Please use this identifier to cite or link to this item:

http://hdl.handle.net/10609/118007
Title: Uso de técnicas de minería de datos para la detección de anomalías y predicción de la producción en una turbina eólica
Author: Pérez Esteras, Javier
Director: Casas Roma, Jordi
Tutor: Parada Medina, Raúl
Keywords: machine learning
energy prediction
anomaly detection
Issue Date: Jun-2020
Publisher: Universitat Oberta de Catalunya (UOC)
Abstract: In recent years, there has been a significant increase in the generation of electricity from renewable energies, especially wind power. This raises special interest on this sector. The present work validates the use of data mining algorithms for anomaly detection and preventive maintenance tasks on turbines. It uses only SCADA control variables, it does not require specific CMS system. The project has validated and compared algorithms determining relationships between parameters during regular operation. These algorithms included random trees, gradient boosting, support vector machines for regression, and multilayer neural networks. We have made use of a proprietary data log for these purposes. Power loss and misalignment are common faults on wind turbines. We have simulated them while applying anomaly detection techniques. These were based on auto encoders and statistical methods, with the latter achieving better results. These techniques were able to detect a 10% power loss with energy production levels above 20% of the nominal value, performing with 100% precision level and showing 97% of sensitivity. Power prediction based on historic series has also been validated and compared using different methods such as auto regression, LSTM, and ARIMA, giving this one the best result. This work has proved and validated the feasibility in the use of data mining techniques for maintenance tasks in wind turbines. We anticipated failures by detecting anomalies in systems without specific CMS, and predicted production levels using historic data.
Language: Spanish
URI: http://hdl.handle.net/10609/118007
Appears in Collections:Bachelor thesis, research projects, etc.

Share:
Export:
Files in This Item:
File Description SizeFormat 
jpesterasTFM0620memoria.pdfMemoria del TFM4.43 MBAdobe PDFView/Open
jpesterasTFM0620presentación.pdfPresentación del TFM1.97 MBAdobe PDFView/Open

This item is licensed under a Creative Commons License Creative Commons