We must share the results of analyses conducted using state-of-the-art methods

Attila Aszódi and his research team investigated the potentials of the solar, wind and nuclear energy mix using machine learning.

The research team was looking at the probabilities of carbon-free electricity share within the electricity system by developing solar and wind energy production, and the possibilities of energy mixes produced by nuclear power plants and weather-dependent renewable energy production units.  The study, published in Applied Energy, one of the most prestigious journals in the international energy science community, with an impact factor of 11.4, has attracted the attention of the a national press.

The penetration of highly weather-dependent solar and wind power generation in Europe poses significant challenges to the electricity system, as their output varies significantly throughout the year and even within a day, and does not match the variation in electricity demand over time. In addition, there may be the periods when both solar PV and wind power plants are unable to generate power or the amount they feed into the system is very low. The use of nuclear energy, which is also carbon neutral, would be a great contribution to the security of energy supply.

We interviewed Attila Aszódi, Dean of the Faculty of Natural Sciences at BME (BME TTK) and a professor of the Institute of Nuclear Techniques about the study and its media coverage.


Have similar calculations been conducted in the field of energetics using machine learning or artificial neural networks?

Attila Aszódi: We have extensive literature on machine learning methods, including the application of artificial neural networks to solve energy problems. Let me mention here the first author of our article, Martin János Mayer, a young researcher of the Department of Energy Engineering at BME’s Faculty of Mechanical Engineering, who has published several articles related to machine learning. He used this method to forecast the production of solar panels, for example.

The research behind the current article followed a major study, conducted at the end of 2021 by myself, Bence Biró and other students specialising in nuclear energy at the Institute of Nuclear Techniques of the Faculty of Natural Sciences, published in Energy Conversion and Management: X, in which we looked at the national energy strategies of 19 European countries and used simulations for longer periods to examine whether the 19 countries, including Hungary, could achieve their climate protection and security of electricity supply goals with the power plants that are included in official government energy strategies. The results have been the topic of extensive professional discussions, and we have decided to take the relevant work of the Faculty and the Department to a new direction: in addition to using machine learning methods to estimate solar and wind power production we also wanted to apply it to calculate the variation of electricity demand over time and to estimate the frequency of so-called Dunkelflaute (dark doldrum) events using many years of data, since it has not been done in Hungary before. Doctoral student Botond Szücs joined Martin János Mayer and Bence Biró in this project. I think it is very important for the development and the future of BME that we have as many multi-faculty and multi-departmental projects as possible. This is one such endeavour, successfully combining the competences available in the two faculties.

What are the advantages of machine learning for such a study?

Machine learning is not about solving deterministic equations that directly describe physical processes (as is so often the case in engineering and science), but about using lots of data to teach a special algorithm what the output “usually is” given a set of input parameters. Then, after learning, if a new data package is sent to the neural network, the model generates the appropriate output parameters based on the learned patterns. The application of this method is still at an intensive research stage, but in many cases it yields results with reasonable accuracy faster than a deterministic model.


How well can the model take into account ongoing climate change?

This model has not been taught specifically for processes resulting from climate change. This was not of interest to us at the moment, but it could be a topic for future studies.


Articles about the study often mention your previous position. Do you think this information is relevant to the conclusions of the study?

I don’t think my previous positions are relevant to this study. I see it as being used by the press as a reference, as if to say, “dear reader, this is the expert from BME you may have heard of earlier in relation to the nuclear power plant”. I do not think it has any more relevance.

I have also spoken regularly on energy, sustainability and nuclear safety issues over the last 30 years. I am invited to give interviews, lectures and participate in debates in many places, at home and abroad. I talk a lot about issues related to energetics, and I feel that this is my mission as a university professor.


It seems that European decision-makers stake everything on renewable energy sources to achieve carbon neutral energy production. How “popular” do you think it is to point out the weaknesses of this concept?

As engineers and academics, I believe that we have a responsibility to communicate to society and to policymakers the results of analytical methods, carried out according to the standards of our profession and considered state-of-the-art even by international standards.

If political and/or economic decision-makers, or indeed society, are going in a direction that is unfeasible or dangerous in terms of a certain goal, then it is imperative to draw attention to it, even if such a move proves to be unpopular. If the doctor finds that you as a patient have a serious illness, they are obliged to tell you, even if it will make them unpopular.

I always tell my students that sustainability has three pillars: environmental, economic and social sustainability. If one pillar is damaged by the other, sustainability as a whole is also damaged. This is not something a professional should ignore, even if it may not be popular!

Figure 1: Heatmap of the probability of Dunkelflaute hours with a 5% threshold for all (8760) hours of the year, calculated from the 42 years of modeled capacity factors. The figure shows the hours per day (vertical axis) and the days of the year from 1 January to 31 December (horizontal axis), while the colours show the probability of the hourly value of the utilisation factor of solar and wind power plants below 5%, based on 42 years of data. 0 probability is indicated by the white colour and any pixel with a non-white colour represents a probability greater than zero according to the right side colour scale.


While the acceptance of nuclear energy in Hungary is growing according to a recent opinion poll, the government has decided to build new gas power plants in Hungary. Do you think they could be suitable for the stated purpose of meeting industrial energy demand and bridging the Dunkelflaute periods?

Nuclear power has long enjoyed over 50% support in Hungary, and people here and in many countries in Europe have recognised the need for a highly reliable and climate-friendly energy source instead of natural gas, the supply of which has become limited due to the war. In my view this is what has led to the increasingly positive attitude towards nuclear power in our country and elsewhere. The planned gas power plant constructions do not contradict this. As I have said many times, in many places, precisely on the basis of our major study published in late 2021: Hungary’s 2020 energy strategy set out to build too few domestic power plants. We need more power plants in the Hungarian system, besides solar power plants and additional nuclear units, because our 30% dependence on electricity imports is too high and very risky. Moreover, gas power plants can be operated flexibly, so they can play a major role in creating flexibility in the electricity system, alongside renewables and nuclear power. Of course, their operating costs will be high and the exposure of Hungarian electricity production to natural gas prices will increase as a result of these developments, but these costs and risks can be undertaken in the current situation. A power plant with a high utilisation factor and controllability can be built fastest on a natural gas base.

The full article is available to the public.

Authors of the article and members of the research team:
Martin János Mayer (BME EGR),
Bence Biró (BME NTI),
Botond Szücs (BME EGR),
Attila Aszódi (BME NTI).


BME – Budapest University of Technology and Economics
EGR – Department of Energy Engineering
NTI - Institute of Nuclear Techniques




Photo: B. Geberle