Multivariate SCADA Data Analysis Methods for Real-World Wind Turbine Power Curve Monitoring

Due to the stochastic nature of source, wind turbines operate under non-stationary conditions and extracted power depends non-trivially on ambient working parameters. It is therefore difficult establish a normal behavior model for monitoring performance turbine most employed approach be driven by data. The curve relation between intensity widely performance. On grounds above considerations, recent trend regarding analysis consists incorporation main parameters (as, example, rotor speed or blade pitch) as input variables multivariate regression whose target power. In this study, method proposed: it based sequential features selection, which employs Support Vector Regression with Gaussian Kernel. One innovative aspects study that set possible covariates includes also minimum, maximum standard deviation important environmental operational variables. Three test cases practical interest are contemplated: Senvion MM92, Vestas V90 V117 owned ENGIE Italia company. shown selection remarkably aspect should taken in consideration order customize data-driven curve. obtained error metrics competitive general lower respect state art literature. Furthermore, operation abundantly selected feature algorithm: result indicates richness measurement channels contained Supervisory Control And Data Acquisition (SCADA) data sets exploited reliably possible.