Speaker
Description
We present evaluations for the prediction of wind power ramping events in the Belgian Offshore Zone. We verify two models from Royal Meteorological Institute of Belgium: the operational ALARO-4km and its version with Wind Farm Parameterization (WFP). Power predictions are generated with power curves and machine learning (ML). As standard metrics like MAE are insufficient for evaluating ramps, our proposed framework incorporates time and power buffers for a flexible assessment by tolerating minor errors. Results show that WFP models improve ramping prediction skill, with ML achieving more balanced forecasts between reducing misses and false alarms. We also introduce a Ramp Alignment Score to quantify temporal errors by forecast lead time and confirm the averaged smaller timing errors of WFP models. Finally, the framework helps to understand that severe precipitation is a strong indicator of large, predictable ramps, while lighter precipitation is associated with more forecast errors.
