Learning lessons from PSE’s residential demand response pilot.
Peter Steele-Mosey is a senior consultant at Navigant Consulting’s Toronto office. His principal area of interest is econometric program evaluation. Email: email@example.com.
Residential direct load control programs intended to reduce peak residential demand on hot summer afternoons have become increasingly common, particularly in more innovative jurisdictions such as Arizona and California. In the last two years, dozens of evaluations of air-conditioning direct load control pilots and programs have been published, and as a result, a fairly solid understanding of potential demand reductions exists. Likewise, the effects and persistence of so-called “snapback”—the increase in demand above normal levels immediately following a curtailment event—are also well understood.
Meanwhile, utilities in winter-peaking areas have received far less attention. Although a small number of studies examining the impact of water-heater direct load control during the winter do exist, robust empirical studies examining the impacts of the direct load control of space heating in winter aren’t nearly as common. Thus, Navigant’s recent evaluation of Puget Sound Energy’s (PSE) residential demand response (DR) pilot and survey data reported by Navigant contractor Energy Market Innovations should be of particular interest to winter-peaking utilities.
Direct Load Control at PSE
The residential DR pilot was a direct load control program running from October 2009 through September 2011, during which time participants’ water heating or space heating—or both—were cycled down on seven winter mornings and two winter afternoons. Afternoon events occurred on days that also had morning events. “Cycling” refers to the strategy by which devices are controlled—for example, a 50-percent cycling strategy is when the controlled device is allowed to operate only half the time it would normally operate during a given half-hour and a 100-percent cycling strategy simply means the controlled device was completely shut off during the period of the event. On the pilot event days, water heater cycling was 100 percent, while space heating cycling was 50 percent, with an adaptive algorithm.
The pilot was conducted in Washington State on Bainbridge Island, in the western portion of the PSE service area. Natural gas service is unavailable on the island, and nearly all of the homes rely on electric space and water heating. The population of the island has increased from approximately 12,000 in 1980 to more than 23,000 in 2010, and the island’s substations are under considerable strain when winter demand peaks.
More than 500 participants participated in the pilot—electing to allow PSE to control their water heaters, space heaters, or both. Figure 1 shows the distribution of controlled devices among pilot participants. Of the 528 participants, quarter-hourly interval data were available for 494. Device-connected data loggers weren’t used in this analysis, meaning that the analysis relied on whole-home, rather than device-specific, demand data.
Curtailment was controlled via broadband using two-way communication, allowing PSE’s implementation vendor to track signal failures. Thus, Navigant was able to exclude devices experiencing signal failure from its analysis.
Program Demand Impacts
Demand reduction and snapback impacts were estimated using an econometric technique known as “fixed effects,” which is frequently used for impact evaluations like the PSE program analysis. Fixed effects make it possible to control for the fact that individuals can have very different levels of electricity consumption that don’t change over time; a participant with a very large house, for example, will consume much more electricity than one with a small house, just as one whose home is poorly insulated will consume more than one whose home is well insulated. Using fixed effects, it’s possible to control for all those differences between individuals that don’t change over time without collecting detailed information about each participant.
The estimated average demand impacts of the various devices on winter morning and afternoon events called by PSE are shown in Figure 2. As noted previously, there were only two afternoon events compared with seven morning events.
The estimated impacts of water heater curtailment weren’t particularly surprising, and are in fact very close to those estimated for PJM in a study by RLW Analytics in 2007 (see endnote 1). Specifically, that study found that the average water heater curtailment load impact between 6 a.m. and 9 a.m. totaled 0.73 kW, and that the average impact between 4 p.m. and 7 p.m. was 0.49 kW. Likewise, there was nothing particularly surprising about the impacts estimated for electric furnaces—outdoor temperatures in morning tend to be colder than the in the afternoon, leading to higher use in the mornings and therefore higher demand reductions during morning events.
The most surprising results are the very high estimated impact resulting from heat pump curtailment in the morning and the very low estimated impact for baseboard heating curtailment in the morning, along with the fact that no statistically significant impacts at all could be estimated for afternoon baseboard curtailment. Interestingly, according to the survey conducted by EMI, while more than half of participants with heat pumps needed to take extra steps to stay warm—e.g., put on a sweater—a large majority of participants whose water heaters, baseboard heating, or electric furnaces were controlled remained comfortable and undisturbed throughout the event. Indeed, many water heater participants were unaware that a curtailment event had taken place.
The morning heat pump impacts are even more striking when plotted (see Figure 3). The left vertical axis shows the average hourly demand observed among pilot participants with heat pumps on the two days in which there was both a morning and an afternoon event, the average hourly demand observed on days in which the temperature was comparable to that of event days, and what the analysis implies the average hourly demand would have been on the event days if no event had been called. The right vertical axis shows the average temperature at each hour of the day for the event days under consideration, and the average temperature at each hour of days with comparable temperature.
Particularly striking is the difference between the magnitude of the impact of morning and afternoon curtailments. While it’s clearly warmer during the afternoon curtailments, the degree to which the impact appears to be sensitive to that difference in temperature suggests that the impact of curtailing a heat pump isn’t linear in outdoor temperature. Or, to put it simply, the demand reduction impact isn’t necessarily proportional to the change in temperature. The incremental impact that would be observed when the temperature decreases from 32 to 30 degrees F, for example, is much greater than the incremental impact that would be observed when the temperature decreases from 35 to 33 degrees.
