Dave McCammant
December 23, 2008
Page 1

Date:December 23, 2008

To:Dave McCammant, MidAmerican Energy Company

From:Scott Dimetrosky and Matei Perussi, The Cadmus Group

Re:Evaluation of the SummerSaver Program

Background

MidAmerican Energy Company (MEC) operates a residential direct load control (DLC) program marketed as SummerSaver. The Program installs a small cycling device (a load control receiver, or LCR), connected to the home’s air conditioner or air-source heat pump. The equipment is wired into the existing thermostat control circuit and allows MidAmerican to cycle air conditioners off and on during peak usage days June through September. When cycling is necessary, air conditioners cycle off and on for 15-minute periods on weekdays between approximately 2 and 7 p.m. (but never on weekends and holidays). When cycling occurs, the air conditioner blower fan continues to operate, circulating cooled air inside the home.

Single-family owner-occupied homes in Iowa with a central air conditioning unit or air-source heat pump are eligible for the program. Participants receive a $40 end-of-season incentive the first year and a $30 incentive in following years. At the end of 2007 there were a total of approximately 60,000 homes enrolled in the program.

In an effort to measure demand and energy savings from the Program, MEC installed demand meters in a random sample of 105 participant homes. This memo presents the findings from the analysis of the metered data.

Methodology

The meters, which collected demand data in five minute increments, were installed in 2005, with the full sample installed for summer 2006 and 2007. The metered data were merged with hourly weather data for six weather stations in the MEC service territory. An indicator variable for direct load control events were then merged in with this dataset.

Data Inspection and Cleaning

Data from each of the meters was then visually inspected to ensure that the meters were functioning properly. The data were examined for unreasonably high or low values, extended periods of inactivity during hot weather, load shapes that were not correlated with weather patterns, and other possible indicators of malfunctioning. In total, seven of the 105 meters needed to be discarded from the analysis due to unreliable meter data.

In addition, the data were inspected on the event days to determine if the load control receivers (LCRs) were working properly. Each meter was checked one day at a time, seeing if they exhibited decreases in use for the five minute increments during the event cycling period. A number of problems were found that identified that the LCRs were not operating properly, including:

  • The meters were on during the entire DLC period;
  • The meters cycled up to 15 minutes and then stopped cycling;
  • The meters cycled one (or two) hour(s) on, only 10 minutes off;
  • The meters cycled off during a 10-30 minute period in the entire DLC period - remaining time meters were on; and
  • The meters cycled correctly one year but not the other year of the study.

In total 27.5% of the LCRs did not appear to be working properly (Table 1). Failures were higher for the FM LCRs compared to the Pager technology. In addition, the failed pager units were all previously documented by MidAmerican, and were already scheduled to be replaced as part of a Motorola Board Replacement Project.

Table 1. LCR Cycling by Technology Type

LCR Technology / N / Number of Meters that did not Show Proper Cycling / % of Meters that did not Show Proper Cycling
FM / 75 / 22 / 29.3%*
Pager / 23 / 5 / 21.7%**
Total / 98 / 27 / 27.5%
* Excludes one cycling event in July 2006 where the Des Moines FM signal did not function properly
** Previously documented and scheduled for replacement by MidAmerican as part of the Motorola Board Replacement Project.

Following the data cleaning, aggregated loadshapes were developed for both days with DLC events and peak (or near peak) days without DLC events. Figure 1 provides an example of the load shapes for DLC days in 2006. On average, participants exhibit reductions in usage during peak DLC periods (from hours 14-19, as shown by the vertical lines). There is a clear drop for each of the DLC days, although it varied based on the temperature preceding and during the event hours. The impact of temperature on the savings estimates will be discussed below.

In contrast, Figure 2 shows the average 2006 loadshapes for high temperature, non-DLC days. As can be seen from this on average the participants don’t exhibit any change of pattern for the non-DLC days. This provides a good reference for what a typical loadshape is like on a non-DLC day.

Figure 1. DLC Day Loadshapes for 2006

Figure 2. DLC Day Loadshapes for 2006

Model Development

Twenty-four hourly regression models were developed, one for each hour of the day. For model estimation purposes, however, only non-DLC days were used in the estimation. Since the weather data was only available at the hourly level, the five minute readings were averaged to obtain an average hourly reading. Since within an hour the air conditioners are cycling on and off at random patterns, this also helps smooth out the five minute loadshapes. Metered data from both 2006 and 2007 were pooled together for the analysis.

