Explainable AI for Energy Systems

Artificial intelligence (AI) is increasingly being used to ensure the stable operation of power grids, predict electricity prices, and manage the energy supply more efficiently. For AI to be used reliably in critical infrastructure such as the energy system, its decisions must remain transparent.
Researchers from the Helmholtz Young Investigator Group DRACOS (Data-Driven Analysis of Complex Systems) at the Karlsruhe Institute of Technology (KIT) have developed a new method called “SHAPformer” for this purpose. The group is led by tenure-track professor Benjamin Schäfer. The study was published on May 27, 2026, in the journal Nature Communications.
SHAPformer combines Transformer models with methods of explainable artificial intelligence (Explainable AI). This makes it possible to visualize the influence that individual factors—such as weather data, holidays, wind forecasts, or historical consumption data—have on predictions. A distinctive feature of this approach is that explainability is directly integrated into the training process. This ensures that predictions remain accurate while making the analysis more efficient. The researchers view this as an important foundation for the future use of transparent AI systems in the energy sector.
“When training our model, we deliberately masked specific pieces of information,” explains Matthias Hertel, a scientist, scientific assistant, member of scientific staff, research assistant, researcher at KIT’s Institute for Automation and Applied Informatics (IAI) and lead author of the study. This makes it possible to trace the contribution of individual inputs to the predictions. The research team tested the approach using, among other things, real-world data from the transmission system operator TransnetBW to predict electricity consumption and prices for periods of up to one week. In the long term, the method is intended to help make AI systems in the energy sector more transparent and widely accepted—for example, in smart charging systems for electric cars or home energy storage systems.