A global sensitivity analysis method applied to wind farm power output estimation models

Abstract

This paper proposes a global sensitivity analysis method applied to wind farm power output estimation models. The relevance of a global sensibility analysis is that it allows quantification of the contribution of the uncertainty of each input variable of the estimation model to the uncertainty of the response of the model. Measures of sensitivity based on Sobol' indices have been used in the field of energy. Sobol’ indices are constructed based on the assumption that the model input variables are statistically independent. The method proposed in this paper uses Shapley effects and a regular vine copula to take into account the probable dependency among the input variables of the models. The model used as case study is fed with sixteen meteorological variables and six operational variables. The following are some of the most important results obtained in the case study: a) Regular vine copulas were more suitable than the drawable vine and canonical vine subclass copulas to simulate the structure of dependency between the random input variables of the used wind farm power output estimation model; b) the Shapley effects were able to overcome the difficulty of interpretation presented by the Sobol' indices with respect to correlations among the input variables of the wind farm power output estimation model. In the case study, the wind speed, active power set-point and turbulence intensity variables explained 98.58% of the variance of the response of the model. The wind direction, nacelle orientation and air density variables only explained 1.42% of that variance.