Analysis of water heater standby energy consumption from ELCAP homes

Energy and Buildings, 19 (1993) 221-234

221

Analysis of water heater standby energy consumption from ELCAP homes R. G. Pratt, B. A. Ross and W. F. Sandusky Pacific Northwest Laboratory, Richland~ WA 99352 (USA)

Abstract The Bonneville Power Administration (Bonneville) routinely prepares forecasts of future energy demands in the Pacific Northwest region of the United States. Bonneville also implements conservation programs to reduce load demands. Results from the End-Use Load and Consumer Assessment Program (ELCAP), undertaken by the Pacific Northwest Laboratory for Bonneville, indicated that single-family homes with electric space-heating equipment consume more than 4700 kWh/yr to heat water for domestic uses. This energy use amounts to about 23% of the total electricity consumed. Additionally, the peak consumption for water heating coincides with regional system peak demands. Detailed analyses of the water heating end-use data acquired for residential buildings in ELCAP reveal that the average standby load for existing homes is 1200 kWh/yr, while homes built as part of the Residential Standards Demonstration Program averaged 1100 kWh/yr. These figures are consistent with the current figure of 1300 kWh/yr that is being used in the regional energy forecast. We also determined that standby loads for some of the participants were behaviorally driven. The data indicated the occurrence of vacancy setbacks in which the participant appears to lower the thermostat to save energy while the house is vacant. Anecdotal evidence from interviews revealed that this does occur. Reasons for setting back the thermostat ranged from not thinking about using the breaker, to fear that the tank would freeze in cold weather. These types of activities also appear to create the occurrence of dueling thermostats where the upper and lower thermostats, after the vacancy period, are not returned to the same temperature. This leads to additional energy use in an attempt to maintain a uniform temperature in the tank.

Introduction Over the past several years, a vast amount of a t t e n t i o n h a s b e e n g i v e n to t h e d e v e l o p m e n t o f c o n s e r v a t i o n p r o g r a m s a i m e d at r e d u c i n g s p a c e h e a t i n g a n d lighting l o a d s in r e s i d e n t i a l buildings. Because these loads represented a majority of the e n e r g y c o n s u m p t i o n f o r electrically h e a t e d resid e n c e s , t h e a p p r o a c h w a s logical. At t h e s a m e t i m e , initial efforts d i r e c t e d t o w a r d r e d u c i n g l o a d s associated with water heating were accomplished t h r o u g h installation o f t a n k w r a p s b e c a u s e a w a t e r h e a t e r w a s v i e w e d as a n e n e r g y - c o n s u m i n g a p p l i a n c e [ 1 ]. This a p p r o a c h w a s d i r e c t e d o n l y t o w a r d r e d u c i n g e n e r g y c o n s u m p t i o n a s s o c i a t e d w i t h the s t a n d b y load. H o w e v e r , h o t w a t e r t a p s also s h o u l d b e c o n sidered energy-consuming appliances, and apart f r o m t h e l o a d a s s o c i a t e d w i t h s t a n d b y losses, e n e r g y c o n s u m p t i o n f o r h e a t i n g w a t e r is b e h a v i o r a l l y driven [ 1 - 3 ] . This is t h e d e m a n d l o a d p o r t i o n o f t h e t o t a l

0378-7788/93/$6.00

e n e r g y c o n s u m p t i o n t h a t is a s s o c i a t e d w i t h h e a t i n g water. Recently, o t h e r a p p r o a c h e s b e y o n d t a n k w r a p s h a v e b e e n u s e d to r e d u c e t h e s t a n d b y load. T h e s e primarily consist of load control for either the tank itself [4, 5] o r u s e o f a t i m e r c o n t r o l f o r o n e o f t h e t a n k e l e m e n t s [6], i n s t a l l a t i o n o f l a r g e r c a p a c i t y w a t e r h e a t e r s [7], o r e l i m i n a t i o n o f s t a n d b y l o s s e s t h r o u g h t h e u s e o f a t a n l d e s s s y s t e m [8]. A l t h o u g h all o f t h e s e a p p r o a c h e s s h o w p r o m i s e , v a r i o u s drawb a c k s a n d c o s t limit t h e i r a c c e p t a n c e b y t h e p u b l i c a t this time. Energy savings from implementation of acceptable conservation measures for reducing the standby l o a d still r e m a i n u n c e r t a i n . L a b o r a t o r y m e a s u r e m e n t s h a v e s h o w n s a v i n g s a v e r a g i n g 570* k W h / y r f r o m u s i n g t a n k w r a p s [9]. M e t e r e d d a t a f r o m t h e *Figures extrapolated from data and temperatures used in this paper are rounded.

© 1993- Elsevier Sequoia. All rights reserved

222

Hood River project indicate the average savings to be 540 kWh/yr, but this included the savings as the result of wraps and pipe insulation. Results from the Residential Standards Demonstration Program (RSDP) conducted by Bonneville indicate the control homes in the study use 2 0 0 - 5 0 0 kWh/yr more electricity for water heating than those built to the proposed model energy conservation standards (MCS) for new, electrically heated residential buildings in the Pacific Northwest [2]. Newer, better insulated water heaters in the MCS homes compared to those in the control homes could explain part of a 200 kWh/yr difference. In addition, a large fraction of the water heaters in MCS homes were located in conditioned space compared to those in the control homes, but savings that could be associated with that effect were not determined. Further, while there were slight differences between water-heating values for the two groups of residences, the results indicate occupants of the residences had similar energy-related behavior. This would indicate the demand load for the two groups should be similar. Another metering program conducted by Bonneville, the End-Use Load and Consumer Assessment Program (ELCAP), provides detailed hourly consumption data at various end-use levels. Loads and load shapes for the first three years of data for each of several ELCAP residential studies representing various segments of the housing population have been summarized by Pratt et al. [10 ]. A more recent study by Taylor et al. [11] on five years of data provides additional consumption data of electricity to heat water. Results of this study indicate a strong morning peak between 08:00 and 10:00 and a secondary peak in the evening hours between 18:00 and 22:00. Thus, peak water-heating loads coincide with regional system peak demands in the Pacific Northwest. The measured residential annual end-use load distribution in ELCAP homes with electric space heating is shown in Fig. 1. The average annual amount of energy used to heat water in ELCAP homes is more than 4700 kWh, and represents 23% of the total amount of energy consumed. This percentage is 8% greater than estimated for the entire US residential population [12]. The hourly data collected from ELCAP provides us with opportunities to gain a detailed understanding of residential loads for water heating. That information is critical for both load forecasting and conservation resource assessments. Although ample data existed to investigate both standby and demand loads, this paper deals only with the standby loads. Besides getting a good understanding of the average standby load for the ELCAP sample, we wanted to

