Forecasting renewables: where machine learning earns its keep
Wind and solar have made generation a weather problem. Probabilistic machine-learning forecasts let us price that uncertainty instead of fearing it.
As renewable penetration rises, short-term power prices are increasingly a function of the weather. A revised wind forecast can move an intraday market more than any fundamental headline.
Machine learning is well suited to this regime. Rather than predicting a single number, our models produce probability distributions over load and renewable generation — capturing not just the expected outcome but the shape of the uncertainty around it.
From forecast to position
A point forecast tells you where to aim. A distribution tells you how much to size. By trading the probability rather than the point estimate, the desk can stay disciplined when the forecast is confident and step back when it is not.
Validation before capital
No model reaches production without out-of-sample validation. We care less about how a model fits history and more about how it behaves on data it has never seen — because that is the only test the market actually runs.