Applied Machine Learning Techniques for Performance Analysis in Large Wind Farms

John Thomas Lyons*, Tuhfe Göçmen

*Corresponding author for this work

    Research output: Contribution to journalJournal articleResearchpeer-review

    169 Downloads (Pure)


    As the amount of information collected by wind turbines continues to grow, so too does the potential of its leveraging. The application of machine learning techniques as an advanced analytic tool has proven effective in solving tasks whose inherent complexity can outreach expert-based ability. Such is the case presented by this study, in which the dataset to be leveraged is high-dimensional (79 turbines × 7 SCADA channels) and high-frequency (1 Hz). In this paper, a series of machine learning techniques is applied to the retrospective power performance analysis of a withheld test set containing SCADA data collectively representing 2 full days worth of operation at the Horns Rev I offshore wind farm. A sequential machine-learning based methodology is thoroughly explored, refined, then applied to the power performance analysis task of identifying instances of abnormal behaviour; namely instances of wind turbine under and over-performance. The results of the final analysis suggest that a normal behaviour model (NBM), consisting of a uniquely constructed artificial neural network (ANN) variant trained on abnormality filtered dataset, indeed proves effective in accomplishing the power performance analysis objective. Instances of over and under performance captured by the developed NBM network are presented and discussed, including the operation status of the turbines and the uncertainty embedded in the prediction results.
    Original languageEnglish
    Article number3756
    Issue number13
    Number of pages28
    Publication statusPublished - 2021


    • Machine learning
    • Performance monitoring
    • Artificial neural networks
    • Long short-term memory
    • Wind farm operation and monitoring
    • Wind farm power curve


    Dive into the research topics of 'Applied Machine Learning Techniques for Performance Analysis in Large Wind Farms'. Together they form a unique fingerprint.

    Cite this