7598 k

ter

HVAC

Fig. 1. Fraction o f residential electricity devoted to water heating in ELCAP homes.

investigate other features that could affect standby loads that previously have not been examined. For example, can standby loads be behaviorally driven?

Overview The nomenclature used to describe hot water electricity consumption in this article is clarified before proceeding with a description of the analysis. The term l o a d generically refers to instantaneous consumption of electricity, as represented here by the hourly metered data. The instantaneous or hourly load is usually expressed in units of power (kW). The cumulative loads over time are usually expressed in units of energy per unit of time (kWh/yr, for example)*. The total hot water load comprises two components: s t a n d b y a n d d e m a n d . These are loads p l a c e d on the hot w a t e r s y s t e m by the occupants, as opposed to electrical loads. The distinction is that the water heater does not necessarily react to the loads at the same time they occur. The standby loads represent the steady heat loss from the tank, which is relatively constant over time even though the electricity consumed to make up for this heat loss typically occurs during only one of several successive hours, as will be illustrated. Thus, this rate o f t a n k heat loss is r e f e r r e d to as the s t a n d b y load, although it is expressed in terms of energy (kWh/yr). The individual bursts of electricity consumption that make up this heat loss are referred to as standby events. Similarly, the average demand *For convenience, most of our calculations, and h e n c e s o m e intermediate r e s u l t s p r e s e n t e d in the next Section, are e x p r e s s e d in average watts. Average watts are simply the average, instead of the sum, of the hourly loads over a given time. They are convenient because the time period need not be accounted for explicitly.

223 load is also expressed in units of energy (kWh/yr) and refers to occupant activities such as showers. ELCAP data used

The metered ELCAP data used in this analysis consist of all data available on hourly water heating for all residential homes in the sample as of April 1989 (typically 3 - 4 years). The metered end-use consumption data are subject to a rigorous test in which the sum of the end uses must equal the total load for the home within the measurement error of the metering equipment. Data failing this test are excluded from the analysis. Another work [13] describes the metering equipment and protocol in detail. Results of the standby load analysis are divided into three segments, one for each of the ELCAP residential studies. The characteristics of these studies are summarized in Pratt et al. [ 10] and described in detail in W'mdell [14]. The Base Study, the largest study in ELCAP, contains 288 of the 449 singlefamily site-built homes in the ELCAP project, and is the most regionally representative of any of the studies. The Base Study forms the backbone of this analysis and is a regional sample of homes that, in addition to being single-family and site-built homes, are detached, owner-occupied, and use electric space heat. The next largest study is the RSDP study, which consists of 105 new homes that were built as a demonstration of the MCS for space heat. As such, these homes do not represent a regional sample of construction practices or occupant energy use. Because hot water tanks in the RSDP study were new at the time of construction (1983-1984), they provide a view of the standby performance of tanks of that vintage. For this reason, they are included in the analysis, but their average standby load is reported separately. The smallest of the ELCAP studies is the case study. This study includes rental homes, existing manufactured homes, gas or off-heated, and one duplex. A small number of homes of each type was metered. Hot water tank characteristics for this analysis are drawn from extensive on-site inspections of each home [ 15]. The number and size of water heaters in each home were recorded, along with notations of standby conservation measures that were in place. The hot water tap temperature was measured by running a hot water faucet for a few minutes and measuring the hot water temperature with a thermometer. The inspections were conducted near the time the metering equipment was installed. Unfortunately, the scope of the study did not allow us

to obtain information regarding water consumption and inlet water temperatures. Data on the number and ages of the occupants were gathered annually through a series of mail and telephone surveys [ 14, 16, 171.

Standby load estimation methodology A principal problem in working with three years of hourly time series data from hundreds of homes is the amount of computational work required. To reduce this work, a filter was designed that eliminates possible standby events that are not significant to the analysis. The time-series nature of the data is maintained by tracking the times of all events retained (in hours from the start of the data time series). The filter that was used is described as a series of steps: • Step 1. Correct the original time series for possible offset drifts in the data logging equipment. The details for such activity are outlined in a report by Pratt and Ross [ 18]. • Step 2. Eliminate all hours with zero loads. • Step 3. Eliminate all periods of missing data except the first and last missing values in each episode. Retain these as a place holder for the event interval calculations. • Step 4. Compute the time interval between all remaining loads in the reduced time series. Time intervals for loads following the missing data place holders are marked as invalid. Intervals for any loads in the hour immediately preceding missing data periods are also considered invalid because the load may have continued but not been recorded. This vector of time intervals will now be filtered along with the hourly data. • Step 5. Eliminate all sequences of two or more continuous hours of nonzero loads. Because warmup events are of very short duration, typically 10 minutes or less, this step eliminates the multihour loads that characterize m ~ o r hot water demands typical of daytime and evening use. Portions of single warmup events may be recorded in two adjacent hours when the warmup event occurs just as the clock hour of the data logger is ending. This double recording will be eliminated by Step 5, but there should be numerous other events that occur entirely within clock hours. At this point, the original volume of hourly data has been reduced to an easily managed size, and remaining loads in adjacent hours can be combined into events and processed further. • Step 6. Following a nonwarmup event, the tank may be left in a partially depleted condition (i.e.,

224

with a temperature between the hot water setpoint and the top of the thermostat deadband). Drop the first of any adjacent warmup events. Note that this step will only keep the second, third, fourth, etc., warmup events in a series of warmup events. Single, isolated warmup events are dropped.

