Uncertainty-Aware Wind Turbine Power Curve Modelling with Rotor Turbulence Intensity Asymmetry Using Two-Stage Gaussian Process Regression

Abstract

Accurate power curve modelling under variable atmospheric conditions is essential for wind energy project design and resource assessment. Current practice assumes isotropic turbulence intensity across the rotor, but rotor-scale turbulence intensity asymmetry can significantly amplify power output variability, yet remains absent from standard IEC specifications and power curve modelling literature. This study systematically investigates how rotor turbulence intensity asymmetry contributes to power variance through aeroelastic simulations of a 15 MW reference turbine. A full-factorial experimental design explores the combined influence of wind speed, wind shear, hub-height turbulence intensity, and rotor turbulence intensity asymmetry, generating a dataset of 5,760 simulations across 36 atmospheric conditions and 20 wind speeds. Statistical analysis reveals that while wind speed and hub-height turbulence intensity dominate power variance globally, rotor turbulence intensity asymmetry exhibits pronounced regime-dependent effects. Variance amplification ranges from approximately 37 to 40% in below-rated and near-rated operation, where power output is tightly coupled to inflow conditions, to 24% in above-rated operation where pitch control attenuates but does not eliminate the effect. A two-stage heteroscedastic Gaussian Process Regression framework is developed to capture this regime-dependent structure by separately modelling mean power and input-dependent variance as functions of all four atmospheric parameters. The proposed framework improves prediction accuracy by 27% and reduces prediction uncertainty by 25% relative to univariate wind-speed-only models, demonstrating the value of atmospheric conditioning for probabilistic power curve estimation. The approach provides a practical tool for design-phase uncertainty quantification and wind resource assessment in operating conditions where rotor-scale turbulence variation is pronounced but remains neglected in standard modelling practice.