Further investigation revealed that the apparent step-change in heat pump curtailment impact is inherent to the design of the heat pump. A heat pump extracts heat from the air and then recirculates it. As a result, the heat pump’s efficiency diminishes as temperatures fall. Once the ambient temperature falls below a certain point, a heat pump typically will engage auxiliary resistance heat to supply the home’s load. This auxiliary resistance heat uses more electricity per Btu of home heating than the heat pump does, hence the step-change in curtailment impact between the morning, when temperatures were just a touch below freezing (when auxiliary resistance heat was required), and the afternoon, when temperatures were on average about 37 degrees F (when no auxiliary heat was required).
After some initial exploratory regressions, it became clear that baseboard curtailment was having little apparent effect on demand during control events, a matter of some consternation among the analysts involved. As noted before, the data-set included only those devices for which the receipt of the curtailment signal had been confirmed. As a result, installation data and interval demand data were investigated in greater depth.
A thorough examination—on a participant-by-participant basis—of individual interval data during the events showed that in many cases there appeared to be no impact on demand whatsoever as a result of baseboard curtailment. In other cases, it appeared as though baseboard curtailment had a very small, almost negligible impact. In a very few cases, however, it was apparent that baseboard curtailment was working as intended and delivering significant demand reductions.
While several theories have attempted to explain this phenomenon, the most plausible emerged from a review of installation data and discussions with installation personnel. The installation protocol established by PSE required that the largest-load baseboards in primary living areas and adjacent rooms be connected to the control switch. Through discussions with installation personnel, it became apparent that these key baseboard units couldn’t always be connected—either due to participant concern for their own comfort, or because the location of the electrical breaker panel precluded the connection of such key baseboards.
It also became apparent that in many cases, not all of the participants’ baseboard units were connected to the control switch. Baseboards not connected to a control switch simply worked harder to replace the controlled baseboards, leading to only a very small net reduction in household demand.
Winter Peaking Lessons
The results of the PSE pilot provide three significant lessons for winter-peaking utilities. The first—and most obvious—is that the direct load control of water heaters and electric furnaces can provide significant and reliable demand reductions during periods of peak winter demand. The second lesson learned is that winter-peaking utilities should carefully consider promoting the installation of heat pumps in their service territory.
In jurisdictions with a high penetration of electric heating, heat pumps are frequently promoted as a conservation measure, with heat pump uptake sometimes incented by the local conservation authority or utility. While heat pumps can deliver significant energy savings in—particularly in milder climates—the PSE pilot has demonstrated that heat pumps can actually increase a household’s winter peak demand when installed in regions where winter temperatures can be expected to drop below freezing on at least some winter mornings. Put another way, in colder climes, the implicit trade-off offered by a heat pump is a reduction in winter energy consumption for an increase in winter peak demand.
The final lesson for winter-peaking utilities is that any prospective direct load control program targeting baseboards should be very carefully planned, with a firm set of protocols to be followed by installers. If participant objections or the construction of the home mean that control device installers will need to deviate from the established protocols, then the device shouldn’t be installed. Perhaps the most important installation protocol that should be adhered to is the requirement that all baseboards in the home be connected.
Innovative and forward-looking summer-peaking utilities and conservation authorities have made great strides in incorporating residential demand response as part of their reliability resources, in some cases recruiting more than 100,000 participants. If deployed thoughtfully and run well, residential demand response can be a very cost-effective reliability resource and one that winter-peaking utilities should consider expanding. The PSE residential DR pilot has added considerably to the overall understanding of the impacts of winter direct load control. Further testing, however, is required to determine the most cost-efficient manner in which to deploy baseboard heating direct load control, such that it produces meaningful demand impacts. Additional experimental pilot programs could provide important information to guide the development of larger-scale residential direct load control programs and ultimately provide winter-peaking utilities with the same cost-effective reliability resource as is currently enjoyed by their summer-peaking counterparts.
1. See for example: Deemed Savings Estimates for Legacy Air Conditioning and Water Heating Direct Load Control Programs in PJM Region, March 2007, RLW Analytics, prepared for Lawrence Berkeley National Laboratory and the PJM Load Analysis Subcommittee, or Ericson, T., “Direct Load Control of Residential Water Heaters,” Energy Policy, Sept. 2009, pp. 3502 to 3512.
2. Impacts were evaluated by Navigant directly, process elements of the program were evaluated by Energy Market Innovations (EMI, www.emi1.com), sub-contracted by Navigant. See “2011 EM&V Report for PSE Residential Demand Response Pilot Program,” Navigant Consulting with EMI Consulting, Feb. 6, 2012.
3. Water heater and heat pump participants also were subject to a number of curtailment events.
4. An adaptive algorithm offers more significant demand reductions than simple cycling by using information regarding the controlled device’s pattern of use.
5. The fact that the data available was whole-home data rather than logger data, and that very few participants had only their space heating equipment controlled, posed an important challenge in the identification of events. Interested readers may obtain the entire report by contacting the author or PSE directly.
6. Figures calculated based on Table 5 in the RLW report.