We have reviewed 11 large North American wind integration studies. This review highlights a number of areas in which wind integration studies have evolved to provide valuable insight, as well as a few areas in which improvements in methods and additional research are needed to facilitate greater insight from future studies. We now summarize our observations and conclusions from this review.
The quality of the wind data used for wind integration studies has improved substantially over time. Almost all of the wind integration studies used data from meso- scale numerical weather prediction models. Early studies used models with a 10-km resolution, whereas the most recent models used 2-km spatial and 10-minute temporal resolution. As shown in Section 17.2, mesoscale models can produce wind speed data that underestimate variability at small time scales. However, the more recent of the two datasets reviewed (the updated EWITS [EnerNex Corporation,
2011] data) showed less evidence of this reduction than the data from the WWSIS model, although the methods used in the updated EWITS data are not documented. The decision to release the data for these two studies publically is commendable, making it possible to perform a variety of validation analyses, which should motivate further improvements in the quality of these data.
Accurately translating wind speed data to statistically representative wind power output data continues to be a challenge. One of the studies (GE Energy, 2010) used a statistical technique (SCORE/SCORE-lite; Potter et al., 2008) to add additional variability to 10-minute resolution power output data to compensate for the reduced variability in the mesoscale data, but there is some evidence that this process may produce data with too much rather than too little variability (Milligan et al., 2012). Several of the studies combined high-resolution data from power plants with 10-minute simulated data in order to look at faster-time-scale phenomena, such as regulating reserves. However, it is not possible to independently verify that the data have the correct statistical properties, since most of the methods and all of the data for this hybridization process are not public.
Finally, there is a need for more research to develop methods that reliably combine high-resolution measured plant data with mesoscale model data to produce hybrid data with statistical properties similar to those of actual wind plants. Some progress has been made in this area. For example, Rose and Apt (2012) introduce a method for synthesizing high-resolution data from low-resolution data, based on the spectral properties of wind power. A significant barrier to this research is the lack of publicly available wind power production data with which to validate candidate methodologies. While it is feasible to obtain proprietary data from some generators, it is difficult to publish reproducible methods using proprietary data. Making some historical, high-resolution power plant production datasets pubic would be tremendously valuable to the electricity industry.