Posters

Abstract

Accurate determination of wind turbine (WT) power curves under real operating conditions remains a major source of uncertainty in Annual Energy Production (AEP) assessment. Current IEC standards construct power curves using a one-dimensional Pmeasured (Vmeasured) representation, in which measured values Pmeasured and Vmeasured are averaged over 10-minute intervals to obtain P10min and V10min, followed by normalization and wind-speed binning to form pairs (Pi,Vi). This procedure implicitly assumes that the instantaneous wind-speed measurand Vt can be treated as a stationary stochastic input, whose uncertainty can be reduced by repeated measurements. In the atmospheric boundary layer, however, Vt exhibits large turbulence intensity (TI), typically in the range of 10–15%, representing an intrinsic physical variability of the wind field that cannot be eliminated by repeated measurements. In IEC practice, this TI is nevertheless transferred into the power domain through V10min and P10min, and wind-speed binning. In contrast, the instantaneous power measurand Pt of a WT varies much more slowly due to the large inertia of the rotor. This fundamental asymmetry is further amplified by measurement resolution: anemometers typically have a resolution of 0.1 m/s, corresponding to about 1% relative uncertainty near rated conditions, whereas modern electronic power meters achieve resolutions on the order of 0.1 kW, resulting in relative uncertainties below 0.01% for multi-megawatt WTs. Consequently, conditioning power measurements on wind-speed measurements, as prescribed by IEC, is methodologically suboptimal. The proposed three-dimensional (3-D) Power-Weighted Curve (PWC) framework replaces the conventional measured one-dimensional Pmeasured(Vmeasured) representation with a 3-D measurement matrix {CP,k, Vj, Pk,j}, forming a discrete representation of the instantaneous power measurand Pt as a function of the instantaneous wind-speed measurand Vt and the power coefficient CP,t. In this matrix, Vj denotes discrete wind-speed levels, while Pk,j represents the corresponding power values at each discrete state (CPk, Vj). The 3-D matrix can be regarded as a digital measurement pseudo-sample within a comparison-based measurement approach. For each measured pair (Pmeasured, Vmeasured), Vmeasured  is first mapped to the nearest discrete level Vj, after which Pmeasured is compared with the set of Pk,j values on the corresponding Vj-slice. The data density of Pk,j across Vj, conditioned on a given CP,k, defines a discrete conditional probability distribution, interpreted as the Power Weight. A weighted wind-speed average Vavg,k over all Vj defines a relation Vavg,k(Pk), which is regressed into V(P), inverted to P(V), discretized in 0.5 m/s steps to obtain Pi(Vi), and finally normalized to Pi,n for comparison with the IEC-derived Pi. Applications to operational wind-farm datasets demonstrate that the PWC framework reduces power-curve measurement uncertainty by a factor of 2–4, from typical IEC-level values of about 10% down to approximately 3%. The resulting power curves yield AEP estimates that are 20–25% lower than OEM-based predictions, indicating a systematic overestimation in current practice. In addition, PWC significantly simplifies the measurement process, reducing the required amount of useful measurement data from at least 25 hours to roughly 8 hours.