Energy and Buildings 36 (2004) 161174

Weather sensitivity in household appliance energy end-use

Melissa Hart, Richard de Dear
Division of Environmental and Life Sciences, Macquarie University, Sydney, NSW 2109, Australia Received 17 July 2002; accepted 31 October 2003

Abstract Data from a Residential Energy Study (RES) were used to examine the weather sensitivity of various household appliances located in households within the Sydney metropolitan area. Thermal environmental indices effective temperature (ET ), standard effective temperature (SET ) and simple air temperature degreedays were used to quantify the dependence of household appliance energy consumption on outdoor weather. Specic appliances included: room air-conditioners, room heaters, refrigerators, freezers and domestic hot-water systems, all of which exhibited some degree of weather sensitivity, particularly space heating and cooling devices. Probit regression techniques were used to predict the degreeday values at which households tend to switch on heating and cooling appliances. All appliances demonstrated weather sensitivity to varying degrees, and this was universally stronger during the cooling season (summer) than during the heating season (winter). The outdoor SET version of the degreeday index demonstrated a stronger statistical association with space-cooling energy consumption than conventional air temperature degreedays. The mean daily temperature associated with minimum heating and cooling energy consumption for Sydney indicated that a temperature of 18 C was the most appropriate base temperature for calculation of both heating and cooling degreedays. 2003 Published by Elsevier B.V.
Keywords: Degree-day; Weather sensitivity; Household appliance; Residential Energy Study (RES)

1. Introduction Australian electricity suppliers Pacic Power and Sydney Electricity carried out a Residential Energy Study (RES) in New South Wales (NSW) during 19931994 [1]. The study involved directly metering the energy end-use of household appliances in a sample of houses across NSW. A preliminary study on the database commissioned by the Australian Greenhouse Ofce (AGO) examined the patterns of usage and energy consumption of the appliances [1]. Whilst weather sensitivity was not the focus of that study, it found, inter alia, signicant summer and winter peaks in energy consumed by many of the appliances and this prompted the present study. The present paper specically examines weather sensitivity in electricity consumed by a selection of the household appliances monitored in that study within the Sydney metropolitan region. The data collected during the RES was analysed alongside concurrent weather data collected in close proximity to the households being studied. Statistical relationships observed between outdoor weather and individual appliance energy consumption for
Corresponding author. Tel.: +61-2-9850-7582; fax: +61-2-9850-8420. E-mail addresses: mhart@els.mq.edu.au (M. Hart), rdedear@laurel.ocs.mq.edu.qu (R. de Dear). 0378-7788/$ see front matter 2003 Published by Elsevier B.V. doi:10.1016/j.enbuild.2003.10.009

an 18-month sample period are used to quantitatively dene the weather sensitivity of appliance energy consumption. 1.1. Weather sensitivity of electricity consumption In most electricity systems the residential sector is one of the main contributors to system peaks [2]. Usage patterns of many household appliances such as heating, ventilation and air-conditioning (HVAC) are expected a priori to be affected by outdoor weather variations. Energy consumption will depend on building envelope characteristics and occupant behaviour. The latter is subject to myriad inuences including householders subjective comfort preferences [3,4], their socio-demographic characteristics [5,6], subtle cognitive factors [7] and even cultural dimensions [3,8,9]. Yet, despite the plethora of human factors impinging on energy consumption, the presence of thermostats in certain appliances such as domestic hot-water systems and refrigerators establishes their sensitivity to weather to varying extents, depending on building envelope thermal performance. Obviously, rates of heat exchange between these appliances and their indoor environment are also affected by the heating, ventilation and air-conditioning (HVAC) system in operation at the time. Fourteen different electrical appliance types were monitored in the original 19931994 Residential Energy Study