Standby load estimation The standby load at any point in time is the average rate of energy consumption when the tank is in the standby mode. This can be computed for any standby event as the energy consumed during the event (Qe) divided by the preceding interval (Ate). The average standby load over a number of events (n) can then be computed either as the mean energy divided by the mean interval

Qstandby----

(1)

or as the mean of the energy divided by the interval of each event Q~dby

l~Qe n Ate

(2)

Equation (1) is used in this analysis. Even under ideal conditions such as vacancies with constant air temperatures surrounding the hot water tank, standby events in hourly time series data will almost always be recorded as occurring with two distinct intervals that differ by one hour. Variations in the surrounding air temperatures over the course of the metering m a y cause further variation in the intervals, approximately equal to a ± 6 °C swing compared to a tank-air temperature difference of about 39 °C ( ± 6 / 3 9 , equal to ±15%). Varying air temperatures can introduce variation of one or more (for tanks with very long intervals between standby events) additional hours. Even ff standby event energies (Q~) are perfectly constant, a true average is not obtained using eqn. (2) because it effectively averages the reciprocal of the time intervals, while eqn. (1) gives a properly weighted average of the time intervals observed. There are additional compelling reasons for using eqn. (1) to estimate standby loads. Air temperature around hot water tanks may differ considerably during vacancy periods. The temperatures may be higher or lower than normal, depending on the season of the year and the tank location. According to data from occupied periods, lower air temperatures are also likely during the

nighttime, when most standby events usable in the analysis are detected. Therefore, standby loads are adjusted for differences in the air temperature for each event and the average air temperature for all data. This adjustment is given by a simple manipulation of the heat loss equation. The heat loss coefficient (UA) is a property of the tank, not the temperatures, so the simple equation of conductive heat loss Qstandby--~ g A ( T h o t -

Tlocal)

(3)

can be rearranged and applied to both instantaneous and long-term average conditions

UA

-~ Q s t a n d b y / ( T h o t

*

- - Vlocal) = Qstandby/(Thor --

*

Tlo¢~)

(4) where * indicates long-term averages, Thor is the hot water setpoint and T~o¢~ is the temperature of the air at the tank location. Equation (4) can be rearranged to give the average standby load under average conditions. * __ * Qstandby -- Qstandby(Yhot -- T l o c a l ) / ( T h o t

- Tlocal)

(5)

These adjustments are applied to each event remaining after the standby event filter is applied. Because the variation in temperature causes variation in event intervals, but not event energies, the reciprocal of this adjustment is applied to the event intervals At* ----Ate(Thot-- T, oc~)/(Thot - T~o¢~)

(6)

This adjustment is made before the average standby load is computed for each home using eqn. (1). To make the air temperature adjustment, local tank temperatures are estimated based on tank location. For tanks in conditioned living spaces (not heated basements), local air temperature and indoor air temperature measured by the ELCAP data loggers are assumed to be equal. Similarly, measured outdoor air temperatures are used for outdoor tanks, and weighted averages of indoor and outdoor temperatures are used for tanks in heated and unheated basements, crawlspaces, porches, and garages as suggested by Hanford et al. [19]. This technique has been used for other analyses [20], and the methodology is described in Pratt and Ross [18]. The fundamental approach used in the analysis is to estimate the average standby load using estimates for individual homes based on data from periods of vacancy, wherever possible. However, for some houses, vacant periods may not be detected, or the homes m a y have water heater characteristics (multiple tanks, timers, solar assist) that invalidate any estimate obtained. It is desirable, however, to

225 dividual homes reveals two types of nonideal patterns: vacancy setbacks and dueling thermostats.

obtain standby estimates for as m a n y metered homes as possible to estimate demand loads (by subtraction from the total hot water load) for the analysis of the influence of number and age of the occupants, and to reduce biases in average standby estimates that m a y result from using only those homes with vacancy-based estimates. To estimate standby loads for homes without vacant periods, a calibration ratio is estimated that adjusts the standby loads from occupied periods to better represent that from vacant periods. The calibration ratio can be computed for each home when estimates for both vacant and occupied periods are available. Using the average calibration ratio, the occupied standby-load estimates for homes without vacant periods can then be adjusted accordingly. This approach implicitly assumes that the warmup events in occupied periods are 'polluted' by handwashing events that form a relatively constant fraction of energy consumed by the set of warmup events identified for the home. Because increased uncertainty results from using these estimates in the analysis, the standby-load estimates include averages for the homes with vacancy-based estimates (for a smaller group of homes) and for all homes (using the adjusted occupancy-based estimates where necessary).

Vacancy setbacks The first type of nonideal standby event pattern, referred to as a vacancy setback, is illustrated in the energy-interval diagram of Fig. 2, which plots event energies and intervals from three consecutive days: an occupied day, a vacant day, and one intervening transition day. The algorithm for vacancy identification is described in Pratt and Ross [18]. This nonideal pattern is characterized by very constant event energies, as expected, but also by an unexpectedly broad band of intervals. In this case, the intervals range over four hours (from two to six hours). Ranges of 10 hours or more were observed. By examining the time-series data shown in Fig. 3 for these same three days, several potential standby events can be identified in the first day on the basis of the standby event filter described earlier (events " a " , "b", " c " , and "h"). All these potential standby events have intervals of two or three hours, as indicated by the energy-interval diagram in Fig. 2. After the last demand load at hours 5 and 6 on the second day (event "i" in Fig. 3), the absence of further demand events indicates the vacancy. All potential standby events from the vacant period (events " k " through " p " in Fig. 3) have intervals of 4 - 6 hours. This pattern is confirmed by examining other occupied/vacancy transition periods for the same home. The interpretation here is that the occupant has lowered the tank thermostat to save energy while the house is vacant. It is at first surprising that the occupant would not simply turn off the tank at the circuit breaker panel. Anecdotal evidence obtained

Investigation of nonideal standby event patterns An unresolved yet interesting issue identified in previously unpublished analyses of ELCAP standby loads is that although many homes exhibit relatively ideal patterns of standby events, nonideal patterns of events also occur. Comparing standby-load estimates from vacant and occupied periods for in12 lO 8

r,,

6

uJ >~ 4 uJ d i 0 1.5

le 2.0

gc

i 2.5

b

,h

3.0 Event Interval (hr)

Fig. 2. Typical energy-interval diagram -- vacancy setback.

