The ability to read the future is priceless in Czech energy. AI can improve it

No complex civilisation can do without the energy that powers it. This is truer than ever in our present, built on all kinds of technologies. Almost nothing would work without electricity. Reliable and stable supply requires not only sufficient resources but also balance in the transmission and distribution grid. Whatever is generated must be consumed. One of the key tasks of a transmission system operator is therefore to manage the system imbalance, or ensure that generation and consumption are balanced. To do this effectively, it must be able to forecast the imbalance as accurately as possible.

This task is becoming more complicated, however. As part of efforts to reduce greenhouse gas emissions and mitigate climate change, large and stable sources such as nuclear, coal and gas-fired power plants are increasingly being replaced by numerous smaller renewable sources. Solar and wind power can generate electricity without emissions, but they cannot be controlled to produce energy when it is in demand. In the European Union, the share of renewable sources in gross electricity consumption rose from just under 16 percent in 2004 to 41 percent in 2022. In Czechia, the share was only 15.5 percent the year before last, but according to the climate and energy plan approved by the government, it is to double by 2030. Maintaining the stability of the energy grid will therefore become more challenging once again.

A range of tools to choose from

High-quality forecasts can make this task easier for transmission system operators. They can also reduce the costs of balancing mismatches between generation and consumption. As the share of renewable sources grows, expenditure on so-called balancing capacity is rising significantly, and consumers pay for it in their bills. Europe is already struggling with high energy prices, so every efficiency gain counts.

Information technology is increasingly able to help with forecasting. More powerful computers are expanding our capabilities, as are the advent of artificial intelligence and machine learning. This has made a range of methods available, including random forests and neural networks in addition to conventional regression models. At Unicorn, in collaboration with Unicorn University, we examined their potential for forecasting system imbalances in the electricity grid. We found that using artificial intelligence can bring interesting improvements. During the period studied, our model’s forecasting ability was about 14 percent better than that of older solutions. The forecast covered the next three quarter-hours, a period based on the new 15-minute trading interval. As expected, the results were best for the first quarter-hour, with the error rate increasing over a longer horizon. Accuracy was not the only parameter we considered; we also looked at the robustness of the models. In this respect, neural networks proved problematic, as they failed more often over a longer period.

Regular training is essential

Machine-learning models have the advantage of being able to improve continuously. However, they need to be retrained so that they can continually account for the fact that relationships between input parameters may change over time. To give just one example, models used by brokerage firms to forecast share-price movements are essentially retrained every day. In our examination of methods to improve system-imbalance forecasts, we drew some inspiration from the financial world.

We must not overlook the crucial importance of input data quality, either. Our testing clearly showed that more data would help improve the results, particularly data with finer geographical granularity and a shorter time resolution. Information about the weather, which affects electricity consumption as well as the output of renewable sources, is especially important. More detailed data on solar irradiation, temperature and wind than is available at the regional level would make it possible to better account for local differences and improve forecasts of generation from renewables distributed across different regions. A longer time series of actual quarter-hourly system imbalances would also help, as would more accurate data on actual photovoltaic generation than the hourly figure for the whole Czech Republic. Fortunately, improving the data foundation should not be an insurmountable task.

The energy sector is undergoing a major transformation, and the entire system is becoming increasingly complex. The most advanced computational methods incorporating artificial intelligence are not a magic solution to the problems this creates. But they can significantly improve forecasts of system imbalances in the energy grid. The practical application of these methods in this area is still at an early stage. At Unicorn, however, we believe in this approach and can guide interested energy-sector companies through every step, from a case study at the academic research stage, through data preparation and integration, to model development, training and maintenance in live operation. We are well placed to do so thanks to our long-standing software development experience and in-depth knowledge of the energy sector, which we help shape across Europe.

About the authors

Jan Korych

Jan Korych works at Unicorn Systems a.s. as a consultant in the analysis and design of information systems for the energy and utilities sector. He specialises in projects for European transmission system operators in the field of balancing, and has also contributed to the delivery of IT solutions for electricity and heat traders and generators. During his career, he has delivered several IT solutions in close collaboration with clients, focusing primarily on meeting their requirements and maximising added value for them.

Karel Šafr

Karel Šafr is a lecturer and instructor at Unicorn University, specialising in artificial intelligence (AI). He teaches Introduction to Machine Learning, Advanced Machine Learning and Probability. His research focuses on data processing and mathematical-statistical models. Karel recently joined the Unicorn Research Centre as a senior researcher, focusing on research and innovation in artificial intelligence. He is actively involved in research projects that push the boundaries of knowledge in AI across various fields.

About Unicorn

Unicorn is a renowned European company providing major information systems and IT solutions. We have long focused on delivering high added value and a competitive advantage to our customers. We have been active in the market since 1990, and over that time have created a range of leading-edge, large-scale solutions that are widely deployed and used by major businesses across different industries. We are also developing Unicorn Universe - Digital Twin Construction Kit. This product is used by the internet service Plus4U to provide a broad portfolio of services based on robust software solutions for small and medium-sized businesses and end users.

In the energy and utilities sector, we focus on implementing large-scale information systems for managing energy markets at national and international levels, as well as related energy trading. Another priority is geoinformatics and solutions for enterprise asset management. We use both our own products and proven platforms from our partners.