162

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

(RES) database upon which the present study is based. Of these, room air-conditioners, room heaters, refrigerators, stand-alone freezers and domestic hot water were selected for analysis in this study. These decisions were made partly on the basis of the a priori logic outlined in preceding paragraphs, but were also informed by preliminary pilot tests on the RES database. The RES involved an initial telephone survey of 600 households, as part of a house selection process and to establish the broad demographic and socio-economic characteristics of typical households in NSW. For households involved in data collection, eight channel data loggers were used with the last channel used to monitor total household load. The duration of data available for each household ranged from a few days to the entire monitoring period of 18 months. Of the appliances monitored only 8% had complete records without any apparent errors or missing half-hourly readings [1]. The present study is the rst of its kind in Australia where directly monitored energy consumption is analysed alongside concurrent weather observations. Awareness of the weather sensitivity in energy end-use can provide more information on actual in-use energy consumption for comparison with laboratory measurements on domestic appliances. This knowledge may also provide practical benets in relation to the implementation of testing procedures for Australias Minimum Energy Performance Standards (MEPS) and energy efciency programs such as star rating energy labels. Similarly, results from this research may be of use to international energy efciency programs such as the European Energy Performance of Buildings Directive. The relationship between heating and cooling energy consumption and outdoor weather, along with threshold temperatures at which occupants begin to heat and cool their house may potentially be useful for power utility companies in predicting system spikes and peaks in the Sydney market. It has been mandatory in Australia for all refrigerator, freezers and electric hot-water systems manufactured since 1999, and air-conditioners manufactured since 2001 to meet MEPS. The introduction of more energy efcient appliances into many Australian households since the original RES monitoring programme may have an inuence on current total energy consumption of appliances; however, the correlation coefcients between energy consumption and outdoor weather discussed in this paper, the weather sensitivity of the appliances, will have no change. Australian households consume a large amount of standby energy, eight percent of residential greenhouse gas emission are contributable to standby power [10]. This level of standby power, which generates greater internal heat during both seasons, will have increased since the original RES. This increase in internal heat may have a slight inuence on appliance energy consumption, or may have a slight inuence on the temperature at which occupants decide to heat or cool, however this inuence would be small and once again these changes would not affect the overall sensitivity between appliance energy consumption and outdoor weather which is the focus of this paper.

1.2. Literature review of electricity end-use weather sensitivity Previous studies have tended to focus on either total household energy consumption or heating and cooling energy consumption patterns. A study in Canada [11] analysed the impacts of appliance efciency and fuel substitution on residential end-use energy consumption. But rather than directly measuring appliance electricity usage, Ugursal and Fung relied mainly on numerical simulations. A similar Residential Energy Consumption Survey [6] by the US Energy Information Administration (EIA) surveyed consumption patterns with questionnaires directed to end-users, including information on the energy-related characteristics of building envelopes, household characteristics and behaviour patterns. A computer program developed by Princeton University, PRISM (PRInceton Scorekeeping Method) [12] uses input of monthly energy billing data and average daily temperatures to produce a weather-adjusted index of energy consumption. The program returns a heating reference temperature for heating or cooling, that is the average outside temperature at which the buildings heating and cooling systems commence. Previous studies have also looked at broad climatic zones [6,13] or used climatological normals [14] from historical weather data, rather than concurrent weather observations. Despite these methodological differences from the present study, key ndings include differences in space heating and cooling energy consumption within different climatic zones [6,13]. The effect of each of the weather variables: air temperature, relative humidity and wind speed, on electricity consumption was found to differ across climate zones [14]. In a similar monitoring program to Pacic Powers Australian RES underpinning the present study, the Household Energy End-Use Project (HEEP) in New Zealand directly monitored individual appliances in over 100 households for a period of 4 years to date, analysis is ongoing. The analysis included building and socio-demographic characteristics of the households [5]. However, emphasis in the analysis and discussion was placed on individual appliances energy consumption in relation to indoor, rather than outdoor temperature. Actual household energy use and temperature measurements are correlated to an energy use model: ALF3, an annual loss factor model used for calculating heating energy requirements of buildings in different climate zones [15]. A domestic electricity end-use measurement campaign undertaken in France [16] over 12 months in 1994/1995, monitored eight appliances every 10 min in 94 households for a duration of 1 month for each household. Appliance energy consumption was compared by season, winter and summer, but the authors did not analyse their data alongside concurrent weather observations. Because the research design was cross-sectional with a short sample duration of just 1 month, comparisons of individual appliance energy consumption could not be made on either synoptic weather

