Solar power forecasting using machine learning and ECMWF weather variables

Soós, Viktória and Markovics, Dávid and Mayer, Martin János (2025) Solar power forecasting using machine learning and ECMWF weather variables. In: 2025 10th International Youth Conference on Energy (IYCE). Institute of Electrical and Electronics Engineers (IEEE), Budapest.

Preview

Text SolarpowerforecastingusingmachinelearningandECMWFweathervariables.pdf - Published Version Download (458kB) | Preview

Abstract

Day-ahead power forecasts play a key role in managing the electricity distribution system, but the weather dependency of renewables is a major challenge. State-of-the-art solar power forecasts are based on numerical weather prediction model outputs. Forecasts of a multitude of weather variables issued by the European Centre for Medium-Range Weather Forecasts are utilized to provide a solid basis for this research. The 31 weather variables, zenith and azimuth angle were first investigated based on correlation coefficients, which resulted in a reduction to 27 variables to avoid multicollinearity. Stepwise regression is used to identify the order of importance of the predictors and to determine the combination of predictors that results in the most accurate forecasts. The post-processing reduced the root mean square error of the global horizontal irradiance (GHI) predictions from 68.1 W/m2 to 64.4 W/m2 . The corrected GHI forecasts are converted to photovoltaic (PV) power by a physical model chain. The PV power forecasts created from the post-processed GHI input data are more accurate than those created from the raw weather forecasts.

Actions (login required)

Edit Item