Revisitando o problema de estimação de curvas de potência de aerogeradores via modelos polinomiais esparsos

Abstract

In this dissertation, the classical polynomial model for wind turbines power curve estimation is revisited aiming at an automatic and parsimonious design. For this purpose, with the help of genetic algorithms we introduce a methodoloy aiming at estimating the optimal order of the polynomial as well the relevant terms in this polynomial. The proposed methodology is compared with the state of the art in estimating the power curve of wind turbines, such as logistic models (with 3, 4 and 5 parameters), artificial neural networks and weighted polynomial regression. We also show that the proposed approach performs better than the standard LASSO approach for building regularized sparse models. The results indicate that the proposed methodology consistently outperforms all the evaluated alternative methods.