformers, capable of detecting anomalies in SCADA data, are applied. To these methodologies, a 2-stage evaluation was implemented to try to predict the occurrence of a fault. Experimental results show that the model is able to predict failures in the IGBT module with an average F1 Score higher than 90% and an average advance of 5.08 months. Keywords Wind turbines, SCADA data analysis, IGBT module, deep learning, Transformers, deep anomaly detection. |
|---|