Augmenting insights from wind turbine data through data-driven approaches

• Published in: Applied energy 2024-12, Vol.376, p.124116, Article 124116
• Main Authors: Moss, Coleman, Maulik, Romit, Iungo, Giacomo Valerio
• Format: Article
• Language: English
• Subjects: Machine learning; SCADA data; Wind farm; Wind turbine
• ISSN: 0306-2619
• DOI: 10.1016/j.apenergy.2024.124116

Data-driven techniques can enable enhanced insights into wind turbine operations by efficiently extracting information from turbine data. This work outlines a data-driven strategy to augment these insights, describing its benefits and limitations. Different data-driven models are trained on supervisory control and data acquisition (SCADA) and meteorological data collected at an onshore wind farm. The developed models are used to predict wind speed, turbulence intensity (TI), and power capture for each turbine with excellent accuracy for different wind and atmospheric conditions. Modifications of the incoming freestream wind speed and TI due to the evolution of the wind field over the wind farm and effects associated with operating turbines are captured enabling modeling at the turbine level. Farm-level modeling is achieved by combining models predicting wind speed and TI at each turbine location from inflow conditions with models predicting power capture. Data-driven filters are also considered in the context of generating accurate data-driven models. In contrast to many current works that utilize simulated data, the proposed approach can describe subtle phenomena, such as speedups, TI damping, and wake-generated turbulence, from real-world turbine data. It is noteworthy that the accuracy achievable through data-driven modeling is limited by the quality of the data; therefore, guidelines are proposed to estimate resultant model performance from a given training set without the need to train or test a model. •Machine learning models are trained from real-world wind turbine data to uncover wind farm phenomena.•Linear models are developed to estimate model performance from the statistics of a given data set without the need to train or test any models.•Machine learning filters for SCADA data are compared and an optimal, new approach is proposed.