Wind Forecasting Quality Metrics for Risk Management

Wind farm operators require wind speed and wind energy power forecast for several purposes:  

1. Trading electricity

2. Program submission to TSO

3. O&M planning

4. Hazardous events prevention 

All these activities necessarily involve strategies on risks management derived from probabilistic forecast outputs, constrained by the current state of the art on Numerical Weather Prediction (NWP) models and statistical downscaling algorithms. Therefore, advanced comparison techniques are required to evaluate the value that a certain forecasting may provide; these techniques usually compare several aspects, such as:

 1) Accuracy, the difference between the mean expected value P50 and the real power

2) Precision, the uncertainty band associated with its corresponding expected mean

3) Granularity, the final time resolution of the forecast provided

4) Reliability, robustness of forecast supply that ensures always its delivery before the market closing

5) Refreshing capability, ability to update the forecasting models before the target hour

These parameters significantly supports the management of risks at wind plants operation through appropriate simple metrics that monitor not just the mean expected forecast but also its probability  distribution, i.e. Normalized Mean Absolute Error (NMAE) for the accuracy of the P50 value or Calibration for the representativeness of the distribution percentiles.

In this context, the proposed work contains comparative examples of several state of the art predictors at several wind power plants through the application of the above-mentioned metrics focused on risks management.

EDPR contributes to the IEC Task 36 on Power Forecast by sharing experimental data and identifying the key points to improve the accuracy and precision of the NWP models and statistical downscaling methods through a transversal benchmark activity over the most critical wind farms.