End User
PPC
Description
The Wind Power Short-Term Forecasting service delivers probabilistic day-ahead and intra-day wind power forecasts at turbine and farm level using LightGBM ensembles trained on multi-model NWP inputs and SCADA telemetry. The service addresses three interconnected challenges: NWP model selection and weighting, non-linear power curve mapping, and SCADA integration for curtailment and maintenance state accounting. Forecasts are generated in day-ahead (13-40h), extended (40-84h), and intraday rolling (0-12h) modes.
Core Capabilities
Predictive & Prescriptive Analytics
Business Need
PPC needs reliable day-ahead wind generation forecasts to reduce imbalance exposure in Greek energy markets, optimise flexibility dispatch, and support grid stability across its wind park portfolio. The cost of imbalance — when actual generation deviates from scheduled positions — represents a significant operational cost that can be substantially reduced by improved short-term forecasting accuracy.
Key Performance Indicators
NMAE below 12% of installed capacity
RMSE below 12% of rated capacity
Skill score improvement of ≥20% over persistence baseline for horizons beyond H+3
Prediction interval empirical coverage between 78% and 82%
Data Provided
Per-turbine and farm-level forecast packages: P10/P50/P90 power and energy at hourly resolution
NWP input summary identifying contributing models per forecast
Traceability metadata: forecast ID, model version, training cutoff, NWP retrieval timestamp
Inputs: SCADA telemetry (wind speed, RPM, pitch angle, power, temperatures) + 5 NWP models
TEF
TEF RES