Density integration approach for probabilistic prediction of wind power generation based on ensemble weather forecast

Abstract

Prediction of wind power output becomes further important to realize compensation for output fluctuation through operation of reserve power sources and energy storage systems. In particular, the probability density prediction (PDP), which derives a conditional probability distribution of the forthcoming wind power output based on the latest observable information, is attractive from the perspective of providing rich information not only on the expected value of power generation output but also on its uncertainty. This study discusses a density integration approach for the PDP task utilizing ensemble weather forecast (EWF) results derived with the individual ensemble members. The approach focuses on sequences of probability densities for the wind power plant (WPP) output derived by individual ensemble members; these sequences are dynamically weighted and integrated according to its current plausibility by matching them with the latest actual power generation at the required prediction timing. The framework enables us to realize frequently updatable short-term PDP by focusing on fresh observations, while taking advantage of ensemble weather forecasts, which are computationally expensive and not easily updated frequently. The usefulness of the proposed framework is evaluated through benchmark experiments on data sets collected at several real-world WPPs.