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

163

or seasonal climatic time scales. The study did not include electrical space heating and cooling nor domestic hot-water appliances. Results included a breakdown of residential electricity consumption by end-use and an estimate of the annual running cost of each end-use, taking into account different hourly electricity tariffs. More recent end-use campaigns in Europe have assessed potential reductions in energy consumption by replacing regular household appliances with the most energy efcient examples available on the European market. The French ECODROME project monitored appliance electricity consumption in 20 households over 2 years [17]. Current plug loads and the electric light circuit consumption were monitored for the rst year. At the beginning of the second year all appliances and light bulbs were replaced by the most efcient alternatives available on the European market. Results indicated that the average annual energy saving per house from the use of efcient equipment was 1800 kWh per year. A more recent project, EURECO, followed a similar methodology and approach to ECODROME. However, rather than replacing individual appliances, the measured consumption of currently installed appliances was compared to that of the most energy efcient model of similar capacity and function, and duration of usage [18]. Appliances were monitored in 100 households in each of Denmark, Italy, Portugal and Greece with 10 min intervals over 1 month. Based on analysis for all four countries, annual savings were calculated to vary from 1000 to 1200 kWh per year (space heating, water heating and cooking appliances were not included). 1.3. Aims of the project (1) Quantify the dependence of residential electrical appliance energy consumption on the outdoor atmospheric environment in Sydney. (2) Dene the most appropriate outdoor thermal climate index affecting energy consumption for residential space heating and cooling. (3) Empirically dene appliance usage threshold temperature and whether they hold true to the current heating and cooling degreeday base temperatures.

Airports inland zone, giving a sample of 136 households in total for the present study. 2.2. Calculation of degreedays Half-hourly outdoor temperature data from two automatic weather stations within the Greater Sydney metropolitan region were used to calculate degreedays for each of the 568 days of the study period. The stations were operated by the Australian Bureau of Meteorology and were selected to represent the distinctive coastal and inland climatic regions within the city. A base temperature of 18 C was used in the following project, as in some earlier Australian research [20], although there is no universal agreement on this. A common approach in the energy sector has been to use 18 C for heating degreedays and 24 C for cooling degreedays [21]. The presumption behind this practice is that there is negligible energy consumed for domestic heating or cooling purposes between mean daily temperatures of 18 and 24 C. An empirical resolution of the question would be to observe the mean daily temperature associated with minimum heating and/or cooling energy consumption, but to date there has been no such analysis in Australia. However, results from this research will show that both minimum heating and cooling energy consumption occurs at a temperature of 18 C, hence this seems to be the most appropriate base temperature for both heating and cooling degreedays in Sydney. In this study degreedays were calculated from degree half-hours in order to match the temporal resolution of the plug-load data. Initial pilot analyses were conducted at this 30 min temporal resolution, including correlation and regression models of average half-hourly energy consumption calculated across all households. However, the signal-to-noise ratio at that scale of analysis was too weak to produce any meaningful trends. Analysis was undertaken on a daily basis (hence the need for degreeday calculations) with the exception of a selection of individual appliance case studies with an hourly temporal resolution. Degree half-hours were averaged to obtain a degreeday, as summarised in Eq. (1): degreedays = 1 48
48

(degree half hours)i

i=1

(1)

2. Methods 2.1. Study location Sydneys climate is mild, humid subtropical. The greater metropolitan Sydney region was split into two zones, one coastal and another inland, each found to be climatically homogeneous [19], and well represented by one of two automatic weather stations operated by the Australian Bureau of Meteorology (Sydney and Bankstown Airports). Sixty-three households were located within the Sydney Airport AWS coastal climatic zone, and a further 73 within Bankstown

where (degree half-hours)i = Touti 18 C, i = 1, 2, . . . , 48, and where Touti is the average ith half-hourly outside temperature in degrees Celsius. Since the same base temperature of 18 C is used in summer and winter throughout this report, a negative degreeday in our analyses denotes a heating degreeday and a positive degreeday, a cooling degreeday. This sign convention centred on 18 C has the advantage of allowing heating and cooling season results to be plotted on the same graph. Degreedays and degree half-hours were also calculated using composite thermal comfort indices, namely standard

164

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

effective temperature (SET ) and effective temperature (ET ) in place of Touti in Eq. (1) above. SET and ET are indices used widely in the heating and air-conditioning industry to combine the effects of temperature, thermal radiation, humidity, wind speed, clothing insulation and metabolic rate on human thermal comfort [22]. ET primarily takes into account the effects of air temperature, thermal radiation and humidity on human thermal comfort, with the remaining variables in the comfort equation set as constants. SET extends the index to include the effects of clothing insulation, metabolic rate and wind speed, but with clo values and met rates set to constant values in the present study (0.6clo and 1.2mets, respectively). Wind data collected from the Bureau of Meteorologys Automatic Weather Stations were measured from an anemometer placed on a 10 m mast in accordance with standard World Meteorological Organisation guidelines. Wind speed in the boundary layer is largely controlled by the momentum exchange (frictional drag) imposed by the surface roughness, causing a decrease in mean wind speed closer to the surface. To take into account wind speeds that will be experienced by people, these wind speeds were downscaled using the power law [23].