I

,

3.5

4.0

jk

4.5

226

~-4 ~3 82 ~

bC

oL gg =

i

24

6

I

k i

i

i

12

i

18

i

i

i

i

24

6

J

i

H i

12

m

w |

=

t

i

24

18

i

n i

i

o

|

i

6

i

12

|

p !

18

i

24

Hour of Day Fi 8. 3. Typical hot water loads -- vacancy setback.

6

e-

ILl LLI

f

e

2

n

c 0

u qa

i

k

t

sb

gl pm

h

o

rj

I

I

I

I

5

10

15

20

25

Event Interval (hr)

Fig. 4. Typical energy-interval diagram -- dueling thermostats.

by interviewing readily available groups of homeowners revealed, however, that some do turn down their tank thermostats when on vacation. Reasons for this ranged from not thinking about using the breaker, to fear that the tank would freeze in cold weather. Turning off hot water tanks with the circuit breakers during vacancies undoubtedly prevents estimation of vacancy-based standby load for some homes, even when these periods of vacancy are detected. In the day after a vacancy with a setback, an extraordinarily large (high energy) warmup event will occur as the tank is reheated to normal operating temperatures. This event will also have an unusually long preceding interval compared to other potential standby events from occupied periods. As a result,

an additional precautionary step is added to the standby filter described earlier. • Step 7. Eliminate the first potential standby event after a period of vacancy.

Dueling thermostats A second type of nonideal standby event pattern, referred to as dueling thermostats, has been identiffed in this analysis. The energy-interval diagram in Fig. 4 illustrates data from a 10-day standby event pattern: two occupied and eight succeeding vacant days. The corresponding time series data are shown in Fig. 5. The standby events at this home show large variation in both energy and intervat, even during the vacancy period that begins after event "f".

227

3

j

g

==

LLI

g

q

1

0

u

,

u

u

u

!

i ,

u

u

,

u

24 12 24 12 24 12 24 12 24 12 24 12 24 12 24 12 24 12 24 12 24 Hour of Day Fig. 5. Typical hot water loads -- dueling thermostats.

A plausible explanation for this seemingly random pattern of standby events is not obvious. A failure in the vacancy detection algorithm was suspected, but the vacancy period clearly lacks the large demand events that occur during normal occupied periods. To investigate possible explanations, a simple thermal simulation of a hot water tank was constructed. Figure 6 shows the results of a simulated hot water tank containing two heating elements, with the bottom thermostat set at a slightly lower temperature than the upper thermostat (hence the term dueling thermostats). The top element heats only the water above it. The bottom element heats only the water below the top element, unless the water in the upper portion of the tank is cooler than the bottom thermostat setpoint plus its deadband. In this case it heats all the water in the tank to this temperature.

Top Tank

%;'%

'',

!"';

Temperature...,

"',,

i"',

",.

!"'-

]""

/

i"., i"",, i",, [""~J56.7

" ~ ' , . = ~ " , i

~ 3

In the case illustrated by Fig. 6, the top element is assumed to be one-third of the way down the tank. The two thermostats are assumed to have deadbands that differ by 2 °(3 which can also represent the effects of cooler water settling to the bottom of the tank. The result is a chaotic pattern of standby events very similar to those observed in the time-series data shown by Fig. 5. If the upper thermostat setpoint is slightly below the lower setpoint, no dueling occurs. Standby events from the upper and lower portions of the tank have different magnitudes because of the uneven portions of the tank they heat. Because of the unequal deadbands, the tanks also have different frequencies, which only occasionally cause the events to occur during the same hour. The unequal deadbands cause the uneven intervals of the standby events and occasional large event ener-

I

55.6 ~" o

e0

54.4

E

e~

~ 2

E

8

di Hour of Day Fig. 6. Simulation of tank with dueling thermostats.

dL

228 gies that are equal to the sum of the normal energies for the upper and lower tank. These effects are observed in the data in Fig. 5. Finally, twice during the hot water tank simulation, the bottom element heated the water in the upper part of the tank somewhat, thus disturbing the upper tank cycle time and adding further 'noise' to the event energies.

Vacancy setback a n d dueling thermostat detection Clearly, vacancy-based standby estimates from homes with vacancy setbacks do not represent the normal standby loads for these homes. Because these estimates must not be used in the analysis, homes that normally set back their tank thermostats during vacancies must be identified. The effect of dueling thermostats is uncertain. The varying time intervals and the correlation of the interval with the event energy (the longest intervals occur when the upper and lower tank events coincide -- the largest possible event energy) strongly support the use of eqn. (1) for computing the average load. However, the shortest possible intervals (in successive hours) are eliminated by the standby filter potentially biasing the result. Visually determining whether a setback has occurred for an individual vacancy for some homes is easy. Setbacks are clearly indicated when the vacancy and occupied event intervals never overlap. When dueling thermostats are also present, the broad ranges of event intervals may overlap considerably. Interestingly, many homes that clearly set back the thermostats also exhibit dueling thermostats. This makes sense, because the resetting of the upper and lower thermostats after the vacancy is unlikely to be uniform. Thus, dueling thermostat patterns may not be consistent throughout the metered time series. Because any given home may not set back the tank thermostat during every vacancy, particularly for short periods of a day or two, deciding whether a home generally exhibits setback behavior is somewhat arbitrary. So, like dueling thermostats, setback behavior may be inconsistent. Because dueling thermostats are most easily detected during vacancies, data are examined for both vacancy setbacks and dueling thermostats for e a c h individual period of vacancy. The steps in this process are as follows: • Step 1. Calculate summary statistics (the maximum, median, mean, and minimum) for the lntel-~als of the potential standby events in each occupied period and each vacancy period. • Step 2. Eliminate periods in which only three or fewer potential standby events were identified.

