Although today microgrids serve a tiny fraction of the market, that share will grow as costs fall. Utilities can benefit if they plan ahead.
The changing architecture of demand response in America.
rates are too low to produce significant impacts. In 15 states, dynamic pricing is estimated to have an impact of 1 to 2 percent. And in one it has an impact of 3 percent.
But the potential in the future is far greater. It ranges from 3 to 16 percent. Figure 1 displays impacts from dynamic pricing by itself and, where it was found to be cost effective, it also displays impacts from dynamic pricing coupled with enabling technology.
The states with the highest potential are shown in Figure 2 . These states tend to have hot summers, high saturation of central air conditioning systems ( i.e., energy consumption can be readily controlled by moving the thermostat either manually by the customer or automatically by the utility) and high capacity costs ( i.e., enables high price signals to be sent out during critical periods).
All of this emphasis on dynamic pricing at the retail and wholesale levels raises the question: How much demand response can we expect from it? The FERC report provides an answer: The impact could range between 14 and 20 percent of peak demand, or 138,000 to 188,000 MW, depending on whether dynamic pricing was deployed on an opt-out basis or on a universal basis. If it was deployed on an opt-in basis, the impact would be much smaller.
It’s important to note that the residential class dominates the FERC results on dynamic pricing. This class hasn’t been the focus of utility dynamic pricing programs.
The FERC projections of dynamic pricing rest on an empirical foundation provided by several dynamic pricing pilots. 13 Since that report was published, results have become available from the Connecticut and Maryland pilots, and an interim report also has been released from the PowerCents DC pilot in the District of Columbia.
Collectively, these pilots embody 67 different tests of dynamic pricing rates (see Figure 3) . The results show a huge amount of variation in affecting load, with the low end being under 5 percent and the high end being greater than 50 percent. Such order-of-magnitude variation stifles policy analysis. Thus, some analytical filters must be applied to the data to improve the resolution of the impacts.
The first filter is simply to group results by pilot. This isolates the geographical variation across pilots and isolates the effects of pilot design and provides some additional resolution (see Figure 4) . But a fair bit of intra-pilot variation still remains.
The second filter is to group the results by rate type. This improves the resolution considerably (see Figure 5) . Critical-peak pricing (CPP) rates vastly exceed the effectiveness of simpler time-of-use (TOU) rates, in large measure because prices during the critical-peak hours of the CPP rate are much higher than those during the normal peak hours in a TOU rate. However, within the family of CPP rates, there remains a huge amount of variation.
The third filter separates those tests that included enabling technology from those that didn’t. The hypothesis is that enabling technology will boost impacts, and this is borne out empirically