3. Results 3.1. Space heating and cooling Thirty-two percent of NSW households currently own air-conditioners, either reverse cycle (heat pump) or cooling only. Most households have some type of space heating device with 42% having electric room heaters (22% of households have gas heating, 15% wood and 4% oil heaters) [24]. 3.1.1. Air-conditioners Daily average energy consumption was calculated across all houses within each of the two Sydney climatic zones, for each of the 48 half-hourly time steps. Sample-wide averages (across all households) for energy consumption (Wh per day) were modelled in relation to degreedays and thermal comfort indices SET degreedays and ET degreedays. Fig. 1 presents the relationship between degreedays and energy consumption over the entire study period, split by season. The relationship is stronger in the cooling season (Fig. 1c) when 56% of the variance in day-to-day energy consumption was explained by degreedays, compared to only 35% in the heating season (Fig. 1b). The choice between straight-line,

Fig. 1. The relationship between air-conditioner average daily energy consumption in Wh per day on the y-axis and degreedays on the x-axis for (a) the entire year, (b) the heating season and (c) the cooling season (empirically dened by the average daily temperature in relation to 18 C). Regression models (solid curves) and 95% condence intervals (dashed curves) were tted with second-order polynomials. Energy consumption was averaged over 47 households in the cooling season and 41 in the heating season.

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

165

quadratic and exponential regression models was made simply on the basis of maximising the variance in energy data being explained by degreedays (maximising R2 criterion). The regression analysis of all-year energy consumption in Fig. 1a produced a parabolic equation with a turning point at 0.25 C degreedays (the point at which the derivative of the function is zero). This point shows the degreeday value when neither heating nor cooling is required, or when they are least in demand. The SET thermal index combines the comfort effects of air and radiant temperatures, relative humidity, wind speed, clothing insulation and metabolic rate into a single variable. Since clo values and metabolic rates were held constant at 0.6 and 1.2, respectively, in this application, the SET index compared to ET mainly reects the impact of wind speeds being included in the calculations by the underlying two-node physiological model. In both ET and SET calculations we assumed a shade condition (i.e. ignored short- and long-wave radiation effects) by setting mean radiant temperature to equal air temperature. The amount of day-to-day variance in air-conditioner energy consumption explained by SET degreedays was marginally greater than with either simple air temperature or ET degreedays (combining

the effect of temperature and relative humidity into one variable). Forty-ve percent of the variance in air-conditioner energy consumption over both seasons was explained by SET degreedays, compared to only 41% by air temperature degreedays and 39% by ET degreedays. As with simple air temperature and ET degreedays, the cooling season produced the stronger relationship between appliance energy consumption and SET degreedays (R2 = 0.59) compared to 0.35 for the heating season. Probit analysis is a statistical technique that models the percentage of a sample responding to various levels of exposure to an environmental agent [25]. In the context of the present project the technique was used to t sigmoidal response functions between the daily temperature or degreeday (stimulus) and the percentage of sample households with their air-conditioners (either in heating or cooling mode) switched on. The decision to turn a heating or cooling system on or off is based on thermal discomfort; probit analysis has been widely used in thermal comfort research [26]. During the heating season probit regression between degreedays and the probability of appliances being switched on (Fig. 2a) produced a small chi-square test statistic for goodness of t and therefore a large P-value

Proportion of appliances switched on

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -30

95% Fiducial Limits Observed Model 50% threshold temperature = -7.2C degree-days

-25

-20

-15

-10

-5

10

15

20

(a)
1

Degree-days (C)

Proportion of appliances switched on

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -15 95% Fiducial Limits Observed Model 50% threshold temperature = 5.5C degree-days

-10

-5

10

15

20

25

(b)

Degree-days (C)

Fig. 2. Air-conditioner probit regression results between a continuous explanatory variable (degreedays) and a binary response variable (whether or not the appliance was switched on at any time during the day). Analysis is performed separately for (a) heating season and (b) cooling season.

