Verification of wind forecasts with special attention to ramping events

The portion of energy generated by wind in the United States has been steadily increasing. As this proportion increases, it becomes increasingly important to have accurate forecasts of wind speeds, or at least forecasts that accurately describe their confidence intervals. Ramping events are sudden changes in wind speeds or power generation which can be very costly if not predicted in advance. Aside from issues resulting from discrepancies in wind speed, errors in the timing of ramping events can also have critical impacts. Since ramping events represent occur over a very small portion of time, their impact can be easily hidden in statistics summarizing forecast performance over extended periods of time.

This paper suggests metrics useful for evaluating wind forecasts in a manner that is useful to both meteorologists and the wind energy industry. Wind ramps are identified through the analysis of first- and second-order differences. Ramps of observed events are first matched to the nearest available forecasts. With matched pairs of forecast and observed events, attributes such as magnitude, duration, peak time and initiation time can be compared. With these comparisons, it is important to consider the potential magnitude of the improvement due to changes in the timing of forecasts. Some changes results in small improvements while others result in more significant changes. Strictly looking at timing errors can obscure this.

These types of metrics are illustrated using wind forecasts created for Xcel Energy for wind farms located across their service area.