Redispatch Without Black Boxes: Load Forecasting With OpenSTEF at E.ON
summary
E.ON replaces vendor black boxes with an OpenSTEF pipeline that forecasts transformer loads using XGBoost per-transformer models, weekly retraining, weather and lagged and partner-transformer features, and a flatliner postprocessing for switched-off units. A 2-year backtest on 10 transformers against an external neural-network model shows a 7% RMSE improvement on average and better short-horizon performance; prediction intervals (5th/95th) are produced and the system targets on‑prem OT deployment while preparing contributions back to OpenSTEF.
AI-written summary · about
chapters (11)
- 0:00Introduction and motivation
- 1:25Redispatch challenges and grid imbalances
- 4:37Why build in-house forecasting
- 7:08Choosing OpenSTEF framework
- 11:03Benchmark setup and data
- 14:37Model features and training
- 18:11Benchmark results and insights
- 22:04Lessons learned from benchmark
- 23:47Moving to production and open source
- 27:23Future operating model for grids
- 28:37Q&A session
related talks
43mMore Renewable Energy Into the Power Grid with Open Source
30mHow Open Source Helps a Grid Operator with the Challenges of...
42mHow Open Source...
35mLF Energy 101: How Open Source Is Powering the Digital Energy Tran...
27mOpen-Source Congestion Mana...
16mLightning Talk: Forecasted Congestion Management Using Plug and Play Model
38mOpen Energy Modeling for Mission-critical Planning Purposes
29mLFE Project PowSyBl: Grid Modeling at the Core of Europe’s Power Grid Planification...