Evaluation of a distributed nowcasting system to support ancillary services and grid restoration with wind power plants

Inhalt:
In this article an approach to distributed wind power forecasts is presented, which addresses the challenges of wind energy integration in grid operation and effectively making use of highly volatile power sources, such as wind power plants (WPPs). Forecasts are generated on-site of the WPP in order to maximize the availability of forecast data to grid operators. To this end an adaptively trained machine learning model uses locally available sensor data to predict the available active power (AAP) signal in a probabilistic fashion. Furthermore, a physics-based forecast can be deposited on-site the WPP from a central computing cluster and combined with the locally generated data-based forecast. We evaluate the performance of the method in a laboratory environment and find that the locally generated forecast significantly improves forecast quality for a short-term horizon, and that the forecast provides a high level of reliability when predicting the minimal AAP signal, which is highly relevant for enabling power reserve provision from WPPs.