2013

Holttinen, Hannele; Miettinen, Jari J.; Sillanpää, Samuli

Wind power cannot be dispatched so the production levels need to be forecasted for electricity market trading. Lower prediction errors mean lower regulation balancing costs, since relatively less energy needs to go through balance settlement. From the power system operator point of view, wind power forecast errors will impact the system net imbalances when the share of wind power increases, and more accurate forecasts mean less regulating capacity will be activated from the real time Regulating Power Market. In this publication short term forecasting of wind power is studied mainly from a wind power producer point of view. The forecast errors and imbalance costs from the day-ahead Nordic electricity markets are calculated based on real data from distributed wind power plants. Improvements to forecasting accuracy are presented using several wind forecast providers, and measures for uncertainty of the forecast are presented. Aggregation of sites lowers relative share of prediction errors considerably, up to 60%. The balancing costs were also reduced up to 60%, from 3 /MWh for one site to 1-1.4 /MWh to aggregate 24 sites. Pooling wind power production for balance settlement will be very beneficial, and larger producers who can have sites from larger geographical area will benefit in lower imbalance costs. The aggregation benefits were already significant for smaller areas, resulting in 30-40% decrease in forecast errors and 13-36% decrease in unit balancing costs, depending on the year. The resulting costs are strongly dependent on Regulating Market prices that determine the prices for the imbalances. Similar level of forecast errors resulted in 40% higher imbalance costs for 2012 compared with 2011. Combining wind forecasts from different Numerical Weather Prediction providers was studied with different combination methods for 6 sites. Averaging different providers' forecasts will lower the forecast errors by 6% for day-ahead purposes. When combining forecasts for short horizons like the following hour, more advanced combining techniques than simple average, such as Kalmar filtering or recursive least squares provided better results. Two different uncertainty quantification methods, based on empirical cumulative density function and kernel densities, were analysed for 3 sites. Aggregation of wind power production will not only decrease relative prediction errors, but also decreases the variation and uncertainty of prediction errors.