Wind Energy Forecasting Using Time Lagged Ensembles

Over the last decade wind energy is the fastest growing energy source worldwide. Improvements in wind turbine technology together with rising costs of conventional energy sources have made wind energy a competitive alternative electrical energy source, with costs now less than 5 cents per kilowatt-hour. As more commercial wind farms come on line the need for high quality near surface wind speed forecasts becomes more critical.

Past investigations (FSL/NREL report, 2001), using the Rapid Update Cycle (RUC) numerical prediction model, have demonstrated that time-lagged ensemble forecasts can reduce forecast errors, when compared to deterministic forecasts. Equally weighted ensembles of forecasts, with common valid times, were used in these studies. While these equally weighted ensembles produced better overall forecasts it is reasonable to assume that differential weighting of forecasts with varying forecast durations could further improve the ensemble forecasts.

This presentation examines advantages and problems associated with several techniques used to derive weighting coefficients for members of a time-lagged ensemble, based on forecast duration. Techniques using linear least squares regression are examined, and issues associated with derivation of weighting coefficients are presented. A second approach, using a synthetic moving average/exponential decay method is explored and refined to address several problems inherent in least squares methodology. Results from several tall towers are presented.