166

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

(2 = 15.5, d.f . = 20, P = 0.8), indicating a good t by the probit model.1 The probit models 50% threshold temperature was 7.2 C degreedays (i.e. a mean daily temperature of 10.8 C), with narrow ducial limits from 7.6 to 6.8 C degreedays (ducial limits are to probit models what condence limits are to regular regression models). This 50% threshold average daily temperature (or degreeday) is the point at which 50% of occupants had their appliance switched on at some stage during the day and 50% had it switched off all day. This is also the point at which the greatest number of households changed their decision from off to on. On the coldest of Sydney days in the study period only 65% of households were observed to have their air-conditioners switched on for heating (Fig. 2a). Fig. 2b shows the results of cooling season analysis in which the probit model produced a large chi-square test statistic and a low P-value (2 = 34.2, d.f . = 20, P = 0), indicating there was a relatively weak match between the probit model and the data, but this was compensated for by widening the ducial limits. The 50% threshold temperature for cooling season analysis was 5.5 C degreedays with 95% ducial limits of 5.1 and 6.0 C degreedays. It should be noted that the implementation of the probit technique that we used [24] SAS was able to compensate for failed goodness of t tests by widening the ducial limits surrounding its estimate of the 50% threshold temperature. On the evidence presented in Fig. 2, the warmest days in the study period had up to 90% of households using their air-conditioner to cool. On a zero degreeday (i.e. mean daily temperature of 18 C) 20% of households still had their air-conditioners switched on, and the same of minimum-usage level was found in the heating season analyses as well. That is, a minimum of about 20% of households used their air-conditioners either in heating or cooling mode, all-year-round. On extending the weather sensitivity analysis by using the ET and SET thermal comfort indices instead of air temperature during the cooling season, the probit regression model of air-conditioner usage and SET degreedays produced the best results in terms of explained variance and maximised the goodness of t test (2 = 7.5, d.f . = 16, P = 0.9). The 50% threshold temperature came in at 3.3 C SET degreedays with tightly dened 95% ducial limits at 3.03.6 C SET degreedays. Load proles were calculated for the air-conditioners in the Sydney sample in order to more closely examine the diurnal variability of energy consumption. Mean hourly energy consumption and concurrent outdoor air temperature observations were averaged across all households for each hour of the day. Due to the two modes of reverse cycle air-conditioners (heating and cooling), separate load proles

were produced for heating and cooling seasons, i.e. averaging across all days in each of the two seasons. Fig. 3 shows the diurnal distribution (time uncorrected for daylight savings) of air-conditioner energy consumption and corresponding outdoor air temperature for the heating and cooling seasons (a) and (b), respectively. The salient feature of the seasonal comparison in Fig. 3 is the mean daily energy peak in winter is twice that for summer. During the cooling season (Fig. 3b) energy consumption begins to rise at 9 a.m., peaking at 4 p.m. and then rapidly decreasing to a minimum at 6 a.m., closely tracking the diurnal outdoor temperature cycle. Assuming the causal link between summer temperature and energy consumption extended to the heating season one might expect the winter diurnal load prole to be a mirror image of the summers, but that appears in Fig. 3a to not be the case. Winter heating energy consumption has two peaks: one at 8 a.m. and the other at 9 p.m., the periods at which houses are most likely to be occupied. During the heating season the daily minimum temperature occurs between 6 a.m. and 7 a.m., but this coincides with the time of minimum, not maximum heating energy consumption. Heating season load proles for room heaters produced a similar twin-peak pattern to that of air-conditioners during the heating season. The previous analyses involved aggregation and averaging across a sample of 47 air-conditioners in the cooling season and 41 during the heating season, thereby masking the intricacies of individual householder behaviour. In order to examine just what was lost in the averaging process we performed a case study on an individual household. Hourly probit analysis of this single reverse cycle air-conditioner indicates that usage during the cooling season conformed to the probit model quite closely (Fig. 4b), although the goodness of t test on the probit model suggests otherwise (2 = 143.9, d.f . = 19. P = 0). The 50% threshold temperature was 27.1 C, and at this daily mean temperature, the household in question had a 50:50 chance of having the cooler on at some stage during the day. In the heating season (Fig. 4a) the probability of the appliance being switched on increased with decreasing outdoor temperature, reaching a peak at 6 C but then decreasing with further reductions in outdoor air temperature. The chi-square statistic for the goodness of t test is highly signicant in the heating season (2 = 239.3, d.f . = 17, P = 0), conrming what is apparent in Fig. 4a, namely that the probit model is inappropriate for this set of data. A similar analysis using room heater data produced similar patterns to these air-conditioner results. 3.1.2. Room heaters For the purposes of this study room heaters are dened as portable electrical heating devices, usually used to heat only one room of a house at a time. Common examples in Sydney are the oil column radiator, the bar radiator and the fan-forced convective heater. Degreedays dened in terms of simple air temperature predicted room heater energy consumption quite well (Fig. 5), producing an R2 of 0.63. Room

