Performance Assessment of Wind Turbines: Data-Derived Quantitative Metrics

Abstract

Deteriorating performance of wind turbines results in power loses. A two-phase approach for performance evaluation of wind turbines is presented at past and future time intervals. Historical wind turbine data is utilized to determine the past performance, while performance at future time horizons calls for power prediction. In phase I of the proposed approach, wind power is predicted by an ensemble of extreme learning machines using parameters such as wind speed, air temperature, and the rotor speed. In phase II, the predicted power is used to construct Copula models. It has been demonstrated that the parameters of the Copula models serve as usable metrics for expressing performance of wind turbines. The Frank Copula model performs best among the five parametric models tested.