Debrecen4U
October 29, 2025
Debrecen researchers develop advanced nodel for more accurate renewable energy forecasting
A researcher from the University of Debrecen has been working on statistical post-processing of weather forecasts, which allows for more accurate forecasts of weather variables required for renewable energy production.
Associate Professor Ágnes Baran is a member of the research group operating at the Faculty of Informatics, which is unique in the country, dealing with statistical post-processing. Their work and a scientific publication based on it were awarded the Publication Award by the Gróf Tisza István Foundation for the University of Debrecen, unideb.hu reports.
The university experts did not deal with a simple forecast of a weather variable but took into account aspects during the research that also have a clearly demonstrable economic utility. The importance of solar energy is continuously increasing in Hungary, and the area of use of renewable energy sources is expanding.
Focusing on forecasts of wind speed and solar radiation measured at a height of 100 meters, the researchers combined machine learning techniques with traditional post-processing methods in order to develop a mathematical model that can provide the most accurate forecasts based on data from a few wind farms and solar farms, as well as from HungaroMet (Hungary’s official meteorological service provider).
“The given models depend on the weather variable we want to forecast and, of course, can also depend on the specific station data. The same models do not necessarily work at a station in the Great Plains as, say, in the Alps, so when building the model, we aimed to make the system applicable to different stations and different data. The validation was specifically tailored to Hungarian data, to Hungarian stations. We worked with real data, so we used part of it to build the model and determine its parameters and the other half for testing, so we could check whether the algorithm was really capable of providing good forecasts,” explained the associate professor of the Faculty of Informatics.
As she said, they worked with a so-called rolling learning period, so the model parameters, the forecasts initialized on a given day, were always determined based on the experiences of the previous few days, 51 in the case of wind, and the data of the previous 30 days in the case of solar radiation.
“The model always has to be retuned; a training phase always has to be included. An earlier version of this work has already been put into operational use by HungaroMet, and it is using it to make forecasts,” she added.
Ágnes Baran highlighted that this is an area of high research interest at an international level, and the research group has direct professional ties with the European Centre for Medium-Range Weather Forecasting in Reading, as well as with the Heidelberg Institute for Theoretical Studies, one of the most important scientific workshops in the field. In Hungary, researchers from the University of Debrecen are collaborating with Budapest University of Technology and Economics (BME). In Hungary, research on weather forecasts modelled using machine learning does not yet have a long history, but UD researchers are considered pioneers in Hungary.
The results are clearly demonstrable, and it can be determined using metrics to determine how much the post-processing technique has improved the raw forecasts. Forecasts that are as accurate as possible also have a significant financial impact.
“In Hungary, solar farms and power generation centers have a scheduling obligation, meaning they must indicate how much energy they will produce in 15-minute time steps for a given time horizon (48 hours), but if they deviate significantly from this, they must pay a penalty. The quality of the forecast may also determine whether energy needs to be purchased or whether other sources need to be relied on in this area. Through the research, we have presented a new technique that, if further developed, can make the forecasts more precise in the case of any weather variable. It is therefore possible to determine more precisely what proportion of the generated electricity can come from solar energy and how much needs to be produced by other methods,” added the associate professor of Faculty of Informatics of UD.
The study entitled “A two-step machine learning approach to statistical post-processing of weather forecasts for power generation,” which won the GTIDEA and Debrecen University Publication Award, was published in the Quarterly Journal of the Royal Meteorological Society of the British Royal Meteorological Society.
Source:dehir.hu | Photo credit: Facebook Debrecen városa
