Accuracy of Wind Energy Forecasts in Ireland and Prospects for Improvement
Author: Kevin F. Forbes, Ph.D.
Abstract
In 2020, the latest year for which official data are available, wind energy accounted for about 38% of Ireland's electricity load. Under Ireland's climate action plan, at least 3.5 GW of offshore renewable energy (mainly offshore wind) and an increase in onshore wind capacity of up to 8.2 GW is targeted for development by 2030. In light of these plans, it is significant that the errors in the wind energy forecasts in Ireland are a likely contributor to the power grid's operational uncertainty. On an energy weighted basis, the wind energy forecast errors are much larger than the load forecasts' errors. While the errors corresponding to an optimal forecast would have the property of "white noise," the errors in Ireland's wind energy forecasts have the property of being statistically related to forecasted weather conditions as well as the forecasted level of wind energy. The errors also have the property of being autocorrelated, a well-established indicator of less than optimal forecasts.
Based on the attributes of the forecast errors, a time-series model is formulated. The model's structural regressors include the level of forecasted wind energy, modeled meteorological conditions, proxies for expected wind energy curtailments, and binary variables for the season of the year and hour of the day. The model was estimated over the period 2 January 2015 through 31 December 2018 using 15-minute data. The model has an R-squared equivalence of about 0.997. The model is evaluated using out-of-sample data over the period 1 January 2019 - 2 June 2020. The 15 minute-ahead out-of-sample predictions have a weighted-mean-absolute-percentage-error (WMAPE) of about 4.8%, less than half the WMAPE associated with the wind energy forecasts reported by the system operator over the same period.
Key Points
There is a high level of operational uncertainty in the Irish power grid as measured by the volatility in the ex-ante measure of the energy imbalance. There is also a high level of ex-post operational uncertainty in the Irish power grid as measured by the volatility in the balancing prices. Forecasting errors are possible contributors to this uncertainty.
The weighted mean absolute percentage error (WMAPE) in the wind energy predictions over the period 2 January 2015 – 31 December 2018 was about 13.57 %, far larger than the error in the load forecasts.
A time-series econometric method to improve the short-run forecast accuracy of the wind energy predictions is presented.
Out-of-sample predictions were calculated that could be made available to a system operator one full period before real-time, i.e., at the end of period t-2. In this case, the out-of-sample WMAPE is 4.8%. Over the same period, the forecasts posted by the system operator have a WMAPE of 16.43%.