Open AccessTransferability of site-dependent wind turbine performance predictions using machine learning
Author(s) -
Florian Hammer,
Sarah Barber
Publication year - 2022
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/2151/1/012006
Subject(s) - transferability , wind power , turbine , computer science , work (physics) , wind speed , power (physics) , machine learning , artificial intelligence , simulation , engineering , meteorology , mechanical engineering , geography , physics , logit , quantum mechanics , electrical engineering
Within this work, machine learning models of site-specific machine learning models of wind turbine power curves of the Beberibe Wind Farm in Brazil, which consists of 32 turbines and one met mast, were developed. Previous work already showed that machine learning models taking into account site-specific effects can increase power prediction accuracy of single wind turbines by a factor of three compared to the standard power curve binning method. The main goal was to investigate the transferability of these models through power output predictions of various turbines depending on the distance from the met mast. It was found that transferring models within a wind farm is possible, but a decrease of prediction accuracy by up to 30% in certain cases could be observed. Neither the combination of various turbine data, nor the incorporation of site-specific data had an apparent effect on the transferability performance. It was thus concluded that a further investigation is needed, where a larger and more distributed subset of turbines should be used.