Argonne is applying machine learning techniques to obtain probabilistic forecasts of wind.
As the use of renewable energy technologies grows, probabilistic forecasting of a wind power generation is needed to help ensure reliable and economic power systems operation.
Argonne is using a normalizing flow deep-learning model to learn the joint distribution between “current wind conditions” and “3-hours future wind conditions.”
We use this model to construct the forecast distribution P(“3-Hours Future” | “Current”). One distinguishing feature of this approach is that forecast distributions are much more informative than forecast point-estimates, and more useful for risk management. Most deep-learning forecasting applications yield point estimates.