• Step 3. Eliminate periods with uncertain occupancy. These are often days of transition between occupied and vacant periods. • Step 4. Eliminate occupied periods in which the potential standby events represent more than 50% of the time series. Under normal occupied conditions, demand loads and their preceding intervals form a large majority of the time series. This step eliminates short periods of occupancy that may be incorrectly identified as such, or that may be occupied only briefly in a "caretaker" scenario. These are dropped because a setback may still be in effect. • Step 5. Combine the statistics of any adjacent occupied periods or any vacant periods that result from elimination of the intervening periods by Steps 2-4. Steps 1-5 create a simple summary of the pattern of potential standby events for the home during each period. This pattern is then analyzed in a process, developed through trial and error, that mimics the process of visually examining the data to make overall judgments about whether vacancy setbacks or dueling thermostats are indicated in each period. Because the judgments about the occurrence of vacancy setbacks are difficult, two levels of j u d g m e n t are applied: a strict criteria indicating certain setbacks, and a less strict criteria indicating suspected setbacks. Continuing the detection process: • Step 6. Flag each remaining vacancy period as indicating dueling thermostats if the range of event intervals is more than three hours. • Step 7. Score each remaining vacancy period as a " s u s p e c t e d " setback if: (a) the maximum vacancy interval is more than 0.5 hour longer than the maximum occupied interval, and (b) the maximum vacancy interval is more than 10% longer than the maximum occupied interval, and (c) the period is flagged as dueling o r the median vacancy interval is greater than the maximum occupied interval. Criteria (a) and (b) must hold if dueling thermostats are detected; (c) ensures the range of intervals in the two periods do not overlap to a large extent if dueling thermostats are not detected. One-half of a point is given to the suspected setback score for comparison with each of the preceding and following occupied periods. • Step 8. Score each remaining vacancy period as a "certain" setback if: (a) setbacks are suspected relative to both the preceding and following occupied periods (i.e., the suspected setback score is 1.0), a n d

229 (b) the minimum occupi ed interval is greater than the m a x i m u m u n o c c u p i e d interval o r dueling thermostats are flagged as certain. • Step 9. Determine the average dueling t h e r m o s t a t and suspected and certain vacancy setback scores for all vacancy periods analyzed for the home. The detection of dueling t her m os t a t s and v a c a n c y setbacks for any given va c a nc y period is intentionally fairly liberal. Averaging across all time periods in Step 9 to get an overall score for the home gives only minimal weight to the 'judgments' applied to e ach individual period. These overall scores axe then used to classify the validity of each h o m e ' s v a c a n c y estimate and develop the calibration ratio.

Occupied~unoccupied calibration ratio The calibration ratio is defined here as the average ratio of the occupied and vacancy standby estimates for all h o m e s with valid vacancy estimates. This ratio is multiplied by the oc c upa nc y- bas e d standby estimate to obtain an adjusted standby estimate. The adjustment a c c ount s for the average 'contamination' by handwashing and single-hour d e m a n d loads remaining in the original o c c u p a n c y estimate after the standby filter has been applied. Number of homes available for analysis Standby estimates were p r o d u c e d for 399 homes, with 305 h o m e s also having a v a c a n c y estimate. Homes with solar water heaters, setback timers, and more than one electric hot water tank are clearly going to p r o d u c e e r r o n e o u s standby estimates and are excluded here. This r e d u c e d the n u m b e r of occupied and vacancy standby estimates to 331 and 265, respectively. Any h o m e s that did not have at least 20 potential standby events from vacancy periods were eliminated, with 177 valid vacancy-based standby estimates remaining. Some h o m e s had unknown tank locations or lacked valid indoor air t e m p e r a t u r e s on which to base the tank location air t e m p e r a t u r e adjustments. This further r e d u c e d t he n u m b e r of

h o m e s with standby estimates ready for analysis to 289 (occupied) and 150 (unoccupied).

Use o f t r i m m e d or u n t r i m m e d m e a n s Recognizing that some handwashing events and one-hour dem and events remain in the events summary after the standby filtering was applied, 10% trimmed means were used in computing the standby estimates. The trimmed m ean standby estimates p r o d u c e d estimates nearly identical to the untrimmed means for vacancies, but significantly higher estimates for occupied periods. This suggests that the use of trimmed means was effective in reducing the effects of the handwashing and one-hour demand events from the filtered occupied standby events. The value of the 25% t ri m m ed m eans beyond the 10% trimmed means a p p e a r e d marginal, so 10% trimmed means were used. Standby estimates based on the 10% trimmed means are summarized in Table 1. Table 1 highlights the fact that the t e m p e r a t u r e adjustments have a negligible effect on the overall results, since the difference between the valid and adjusted estimates is very small. D e t e r m i n a t i o n o f the c a l i b r a t i o n ratio The distributions of the scores for dueling thermostats and both suspect ed and certain vacancy setbacks obtained from the detection pro c e ss described earlier do not cluster into two groups that distinguish t hose h o m e s that do from those that do not show these patterns of standby events. If selected criteria are t oo tight, the size of the remaining sample used for determining the calibration ratio will be greatly reduced. If criteria are too loose, bias in the final estimates will result. To help select appropri at e cutoff points for these criteria, plots were m a d e of the calibration ratio resulting from the full range of setback scores and the resulting calibration sample size. Figure 7 shows the calibration ratio as a function of the cutoff values selected for the two setback criteria. Each line

TABLE 1. Standby estimates (10% trimmed mean) and calibration ratios for ELCAP homes in the analysis Sample analyzed

Vacant (Avg. W)

Occupied, all est. (Avg. W)

Occupied and vacant (Avg. W')

Calibration ratio

Total no. of homes

No. of homes vacant

All Valid A~us~d Calibration

124 116 114 123

176 174 172 156

176 164 162 156

1.57 1.49 1.52 1.30

399 321 289 104

305 177 150 104

230

1.55 / | .50

Certain S e l b a c k s < = j

/

~"~'-"01.40

o

o ~

__%2--

,.35

-

--_-ZU_-_-

......... .

