Comparison of individual and combined forecasting of wind power output of two wind farms in Western Australia

Abstract

Large penetration of wind farms in the electricity grid is expected to be an unknown challenge in the near future. The biggest factor is the intermittency of wind turbine output and the difficulty in forecasting it accurately. A number of studies on wind forecasting have shown that the error of statistical methods increases with forecast horizon. However, combined forecasting of geographically dispersed wind farm output, sees a reduction in error. This paper extends this further by analysis of the confidence interval and uncertainty of forecasts when combined forecasting is used. It is observed that both the forecast error as well as uncertainty is reduced when combined forecasting is used for two wind farms in Western Australia. For a short term forecasting horizon of 6 hours, maximum forecasting error is reduced by 6% while uncertainty is reduced by up to 20%. This is expected to improve if more wind farms are considered.