Almost all of the studies in this review used net load step-change statistics to estimate the need for additional balancing (regulation and load-following) reserves. Most of the studies implicitly or explicitly assumed that load and wind are uncorrelated and that the data fit Gaussian statistical models, neither of which is accurate. Methods such as the one proposed in Charles River Associates (2010) that use the magnitude of low-probability ramping events rather than standard deviations are likely to produce balancing resource estimates that more accurately predict what will be needed to maintain system reliability. An even more useful improvement would be to build on the methods in KEMA (2010), which used a dynamic power system model to simulate the effect of different amounts and types of balancing resources.
High-penetration wind scenarios, as they move from concept to deployment, are very likely to motivate substantial changes in the ways that ancillary resources are scheduled, purchased, and dispatched. Operational changes, such as the large – scale use of pumped hydro in high-renewable penetration countries such as Portugal, Germany, and Ireland, illustrate the types of changes that are likely to be needed. Many of the studies in this review explicitly assumed that the structure of the energy and ancillary services markets will be largely unchanged, with the exception of more frequent dispatch intervals and balancing area aggregation. In future studies, it may be valuable to think more broadly about and model more explicitly the ways in which different types of balancing services can be purchased from different types of power plants. For example, storage and demand response could provide highly responsive balancing services but will be deployed only if electricity markets reward power plants for their responsiveness. Modeling studies of fast-ramping resources, such as in KEMA (2010), pushed the state of the art, making it increasingly feasible to quantify the benefits of responsive resources.
A growing number of research papers estimate balancing reserve requirements from historical data. For example, Mauch and colleagues (2013a) estimated total balancing reserve requirements in Texas and the U. S. Midwest and estimated that 0.07 to 0.30 MW of day-ahead reserves are needed per MW of wind. These reserve requirements are based on day-ahead load and wind forecast uncertainty. Forecast uncertainty is greatest on days when the wind is forecast to be blowing strongly (Mauch et al., 2013b); thus, there are some days when significant reserves are required and other days when much smaller reserves are needed. Because of the additional uncertainty associated with estimating the reserves that are needed to cover 95% of the day-ahead forecast errors, the reserves estimates in Mauch and colleagues (2013a) are substantially higher than those reported in many of the studies reviewed here and in Figures 17.7. Mauch and colleagues (2013a) suggest that a dynamic method should be used to schedule reserves based on the day-ahead forecast values. Additional research is needed into methods that can dynamically predict the reserve requirements for high-wind scenarios. For example, one could imagine scheduling reserves at a level of half the uncertainty in the day-ahead forecast and rescheduling after an intraday unit commitment when uncertainties are lower.
Another area for which improvements in methods are needed is in the modeling ofwind forecast errors. Some early work (e. g., Doherty & O’Malley, 2005) assumed that wind and load are uncorrelated Gaussian random variables, which, as shown in Section 17.2, is not supported by the data. More recent research in this area (e. g., Hodge et al., 2012; Mauch et al., 2013b) points out the heavy-tailed nature of the distributions. Appropriate use of these results should allow for increasing statistical accuracy and thus more insightful results in future integration studies. This topic is also discussed in Chapter 9 .