Fraunhofer IPA: AI-supported price forecasting increases revenue by up to 37 percent

On the one hand, fluctuating electricity generation from wind and sun poses challenges for grid operators. If there is an imbalance or bottleneck in the power grid, for example if there is temporarily less electricity available from photovoltaic systems than originally planned, urgent measures must be taken to relieve the grid. Otherwise, there is a risk of a blackout like the one in Spain and Portugal at the end of April 2025. The so-called balancing energy market plays an important role here, on which short-term increases or reductions in the quantities of electricity demanded can be traded and called up as required.

On the other hand, it is precisely these uncertainties that open up new opportunities for industrial companies. Because those who can flexibly adjust their electricity requirements can earn money on the balancing energy market. Researchers from the Fraunhofer Institute for Manufacturing Engineering and Automation IPA have now presented a new type of forecasting method in a scientific paper that uses machine learning to predict the bid prices for balancing power more reliably.

“Balancing power is traded on a so-called ‘pay-as-bid’ market,” explains Professor Alexander Sauer, Director of Fraunhofer IPA. “There is a bidding process in which each provider is paid the price at which they submitted their bid. So anyone who significantly undercuts the actual electricity price is forgoing money. And worse still, anyone whose bid is higher than this will go away empty-handed. Companies can therefore significantly increase their revenues with our newly developed forecasting method.”

More revenue, more grid stability

Many industrial companies have so far relied on simple, static bidding strategies. They set their bid once and then stick to it. Or they base their bid on the price of the previous day or the previous week. By using various machine learning methods, the scientists at Fraunhofer IPA have now succeeded in better predicting this price.

In a second step, they have supplemented their AI-supported price prediction with a specially developed offset method. “To a certain extent, this is the post-processing of the predicted electricity price so that the submitted bid is slightly lower,” explains Vincent Bezold from the Data-Driven Energy System Optimization research team at Fraunhofer IPA. “This has to do with the rules of the game on the ‘pay-as-bid’ market. If your bid is too high, you lose out. That’s why it pays to specifically undercut the actual electricity price – and that’s exactly what we achieve with our offset method.”

With this optimized bidding method, revenues can be increased by up to 37 percent compared to other strategies, because the specified bid is less often higher than the actual price and is therefore awarded more often. In their paper, the scientists examined four sub-markets of German balancing energy and showed that a forecasting error that is one euro less per megawatt hour can – depending on the market – generate up to 3631 euros in additional annual revenue per megawatt. “The data-driven optimization of the bidding strategy is therefore worthwhile and also contributes to the stability of the electricity grid,” summarizes Lukas Baur from the Data-Driven Energy System Optimization research team at Fraunhofer IPA.

Future even higher forecast quality

In principle, the AI-supported forecasting method can be transferred to other, similarly structured markets – such as securities trading – even beyond Germany’s borders. In future, the researchers at Fraunhofer IPA want to use even more complex AI models and also take into account external factors such as weather data and probability-based forecasts in order to further increase the forecast quality.

– – – – – –

Further links

👉 www.ipa.fraunhofer.de  

Photo: Fraunhofer IPA