Presseinformation 051/2026 (German only)

Researchers from two working groups at the Institute for Automation and Applied Informatics (IAI) of KIT have developed a new method called “SHAPformer” that allows AI-based predictions for energy systems to be analyzed much more transparently.

In the study, which was funded by the Helmholtz Association within Helmholtz AI, tenure-track professor Benjamin Schäfer, head of the Helmholtz Young Investigator Group Data-Driven Analysis of Complex Systems (DRACOS), presents the new method. The researchers have developed it for time series forecasts—i.e. for forecasts based on successive periods of data, such as electricity consumption or electricity prices. The target is to make AI-supported predictions as precise and comprehensible as possible.

The method combines transformer models—known from modern language models—with Explainable AI methods. The name “SHAPformer” refers to the combination of transformer models with SHAP methods. These are based on game theory concepts and reveal the influence of individual factors on a forecast. These include temperatures, public holidays, wind forecasts and previous consumption data.

Doctoral candidate Matthias Hertel, first author of the study, presents as a special feature of the approach that the explainability is directly integrated into the training process, thus maintaining the accuracy of the predictions while increasing the efficiency of the analysis.

Original publication
Matthias Hertel, Sebastian Pütz, Ralf Mikut, Veit Hagenmeyer & Benjamin Schäfer: Explainable time-series forecasting with sampling-free SHAP for Transformers. Nature Communications, 2026. DOI 10.1038/s41467-026-73243-5

Presseinformation 051/2026 (German only)