In the context of this particular statistical method, failing the 2 goodness of t test at the 0.05 level indicates that the data were well approximated by the probit model.
1

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

1300 1200 1100 1000 900 800 700 600 500 400 300 200 100 0 0:00 25
Energy Co nsumptio n (W) T emperature (C)

167

Energy Consum ption (W)

20 Tem perature (C ) T e m p e r atu r e ( C )

15

10

0 2:00 4:00 6:00 8:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00 Time 25
Energy Co ns umptio n (W) Temper ature ( C)

(a)

(b)

1300 1200 1100 1000 900 800 700 600 500 400 300 200 100 0 0:00

En e r g y C o n s u m p tio n (W )

20

15

10

0 2:00 4:00 6:00 8:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00 T ime

Fig. 3. The diurnal distribution of mean hourly air-conditioner energy consumption and mean hourly outdoor temperature for (a) the heating season and (b) the cooling season. Energy consumption is averaged across 47 appliances for the cooling season and 41 appliances for the heating season.

heater energy consumption increases gradually with a decrease in degreedays below a value of +2.8 C degreedays (i.e. mean daily temperature of 20.8 C). Room heater probit analysis produced a signicant difference between observed data and the probit model (2 = 35.9, d.f . = 26, P = 0) indicating a sub-standard goodness of t between data and model. The 50% threshold temperature for room heaters was 7 C degreedays, with widened 95% ducial limits (8.1, 6.3 C degreedays) to compensate for the poor goodness of t. On the coldest of days observed during the entire study, barely two thirds (63%) of households had their electrical room heaters switched on (Fig. 6). 3.2. Refrigerators/freezers Refrigerator and freezer penetration rates in NSW currently stand at 1.3 and 0.4 appliances per household, respectively [24]. Refrigerator energy consumption has a positive, linear dependence on degreedays (Fig. 7a). There appears to be two distinct sets of data in Fig. 7a; a cloud of data points appearing just above the tted regression line and another just below it. There also appears to be a slight change in gradient moving from heating season (ve degreedays) to cooling season (+ve degreedays). According to the coefcients of determination (R2 ) in Fig. 7a, outdoor temperature bears a closer relationship to

refrigerator energy consumption during summer than in winter (R2 = 0.31 and 0.06, respectively). An increase of one degreeday in summer was associated with energy consumption increase of 95 Wh per day, whereas the same increase in winter increased energy consumption by barely half that amount (42 Wh per day on average). Freezer daily energy consumption also increases linearly with an increase in degreedays (Fig. 7b). The relationship between energy consumption and degreedays is quite strong with two thirds (67%) of the day-to-day variance in energy consumption being explained by the degreeday index. This relationship is stronger than that for refrigerators (R2 = 0.42). Fig. 7b depicts the relationship, split by season. As with refrigerators the relationship is stronger in summer (R2 = 0.51) than in winter (R2 = 0.36). In summer an increase of one degreeday was associated with an increase energy consumption of 75 Wh per day whereas in winter energy consumption only increased by 53 Wh per day per heating degreeday; in effect about 30% less weather sensitive than in summer. The single house case study in Fig. 8 explores the relationship between degreedays and refrigerator energy consumption, and includes the effects of space heating and cooling in the household. Energy consumption data for the households reverse cycle air-conditioner were used to dene the outdoor temperatures at which the occupants of this particular house began to heat or cool. Refrigerator energy consumption was

168
1

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

Probability of appliance being switched on

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -20

Model 95% Fiducial Limits Observed

-15

-10

-5

10

15

20

(a)
1 Probability of appliance being switched on 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 10 15 20

Temperature ( C)

Model 95% Fiducial Limits Observed 25 Temperature (C) 30 35 40

(b)

Fig. 4. Single-household case study of a reverse cycle air-conditioner. Hourly probit analysis examines the relationship between mean hourly temperature and the fraction of occurrences, for each temperature bin, of the appliance being switched on at anytime during each hour of analysis, during (a) the heating season and (b) the cooling season.

y = 28.275x 2 - 349.28x + 809.36 R2 = 0.63

Energy Consumption (Wh/day)

12000 10000 8000 6000 4000 2000 0 -12

-10

-8

-6

-4

-2

Degree-Days ( C)

Fig. 5. Average daily room heater energy consumption in Wh per day as functions of degreedays. Regression models (solid curve) and 95% condence intervals (dashed curve) were tted with second-order polynomials. Seventy-one room heaters (i.e. households) were included in the analysis.

