Forecasting & Wind Upscale
The Forecasting product aims to channel many different types of forecasts from internal and external sources and make them available in the user interface as well as to other modules, products and planning processes.To support forecast experts in ensuring a consistent quality, forecasts are processed in groups and validated with consistent rule sets before publishing them to other modules and products. All individual forecasts are compared to observations and the resulting quality indicators stored and visualised in a quality and error monitor. Currently, the Forecasting product processes the following forecasts at substation level:
- Weather forecasts and observations from external data providers
- Wind generation forecasts from external data providers
In parallel, we are also building 50Hertz-owned forecasts. We are working on implementing an interface that accepts forecasting models in an MLflow format and deploys them for inference to MCCS. MLflow is an open standard for exchange between machine learning platforms.
Wind Upscale as a starting point for wind generation forecasts
Wind Upscale shows grid operators in real time how much power is generated by all wind turbines installed in the 50Hertz grid area. To generate this information, Wind Upscale processes several hundred real-time measurements from wind farms that are supplied by the connected DSOs.
This is done in three steps:
- Validate measurements in real time
- Integrate all valid measurements into forecasts
- Execute regional aggregation (e.g. substation level)
Every measurement is checked for
- validity flags;
- completeness of values for at least 24 hours;
- non-zero values;
- correct value bandwidth.
Wind Upscale will support our forecast experts in quickly adding new generation units and new real-time measurements, ensuring an up-to-date algorithm that reflects the addition of more and more wind farms with little delay.