OpenSTEF

by LF Energy

AI Application

Low Confidence

OpenSTEF is an open-source Python package with end-to-end machine-learning pipelines for short-term energy forecasting, producing probabilistic forecasts hours to days ahead for consumption, generation or mixed load. Alliander built it to anticipate congestion in its distribution grid and then contributed it to LF Energy for neutral governance. At Alliander it runs as the short-term forecasting module of its 'Grid as a Service' congestion-management system, delivering operational forecasts at hundreds of grid locations (LF Energy's project page now says thousands). Forecasts feed grid safety analyses, DSO-to-TSO interface communication and congestion actions such as agreeing with renewable producers to lower output during peaks. Grid operators consume the forecasts through an API and an expert user interface; when a model cannot produce a forecast, labelled fallback forecasts are issued so it can be reconstructed which forecast a decision was based on. The stack is containerised and works with any scikit-learn compatible model, with tree-based, gradient-boosted and foundation time-series models supported.