.

.

.

.

.

.

"

.

.

.

.

.

.

.

.

.

.

.

.

1.30

, 0.4

, 0.6 Suspected Setbacks

1.25

0.0

0.2

, 08

1.0

Fig. 7. Calibration ratio as a function of setback frequency criteria.

160

=

,

i

i

i

c,.=.2.~

140 120 u)

o0 "6 d z

100

~>" 8O

_ _____02 . . . . . . . . . . . . . . . . . . . .

~

i

. . . . . . . . . ~,J . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6O 40 20 0.0

i

I

I

I

0.2

0.4

0.6

0.8

1.0

Suspected Setbacks Fig. 8. Number of sites passing setback frequency criteria. r e p r e s e n t s the result for a constant value o f the certain setback frequency, which is indicated as a label on each line. Similarly, Fig. 8 shows the calibration sample size as a function of the same criteria. As both criteria b e c o m e stricter, Fig. 7 shows that the calibration ratio stabilizes at a little less than 1.3 for a criteria of certain setbacks detected for less than 40% of the vacancies (the line labeled 0.4). The calibration ratio varies only slightly ff stricter criteria for suspected setbacks are used. Figure 8 indicates the sample size at this point is about 105 h o me s (actually 104), and begins to fall off rapidly for certain suspected setback detection frequencies of 0.5 or less. About one-third of the potential sample has been eliminated by setting the calibration ratio with this criteria. Use of a h o m e

with vacancy setback frequencies less than or equal to (7.4 seems appropriate. Similarly, Figs. 9 and 10 show the calibration ratio and sample sizes resulting from a combination of cutoff criteria for certain setback and dueling t herm ost at frequencies. Figure 9 indicates that the calibration ratio based on h o m e s with certain setback ratios less than 0.4 is almost completely independent of the dueling thermostat frequency cutoff criteria used*. A calibration ratio of 1.3, based solely on a calibration sample for hom es with certain thermostat setback frequencies less than or equal to 0.4, is therefore used in this analysis. *The s a m e is n o t i~ue at higher cutoff values o f the s e t b a c k criteria, indicating t h e degree to which dueling t h e r m o s t a t s a p p e a r to be associated with t h e r m o s t a t setbacks.

231

1.55 / O

1.50

CertainDuellng<-l~ / ~"'-'~o.,~

O= e~ "O

1.4S

~

e-

09 t-

....................

O

.

o 1.30 1.25

~ 0.0

0.2

i

,

I

0.4

0.6

0.8

1.0

Suspected Setbacks

Fig. 9. Calibration ratio as a function of dueling and setback frequency criteria. 160

.

.

.

.

.

140

..~L__ ~ - ~ . . ~ - - - ~

120

~

~

~

...........

_

. . . . .

. . . . . . . . . . . . . . . . . . . .

loo

...............

.

~ 40 20

~

.

.

.

.

.

.

.

.

.

.

.

~

.

_- ......

. ................................:'0.............................................

......

'

I

I

I

,

0.0

0.2

0.4

0.6

0.8

.

.

.

.

.

.

...................

1.0

Suspected Setbacks Fig. 10. N u m b e r o f sites p a s s i n g s e t b a c k and d u e l i n g criteria.

An analysis was c o n d u c t e d of the homes with dueling th er mo st a t s that remained in the calibration and noncalibration samples. Results of this analysis are summarized in Table 2 which shows that the h o m e s with t h e r m o s t a t setback frequencies above 0.4 have v er y low standby-load estimates. It also d e mo n s tr ates that including h o m e s with dueling t h e r m o s t a t frequencies above 0.4 has a negligible effect on the calibration ratio*. * T h e m o s t n o t a b l e r e s u l t n o t s h o w n h e r e is f o r n o n d u e l i n g t h e r m o s t a t h o m e s . W h e t h e r exhibi{~ng t h e r m o s t a t s e t b a c k s or not, t h e s t a n d b y e s t i m a t e f r o m t h e m e a n e v e n t ratio, e q n . (2), is n e a r l y e q u a l to t h e r a t i o of t h e m e a n e v e n t e n e r g y to t h e m e a n e v e n t interval u s e d in t h e a n a l y s i s . T h i s r e s u l t s u p p o r t s t h e d u e l i n g t h e r m o s t a t d e t e c t i o n c r i t e r i a a n d is c o n s i s t e n t with h o w t h e m e a n e v e n t r a t i o is e x p e c t e d to i n t e r a c t w i t h d u e l i n g thermostats.

Standby load results

Table 3 shows the temperature-adjusted mean standby-load estimates for the hot water tanks in the calibration sample by ELCAP study (Base, RSDP, and other). Also shown are standby estimates extrapolated to the entire sample of valid homes. This extrapolation is accomplished by using the p ro d u c t of the calibration ratio and the standby estimate from occupied periods for each hom e to provide an estimate of the true standby load for homes without a valid vacancy-based estimate (the noncalibration homes). This adjustment is made to obtain a larger sample size and so demand loads can be estimated by subtraction.

232 TABLE 2. Summary of results: standby estimates (1096 trimmed mean) and calibration ratios for ELCAP homes in the analysis Sample analyzed

Vacant (Avg. W)

Occupied, all est. (Avg. W)

Occupied and vacant (Avg. W)

Calibration ratio

Total no. of homes

No. of homes vacant

Calibration Dueling Not dueling

123 122 124 92 89 101

156 157 155 177 182 163

156 157 155 177 182 163

1.30 1.31 1.29 2.03 2.15 1.62

104 32 72 46 35 11

104 32 72 46 35 11

Noncalibration Dueling Not dueling

TABLE 3. Calibration and extrapolated standby estimates in kWh/yr Sample

Calibration homes Extrapolated Mean

ELCAP -- all homes Base Study homes RSDP homes Case studies

n

S.

