Copula-based Probabilistic Prediction of Grid Frequency Dynamics

With the growing share of renewable energy sources in the electricity supply, monitoring the grid stability has become increasingly important. A major challenge is the modeling and prediction of grid frequency as a stability indicator based on external technoeconomic features recorded on an hourly basis.
Common models in the literature are based on the assumption of Gaussian distributions in which either no dependency between the points in time is modeled or, if a dependency is taken into account, it is only captured linearly. We present a data-driven approach to modeling grid frequency that constructs a copula-based probabilistic predictor from an existing point predictor that is able to account for the nonlinear dependence between time points. The construction is based on the probabilistic correction of the point predictors by feature space driven error estimation. Models that are corrected using the error distribution perform better on probabilistic evaluation measures than baselines that assume independence. In addition, our best copula-based model performs also better than a Gaussian prediction model that takes dependence by correlation into account. ... mehrOur simulation study also shows that the conventional models underestimate the uncertainty of the grid frequency evolution and our copula-based predictors can provide a more accurate representation of this uncertainty