The final model specification was:[1]

DEMANDit = i +1 CDHtauit2 DLCt* CDHtauit + 3 MISERYBUILDUPit + 4 MAYTit + 5 JUNTit + 6 JULTit + 7 AUGTit + 8 SEPTit + 9MONt + 10 TUEt + 11 WEDt + 12 THUt + 13 WINDit + it

Where, for hour i on day t:

  • DEMANDit= Average hourly demand (kW) across the DLC participants
  • i = Intercept term
  • CDHtauit = The number of degrees the average temperature is above a reference cooling temperature value (tau) during hour i. If the temperature is less than or equal to tau, the value is set to zero. Thresholds ranged from 64 in night time hours to 80 in the peak hours.
  • DLCt = A binary variable, set to “0” for non-DLC days, and set to “1” during DLC days.
  • MISERYBUILDUPitA “misery index” that includes humidity and heat build-up calculations
  • MAYT it = An interaction term of May and temperature
  • JUNT it = An interaction term of June and temperature
  • JULT it = An interaction term of July and temperature
  • AUGT it = An interaction term of August and temperature
  • SEPT it = An interaction term of September and temperature
  • MON it = A binary variable that is 1 on Mondays, 0 otherwise.
  • TUE it = A binary variable that is 1 on Tuesdays, 0 otherwise.
  • WED it = A binary variable that is 1 on Wednesdays, 0 otherwise.
  • THU it = A binary variable that is 1 on Thursdays, 0 otherwise.
  • WIND it = A variable measuring the wind speed in miles per hour

The misery buildup index takes into account temperature and humidity from previous periods as well as the current period and better explains current usage than variables dependent on current conditions, which tend to be collinear with CDH. The only difference between the hourly model specifications from one hour to another is the threshold temperature used to calculate CDH, with the temperature cutoff for each peak hour based on the model that gave the highest adjusted r-squared.

Once the final models are estimated, the predicted values for all the historical and model estimation hours including the DLC hours are estimated from the model predictions. The difference between the predicted load shape and actual load shape on the DLC days provides the savings estimates during the actual DLC events. In addition, the regression model approach allows for savings estimation at any temperature, so savings can be modeled based on system planning temperature assumptions. 2 yields the kW savings during DLC hours or the take-back for non-DLC hours for every degree above the base temperature.

Energy savings were determined by examining the 24-hour load shape and comparing load shapes with and without DLC events. By examining all the hours of the day – particularly the hours just prior and subsequent to the DLC periods – the analysis accounts for any potential “take-back” from participants attempting to either pre-cool their homes prior to an event or increasing usage following an event to get their home back to a desired temperature.

Findings

Savings were examined for both planning purposes and the 2006 and 2007 program years. Results for both of these analyses are presented here.

Long-Term Planning Savings for SummerSaver

The hourly regression models were run based on four different average temperature assumptions for the DLC hours (2 p.m. through 7 p.m.): 90 degrees, 93 degrees, 96.5 degrees, and 98.7 degrees. These temperatures were selected based on planning assumptions regarding peak demand.

As shown in Figure 3, expected demand savings go up as the assumed temperature on the DLC go up.[2] Hourly savings reach their peak at 7 p.m., and then show a negative savings (i.e., increased consumption) are participants cool off their homes following the DLC event.

Figure 4 summarizes the expected demand savings for each temperature assumption. The savings are examined both as an average over the DLC period, as well as at hour ending 17, the historical peak hour for MidAmerican. Savings are highest at the “extreme” temperature of 98.7 degrees, at 1.048 kW per participant based on the average savings over the DLC period, and 1.1175 kW at hour ending 17. For all the temperature assumptions savings are higher at hour ending 17 compared to the average savings over the entire five hour DLC period.

A summary of the energy savings is presented in Figure 5. Like demand, energy savings is highest as the assumed temperature increases: at 90 degrees the expected energy savings is 1.137 kWh per day, while at 98.7 degrees the expected energy savings is 3.151 kWh per day. Note that despite the “take back” effect there is still a clear net positive energy savings (i.e., the energy savings during the DLC hours far offset any increase in consumption in the hours proceeding and following the event).

Figure 3. Demand (kW) Savings by Hour and Temperature Assumptions

Figure 4. Summary of Demand (kW) Savings by Temperature

Figure 5. Summary of Energy (kWh) Savings by Temperature

Savings from the 2006-2007 SummerSaver Program

While the long-term planning savings examine various scenarios based on a number of temperature assumptions, the 2006-2007 SummerSaver Program achieved savings based on the actual temperatures at which events were called. In addition, the savings during those years were impacted by known LCR failures.

In 2006 the average temperature during DLC events was substantially hotter (94.6 degrees) compared to 2007 (90.1 degrees), resulting in both higher demand and energy savings in 2006 compared to 2007 (Table 2). For example, demand savings – excluding LCR failures – was .8080 kW/participant in 2006, compared to .5724 kW/participant in 2007. If failed LCRs are taken into account, demand savings drop to .6656 in 2006 and .4839 in 2007.

Table 2. SummerSaver Demand and Energy Savings (2006-2007)

2006 / 2007
Average DLC Temperature / 94.6 / 90.1
Demand Savings (excluding failures) Average kW (hrs 15-19) / 0.8080 / 0.5724
Demand Savings (Including failures) Average kW (hrs 15-19) / 0.6656 / 0.4839
Energy Savings (Excluding failures) kWh/day / 1.8125 / 1.2429
Energy Savings (Including failures) kWh/day / 1.3483 / 0.9859

Conclusions

Savings for the SummerSaver Program vary significantly based on the temperature at which an event is called. Limiting events to days where temperatures exceed system planning “hot days” (96.5 degrees) will increase both the demand and energy savings achieved by the program. In addition, replacing the older FM LCRs will have a positive impact on savings by decreasing the failure rate.

[1] Final model output is included in Attachment A.

[2] Note that the long-term planning savings do not account for LCR failures.