Utilities are leaving no stone unturned in their search for ways to save electricity. Federal incentives will support new technologies and projects, but can those incentives overcome structural...
Planning for Efficiency
Forecasting the geographic distribution of demand reductions. Copyright © 2011 Consolidated Edison Company of New York, Inc.
(L) for converting energy (MWh) reductions to peak demand (MW) reductions.
Taking the dot product of each row in S with the corresponding row in L produced the expected peak DSM demand reductions by network.
The last step in the process recognizes the fact that the DSM calculations so far are expectation values. That is, even if top level energy savings goals are achieved and our load curves are accurate, the inherent geographic variability in the penetration of programs means we should only expect to achieve or exceed the calculated values 50 percent of the time (P50) in any given network. However, maintaining system reliability requires that we forecast a DSM reduction we are much more than 50 percent confident will be achieved for each network ( e.g., P90 or P95).
The solution to this is to move a few standard deviations below the expectation value to calculate a P95 DSM reduction. The difficulty here is that these programs are new and there’s no data from which to estimate standard deviations. As a conservative approach, Con Edison applied a 50 percent allocation uncertainty (reduction) to the DSM expectation values to approximate two standard deviations. 4 With experience, it might become possible to measure the variability in geographic penetration of programs and therefore better estimate the standard deviations.
One final note is that this reduction for allocation uncertainty isn’t applied to the system peak forecast, because Con Edison assumes that the top-line forecasted energy savings will indeed be realized. For the system peak, then, the program energy savings are simply converted to demand reductions using the load curves we derived for each program. In this case, we sample the load curve at 4 p.m. on a summer weekday, which is the time of the system peak. No geographic uncertainty reduction is applied.
Allocating and Aggregating
The most important question about this approach is whether energy efficiency gains will mirror consumption patterns within each market segment. We believe this is a very reasonable assumption for sufficiently large aggregations of demand over the long term. 5 However, short-term achievements might follow very different patterns. For instance, we’ve already noticed that implementation contractors, particularly in the small commercial segment, have focused outside Manhattan, where they generally find both a lower cost of doing business and, probably, less-efficient existing equipment.
An alternate approach—for established efficiency programs—might be to quantify the geographic distribution of past achievements and use this as an allocating function. The danger here is that distributions likely will shift as some areas saturate, skewing longer term projections. In any event, this approach won’t help allocate new DSM programs where historical data are unavailable.
A key feature of this approach is the choice to convert energy to demand after—rather than before—allocating the savings to the networks. We think this is the right approach, because the coincidence of most DSM measures is highly dependent on time of day, and our networks peak at very different times. For instance, converting the energy savings of a residential lighting program to demand and then allocating it likely would