Towards seamless and high-resolution renewable energy forecasting by the combination of approaches and data sources

Operational forecasting models of Renewable Energy Sources (RES) are usually developed for specific time frames, using a reduced set of data sources as a function of the forecasting horizon. This results in discontinuities in predictions between the different time frames from the next minutes to the next days, which is detrimental to decision-making in power systems. Although advanced forecasting model may lower forecasting error, an alternative and maybe more impactful option consists in expanding the set of potential sources for RES forecasting, going beyond recent power measurements, classical Numerical Weather Prediction and information from satellites. It is therefore crucial for next-generation RES forecasting to propose continuous, seamless predictions that harvest the full potential of heterogeneous data sources. This presentation highlights the different solutions developed in the Horizon2020 project Smart4RES for seamless RES forecasting based on the combination of multiple data sources, including high-resolution weather measurements and forecasts.

A step towards high-resolution RES forecasting, i.e. temporal resolutions below 5 min and spatial resolutions at the scale of 100 m, is achieved by the integration of high-resolution into RES forecasting models. The use of lidar measurements for the minute-ahead power forecasting of wind turbines improves RMSE against persistence. The production of wind speed forecasts at a 100m-30s resolution thanks to Large Eddy Simulation (LES) at different wind farms of an isolated power systems enables to predict the total variability of wind power production on 10-min rolling intervals, with higher reliability than an approach based on traditional NWP that have a lower spatio-temporal resolution.

The combination of multiple data sources has proven to be efficient for the improvement of RES forecasting, especially at intraday horizons. Smart4RES proposes optimal combinations for both weather and RES forecasting. The combination of high-resolution irradiance maps derived from a network of All-Sky-Imagers (ASI) with satellite images outperforms the predictions of the ASI network only. Similarly, the combination of filter-based models for Wind and PV forecasting improves the forecasting scores of the individual models.