Mean

n

S.

Dev.

Dev.

1078 104 289 1092 63 284 1013 30 280 1176 11 334

1 1 6 2 289 418 1 1 9 1 182 457 1075 75 301 1202 32 401

Table 3 shows that the average extrapolated standby load for the Base Study is just under 1200 kWh/yr, and for the RSDP study the load is just u n d e r 1100 kWh/yr. On-site inspection data indicate that 41% o f water heaters in the ELCAP sample are wrapped, 9% have bot t om boards, and 2% were n o t e d to have thermal traps. How these figures c o m p a r e d to the region as a whole in 1984 is unknown. Standby estimates for the calibration sample tend to be about 9% and 6% less for the ELCAP and Base studies, respectively. This m ay r e p r e s e n t real differences between hot water tanks and conditions in the calibration and extrapolated samples or an e rr o r in the calibration ratio caused by use of a setback criterion that is too strict. Further means o f investigating reasons for this difference were not developed.

Hot water temperatures and tank location temperatures Standby estimates obtained for the ELCAP h o m e s and th o s e used for regional planning m ay differ ff average hot water t e m per a t ur es or the mixture of tank locations (and hence the surrounding air temp er atu r es ) for the ELCAP hom e s differ markedly from the homes in the region.

Hot water tap temperatures, approximately equal to hot water setpoints, average 59 °C for the Base Study and 55 °C for the RSDP study. The average hot water tap temperature in h o m e s in the case study is even higher - 61 °C. This raises the overall ELCAP average to 58 °C. The m o s t c o m m o n tank locations for the Base Study are in primary occupied spaces (34%), heated and daylight basements (27%), and both unheated basem ent s and garages (29%). The remainder (10%) were located on the porch/ sunspace, crawlspace, or outdoors. The regional estimate assumed that half were located in conditioned space and half in nonconditioned space. The RSDP Study, which may be an indicator of current construction trends, has very few unheated basements. The p e r c e n t a g e o f hot water tanks located in unheated basements in the Base Study appears, for the RSDP study, to have been installed in primary occupied zones, t h e r e b y raising the percentage in occupied zones to 489/0. E x c e p t for occupied zones, tank location temperat ures are estimated, not measured. Nevertheless, t hey should reasonably approximate the actual air t e m p e r a t u r e s at the tank locations. For the Base Study, the tank location t em perat ure averaged 17 °C, ranging from 21 °C in the occupied zones to 11 °C for the one tank located outdoors. Similar t e m p e r a t u r e s are indicated for the RSDP study, although the overall average is higher due to the shift in tank locations. The t e m p e r a t u r e difference driving standby heat loss is about 24 °C in the Base Study, 19 °C in the RSDP study, and 23 °C for the ELCAP h o m e s as a whole. The average t e m p e r a t u r e difference used t o develop regional planning estimates for standby losses is 21 °C and 27 °C for demand. The temperat ure differences in the ELCAP h o m e s clearly do not explain why regional planning estimates are higher than the ELCAP standby-load estimates. The differences are m ore likely due to conservation

233

measures already installed in the ELCAP Base

Study. C o n s e r v a t / o n m e a s u r e e~ects To determine conservation measures that reduce standby loads, standby-load estimates for homes with and without conservation measures were compared using a series of regression models. Four types of models were tried: (1) regression models of standby loads against a list of key explanatory variables including tank surface areas, tap temperature/tank location temperature differences, and conservation measures; (2) regression models of standby loads against conservation measures listed in Table 4, multiplied b y the product of their effective surface areas and the difference in hot water tap temperature and tank location temperature; (3) regression models of tank heat loss coefficients (UAs) against conservation measures listed in Table 4; (4) regression models of tank U-values against the list of conservation measures in Table 4, divided by their equivalent surface areas. The effective surface area for tank wraps and tanks with nominal R-values greater than R-3 is the tank surface area. The effective surface area for b o t t o m boards is the tank b o t t o m area (half the area of the ends of the tank), equal to 9% of the total surface area on average [18]. The effective surface area for pipes is estimated as 0.35 m ~ based on laboratory tests [21]. Only those h o m e s with valid standby estimates and a complete set of explanatory variables were used in the models. The explanatory variables used in the regressions were eliminated in a stepwise procedure, dropping from each step the variable with the lowest T-statistic. The results of constructing these models are relatively disappointing, explaining low fractions of the variance, R 2 values of 0.10 and less [18]. Nevertheless, some statistically meaningful coefficients are indicated by T-statistics near or above 2.0. Generally, the only conservation measures shown to be statistically significant are tank wraps and T A B L E 4. Conservation measure variables used as explanatory variables in standby load models

Tank wraps Tanks noted to have R-values> 3 Bottom boards Traps/anticonvection valves (ACVs) only Pipe insulation only Traps/ACVs and pipe insulation

bottom boards. The nonphysical form of the standby model clearly indicates that the surface areas and temperature differences explain significantvariance. However, itis difficultto interpret.Addition of other data regarding water heaters, such as inlet w a t e r temperature and actual amount of water use, would have helped. The UA model coefficients for tank wraps and b o t t o m boards appear to roughly confirm laboratory effects [21 ]. The magnitude of the tank-wrap savings is approximately as the laboratory experiment indicates, if the benefit of tank wraps is discounted to reflect the sizable fraction of wraps that only partially cover the tanks. The savings value for the b o t t o m boards is remarkably close to the average of the laboratory test results.

