EU-AI Act Compliant Load Forecasting Beats Baseline

Short-term load forecasting (STLF) is critical for the electric power industry, especially for infrastructure designated as safety-critical under European and German law. This means that beyond mere accuracy, STLF solutions must meet stringent engineering requirements for determinism, reproducibility, and auditability, aligning with the EU-AI Act. A recent 41-day live challenge evaluated a complete STLF pipeline designed with these compliance needs in mind. The pipeline, built on the open-source Python library `spotforecast2-safe`, forecasts 24 hourly load values for the aggregated German transmission-grid using ENTSO-E data. It incorporates anomaly detection, data preparation, various covariates, a recursive forecasting algorithm, and hyperparameter tuning. The results showed that this EU-AI Act compliant pipeline significantly outperformed the official ENTSO-E day-ahead forecast baseline. Notably, transparent and auditable local models within the pipeline proved competitive with much larger, energy-intensive foundation models, highlighting that compliance and efficiency can go hand-in-hand with strong performance in safety-critical AI applications.