Seminar, Hosted by Asymptotic Methods in Statistics Series
Renewable energy-related aspects of statistical post-processing
Department of Applied Mathematics and Probability Theory, Faculty of Informatics, University of Debrecen, Hungary
- Statistics, Meteorology and
Abstract
By the end of 2025, the renewable energy share of global electricity capacity reached 49.4%, and new installations are dominated by solar energy. Because photovoltaic (PV) energy sources are highly volatile, integrating them into the electrical grid requires accurate short-term PV power forecasts. PV power is predicted from forecasts of solar irradiance, where the most advanced method is the probabilistic approach based on ensemble forecasts. Ensemble forecasts, however, are often underdispersive and subject to systematic bias, so they require statistical post-processing. The talk first introduces a general two-step machine-learning-based approach to calibrating ensemble forecasts; in a case study based on global horizontal irradiance forecasts from the operational ensemble prediction system of the Hungarian Meteorological Service, its predictive performance is compared with the raw ensemble and with state-of-the-art parametric approaches. It then investigates direct post-processing of PV power ensemble forecasts, using power output data from seven PV power plants in Hungary and corresponding PV power ensemble forecasts derived from ECMWF ensemble weather predictions to compare modern parametric and nonparametric approaches. Joint work with Ágnes Baran (University of Debrecen), Nina Horat (Karlsruhe Institute of Technology), Sebastian Lerch (Marburg University, HITS) and Martin János Mayer (Budapest University of Technology and Economics).