Conclusions

The primary objective of the analysis is obtaining an estimate of the average standby load (heat losses) of the current population of electric hot water tanks. The average standby load is an important element of the residential forecast, because much of the reduction in water-heating energy consumption in the future is the difference between the current loads and those projected to result from tank efficiency standards and conservation programs. Standby loads were estimated for the three groups of ELCAP homes: (1) Base Study homes (all single-family, detached, owner-occupied with electric space-heating equipment); (2) Residential Standards Demonstration Program CRSDP) study homes (all constructed in 1 9 8 3 - 1 9 8 4 as part of the RSDP); (3) All case study homes in ELCAP. All homes analyzed have a single active electric hot water heater. The results of this analysis are: • The a v e r a g e s t a n d b y load f o r the E L C A P B a s e S t u d y homes is a b o u t 1200 kWh/yr, which is less than used in previous forecasts; • The a v e r a g e s t a n d b y load in the R S D P s t u d y h o m e s /s a b o u t 1100 k W h / y r , less than that for the Base Study. • The B a s e S t u d y homes h a v e m o r e w a t e r heating conservation m e a s u r e s installed t h a n i n d i c a t e d i n the regional a v e r a g e for similar homes. In ELCAP homes, 41% of the water heaters axe wrapped, 9% have bottom boards, and 2% have thermal traps.

234

Effects o f c o n s e r v a t i o n m e a s u r e s To determine effects of conservation measures, a series of regression models were used to compare standby-load estimates for sites with and without conservation measures used to reduce standby loads. • Amodelofheatlosscoefficientsproducedsavings e s t i m a t e s f o r t a n k w r a p s a n d bottom boards. The results confirm laboratory tests [21], if the value of tank wraps is discounted to reflect a sizable fraction of wraps only partially covering the tank. The value for the bottom boards is remarkably close to the laboratory test results. Other observations In completing the analysis, several nonideal events were uncovered. The first was the occurrence of occupants lowering the tank thermostat to save energy while the house is vacant. It appears this approach is used instead of turning off the breaker to reduce the potential of the tank freezing in cold weather. It is unknown how widespread this approach is used, or if it is a result of sample selection. Several of our participants were retired and vacation in the southwestern United States during the winter. As the result of these types of thermostat adjustments or other maintenance activities, such as element replacement, we observed the occurrence of dueling thermostats. As the name implies, improper temperature setting can increase the amount of energy for heating water.

Acknowledgement This work was supported by the Bonneville Power Administration under a Related Services Agreement with the US Department of Energy (DOE) under Contract DE-AC06-76RL0 1830. Pacific Northwest Laboratory is operated for DOE by Battelle Memorial Institute.

References 1 W. Kempton, Residential hot water: A behaviorally driven system, Energy, 17 (1) (1988) 107-114. 2 A. K. Meier and B. Nordman, A thermal analysis of the model conservation standards for n e w homes in the Pacific Northwest U.S-~., Energy, 17 (11) (1988) 833--844. 3 J. S. Weihl and W . Kempton, Residential hot water energy analysis: instruments and algorithms, Energy Build., 8 (1985) 197-204.

4 V. A. Rabl and R. P. Blevins, Load management experience in the United States, in D. R. Limaye and V. Rabl (eds.), Inte~vzational Load Management: Methods and Practices, Fairmont Press, Atlanta, GA, 1988, pp. 7-36. 5 B. F. Hastings, Manage loads to reduce capacity needs, Electrical World, 184 (10) (Nov. 15) (1975) 92-95. 6 J. M. Akridge and D. Keebaugh, An investigation of off-peak domestic hot water heating, ASHRAE J., 32 (1) (1990) 32-37. 7 S. Sticker and R. L. H~jas, Canadian experience in demandside management, in D. R. Limaye and V. Rubl (eds.), International Load Management: Methods and Practices, Fairmont Press, Atlanta, GA, 1988, pp. 37-49. 8 P. duPont, Going tankless, Home Energy, 6 (5) (1989) 34-38. 9 A. Meier, Saving water heating energy, Energy Auditor & Retrofitter, 2 (1) (1985) 9-14. 10 R. G. Pratt, C. C. Conner, E. E. Richman, K. G. Ritland, W. F. Sandusky and M. E. Taylor, Description of Electric Energy Use in Single-Family Residences in the P a c ~ Northwest, DOE/BP-13795-21, Bonneville Power Administration, Portland, OR, 1989. 11 M. E. Taylor, K. G. Ritland and R. G. Pratt, Hot Water Electric Energy Use in Single-Family Residvnves in the Pacific Northwest, DOE/BP-13795-27, Bonneville Power Administration, Portland, OR, 1991. 12 U.S. Congress, Office of Technology Assessment, Building Energy Efficiency, OTA-E-518, US Government Printing Office, Washington, DC, May 1992. 13 G. B. Parker, E. W. Pearson and W. F. Sandusky, The Residential Pilot Study, DOE/BP/13795-7, Bonneville Power Administration, Portland, OR, 1985. 14 P. A. Windell, Overview of ELCAP study objectives, in Workshop I1 Abstract, DOE/BPA-13795-10, Bonneville Power Administration, Portland, OR, 1985. 15 Procedures Manual f o r ELCAP ResidentialBuilding Characteristics Survey, DOE/BP-13795-9, Bonneville Power Administration, Portland, OR, 1986. 16 R. F. Darwin, D. L. Ivey, M. S. Klan, S. A. Shankle and B. L. Mohler, Pacific Northwest Residential Energy Survey: 1985 Residential Occupant Survey - Telephone, DOE/ BP-13795-15, Bonneville Power Administration, Portland, OR, 1986. 17 D. L. Ivey and P. K. Alley, 1986 Residential Occupant Survey, PNL--5138, Pacific Northwest Laboratory, Richland, WA, 1987. 18 R. G. Pratt and B. A. Ross, Measured Electric Hot Water Standby and Demand Loads f r o m Pacific Northwest Homes, PNL-SA-7889, Pacific Northwest Laboratory, Richland, WA, 1991. 19 J. Hanford, M. Kennedy, M. J. DeLaHunt and L. Palmiter, Heat Pump Water Heater Field Test (DraJt Final Report), Ecotope, Seattle, WA, 1985. 20 Technical Appendix to Conservation Supply f o r the 1990 Power Plan, Northwest Power Planning Council, Portland, OR, 1989. 21 C. W. Ek and C. D. Auberg, Electric water heater standby losses: comparison of conservation strategies and their energy savings, Proc. ACEEE S u m m e r Study on Buildings, American Council for an Energy Efficient Economy, Washington, DC, 1984.