Short-term and regionalized photovoltaic power forecasting, enhanced by reference systems, on the example of Luxembourg. (March 2019)

• Record Type: Journal Article
• Authors: Koster, Daniel; Minette, Frank; Braun, Christian; O'Nagy, Oliver
• Is Part Of: Renewable energy. Volume 132(2019)
• Page Start: 455
• Page End: 470
• Publication Date: 2019-03
• Subjects: Photovoltaic forecasting -- Forecasting performance -- RMSE -- Photovoltaic integration -- Solar forecasting -- Solar energy integration; Renewable energy sources -- Periodicals; Power resources -- Periodicals; Énergies renouvelables -- Périodiques; Ressources énergétiques -- Périodiques; 333.794
• DOI: 10.1016/j.renene.2018.08.005
• Languages: English
• ISSNs: 0960-1481

Abstract

Abstract: The authors developed a forecasting model for Luxembourg, able to predict the expected regional PV power up to 72 h ahead. The model works with solar irradiance forecasts, based on numerical weather predictions in hourly resolution. Using a set of physical equations, the algorithm is able to predict the expected hourly power production for PV systems in Luxembourg, as well as for a set of 23 chosen PV-systems which are used as reference systems. Comparing the calculated forecasts for the 23 reference systems to their measured power over a period of 2 years, revealed a comparably high accuracy of the forecast. The mean deviation (bias) of the forecast was 1.1% of the nominal power – a relatively low bias indicating low systemic error. The root mean square error (RMSE), lies around 7.4% - a low value for single site forecasts. Two approaches were tested in order to adapt the short-term forecast, based on the present forecast deviations for the reference systems. Thereby, it was possible to improve the very short term forecast on the time horizon of 1–3 h ahead, specifically for the remaining bias, but also systemic deviations can be identified and partially corrected (e.g. snow cover). Highlights: A hybrid approach for PV-power forecasting, using metered PV-systems as references. Demonstrating a comparably accurate forecast performance on our case study over a two years period. Bottom-up model, able to reach high spatial resolutions if the data is available.