Presentations
Siblings:
ProgrammeSpeakersPostersContent PartnersCall for university proposalsPresenters’ dashboardEDiGaPS: A novel machine learning-based approach to model the influence of turbulence intensity on wind turbine power curves
Nathianne Andrade, Head of Performance Engineering, Delfos Energy
Session
Abstract
Wind turbines are installed in various wind climates, each with unique conditions, but they are often developed and tested in locations with different wind characteristics. This creates a need for normalization across different wind conditions, including factors like air density and turbulence intensity. Turbulence intensity, in particular, impacts wind turbine power output in multiple ways. Some effects are physical, such as the aerodynamic and inertia-related forces on the turbine, but another effect is the bias induced by averaging measured power output and wind speed over 10-minute intervals. Annex M of the IEC 61400 12-1 standard offers a complex methodology that only addresses the correction for the 10-minute averaging bias, covering allegedly 50% of the total turbulence effect. This paper introduces the Equivalent Difference in Gaussian Process Surface (EDiGaPS), a novel machine-learning approach for correcting turbulence effects on wind turbine power curves using data from the SCADA system of wind turbines. The corrections provided by EDiGaPS were compared with the ones obtained from the methodology proposed by the IEC standard. On average, the new procedure corrected 60.2% of the total turbulence influence in the tests performed, which is good evidence that it can handle not only the effect of 10-minute time-averaging, but also other possible effects of turbulence. Furthermore, implementing and comparing both methods, we noticed an already mentioned effect in the literature: the renormalization method tends to cause an overcompensation that leads to too high AEPs; the novel proposed approach does not present such an effect. To the best of our knowledge, our work represents an important contribution, due to the fact that, although the renormalization method has existed for some years, there are very few works in the literature addressing the topic, possibly due to the limited adoption by the community.