The accuracy of wind energy forecasts and prospects for improvement

Abstract

Wind energy forecast errors, while modest when weighted by wind energy capacity, are quite large relative to the average level of actual wind energy generation. For example, while the capacity weighted root mean squared error (CWRMSE) of day-ahead wind energy forecasts for the 50Hertz control area in Germany over the period 1 January 2015 through 31 December 2015 is just 4.5 percent, the energy-weighted root-mean-squared-error (EWRMSE) is almost five times as large at 21.67 percent. Our analysis also indicates that the errors in 50Hertz's wind energy forecasts are statistically related to forecasted weather conditions. Based on this finding and the time-series attributes of the forecast errors, an ARCH/ARMAX model was formulated to predict wind energy generation. The model's forecasting accuracy was evaluated using out-of-sample data over the period 1 January 2015 through 31 December 2015. The out-of-sample period-ahead predictions have a EWRMSE of about 2.93 percent and CWRMSE of about 0.60 percent.