Are Policies to Encourage Wind Energy Predicated on a Misleading Statistic?

In response to the very real challenge of climate change, the share of electricity generation from wind turbines is expected to increase substantially over the next few decades. For example, in October 2008, the California Public Utilities Commission and the California Energy Commission recommended a 33 % renewable energy requirement as a key strategy to reduce greenhouse gases. The plan, if implemented, would increase the wind energy capacity available to the California Independent System Operator (ISO) by almost 500 percent by 2020 with wind energy capacity accounting for approximately 18 percent of installed nameplate capacity (Hawkins, 2008). According to the American Wind Energy Association, renewable energy policies currently exist in 28 U.S. states with wind energy being a significant beneficiary. Growth in wind energy production is also enhanced by an income tax credit of 2.1 cents per kilowatt-hour of wind energy production. Legislation is also being considered in the United States Congress that could increase the share of electricity from its 2009 share of total electricity generation of about 1.8 percent to 20 percent. The European Union has a goal of 20 percent renewable energy by 2020 with wind energy serving as a key source of the increase.

These proposed increases in wind energy are implicitly predicated on the belief that wind energy, while not capable of “upward dispatch”, is fairly predictable. But is it? According to most analyses, the answer clearly is yes. Cali, et. al. (2006) have indicated that the root-mean-squared-errors (RMSE) of the day-ahead wind energy forecasts at one of the German TSOs (unidentified) have declined from approximately 10 percent of installed wind capacity in 2001 to approximately six percent of installed wind energy capacity in 2006. This finding has been cited by the European Wind Energy Association (2007) as evidence that wind power is a reliable source of electricity supply. Lang (2006) presented a forecasting approach that had a weighted mean forecast error of approximately six percent for Ireland. It was noted that the forecasting method offered the system operator the ability to accommodate higher wind penetration levels. The Midwest ISO has reported that the mean absolute percentage error of its day-ahead wind forecasts is approximately 7.06 percent of its installed wind energy capacity. In a report commissioned by the National Renewable Energy Laboratory , Porter and Rogers (2010, p. 5) have reported the mean absolute wind forecasting errors in ERCOT, the system operator that serves the vast proportion of Texas, ranged from 8.28% to 10.73% of capacity for all hours over the period May 2009-August 2009.

This paper explores whether it is appropriate to rely on and report wind forecasting errors weighted by wind energy capacity. The starting point is that load forecasting errors are never weighted by the capacity of the equipment that consumes electricity. If they were, then load forecasting errors would appear to be trivial when in fact they are not. The paper then points out that a wind forecast error weighted by capacity can decline over time, creating the impression that forecast accuracy is improving, even when the unweighted error is not declining.

The paper examines the accuracy of wind forecasts using data from Western Denmark, Eastern Denmark, the 50Hertz system in Germany (formerly known as Vattenfall), the Transpower system in Germany (formerly known as E.ON Netz), the Amprion system in Germany (formerly known as RWE), the Italian power grid, the power grid in the Republic of Ireland, the Midwest ISO, and ERCOT, the power system that serves the vast proportion of Texas. Our analysis reveals that the forecast errors are significantly larger than those implied by the capacity-weighted errors and are also significantly larger than the errors in load forecasting. Econometric analysis provides evidence that these forecasting errors have implications for power system operations. The analysis also indicates that there is a nontrivial systematic component to the errors and that the modeling of this component may produce more accurate forecasts. The paper's results are expected to be of interest to wind energy forecasters, system operators, policy makers, and individuals interested in wind energy.