Predicting wind power variability events using different statistical methods driven by regional atmospheric model output

摘要: Variability in power generation from wind farms is an important issue the energy industry. If sub-hour variability events can be predicted, potential disruptions to grid operations might mitigated. Using 4 years of 5 min data Australian Energy Market Operator for 80 MW farm south-east Australia, we fit statistical models on meteorological reanalysis US National Centers Environmental Prediction. The Prediction fields were transformed into spatial empirical orthogonal functions, and 6 h projections onto these became explanatory covariates generalized linear, random forest (RF), gradient boosting support vector machine classification models. Other considered local speed 6 h-lagged function differences. Models selected by minimizing cross-validated misclassification rate assessed using area under receiver operating characteristic curve reliability score. Considering performance ease tuning, RFs preferred. Performance was poorer larger ramps. accurately predicted their validation set. For asymmetric costs (miss-to-false alarm cost ratio = 10), yielded competitive low-cost Support machines produced slightly superior but needed tuned manually. RF atmospheric model output provide a robust approach predicting relatively large ramp events. We recommend as practical skilful method feed early warning system energy/electricity operators. Copyright © 2014 John Wiley & Sons, Ltd.