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

1

169

Proportion of appliances switched on

0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 -30

Model 95% Fiducial Limits Observed

50% threshold temperature = 7C degree-days

-25

-20

-15

-10 -5 0 Degree-days (C)

10

15

20

Fig. 6. Room heater probit regression results between a continuous explanatory variable (degreedays) and binary response variable (whether or not the appliance was switched on at any time during the day).

split into three sections depending on whether the household was being articially heated, cooled, or free-running. On days below 2 C degreedays this particular household tended to heat, while on days above 4 C degreedays the household tended to cool. On days between the threshold temperatures of 2 and 4 C degreedays the house was assumed to be free-running, neither heated nor cooled. When the household was free-running, that is the building is not using energy for either heating, nor cooling [27], the relationship between refrigerator energy consumption

and outdoor temperature was strongest (R2 = 0.42) and the gradient (weather sensitivity) was greatest (Fig. 8). These points show a change in weather sensitivity of refrigerator energy consumption, which tends to plateau once the house is articially heated or cooled (Fig. 8). 3.3. Domestic hot-water systems Domestic hot-water analysis was performed separately for the different energy rates available to consumers (anytime

5000

Energy Consumption (Wh/day)

4000

Heating Season y = 42.259x + 2800.8 R2 = 0.06

3000

2000

1000

Cooling Season y = 95.443x + 2982.2 R2 = 0.31

-12

-10

-8

-6

-4

-2

10

12

(a)
5000

Degree-Days ( C)

Energy Consumption (Wh/day)

4000

3000

Heating Season y = 52.547x + 1807.1 R2 = 0.36

2000

1000

Cooling Season y = 75.254x + 1776.3 R2 = 0.51 -10 -8 -6 -4 -2 0 2 4 6 8 10 12

-12

(b)

Degree-Days ( C)

Fig. 7. The relationship between daily energy consumption and degreedays, split by season (dashed model tted to the heating season; solid model tted to the cooling season) for (a) refrigerators and (b) freezers. Ninety-three refrigerators and 39 freezers (i.e. households) were included in the analysis.

170
1800 Energy Consumption (Wh/day) 1600 1400 1200 1000 800 600 400 200 0 -10

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

-2C Degree-day Point at w hich household tends to heat

Cooling R2=0.23 4C Degree Day Point at w hich house begins to occasionally cool Neither R2=0.42 Heating R2=0.26

-8

-6

-4

-2

10

Degree-Days ( C)

Fig. 8. A single house case study of refrigerator energy consumption from a household located in coastal Sydney. Refrigerator energy consumption was split into three sections depending on whether the household was being articially heated, cooled, or free-running.

30000

Energy Consumption (Wh/day)

25000

20000

H eating Seaso n y = -139.39x + 8806.4 R 2 = 0.05

Co oling Season y = -326.59x + 8254.7 R 2 = 0.30

15000

10000

5000

H eating Seaso n Co oling Season

-12

-10

-8

-6

-4

-2

10

12

(A)
30000

Degree-days (C)

Energy Consumption (Wh/day)

25000

Heat ing Seas o n y = -261.25x + 12792 R 2 = 0.05

20000

Co oling Season y = -331.05x + 10838 R 2 = 0.17

15000

10000

5000

Heat ing Seas o n Co oling Season

-12

-10

-8

-6

-4

-2

10

12

(B)
30000

Degree-days ( C)

Energy Consumption (Wh/day)

25000

Heat ing Seas o n y = -182.85x + 9646.4 R 2 = 0.10

20000

Cool ing Season y = -293.96x + 8955.8 R 2 = 0.31

15000

10000

5000

Heat ing Seas o n Cool ing Season

-12

-10

-8

-6

-4

-2

10

12

(C)

Degree-Days (C)

Fig. 9. The relationship between hot-water energy consumption and degreedays, split by season, for (A) anytime, (B) off-peak and (C) night-rate systems.

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

171

heating, off-peak heating and night-rate heating). Anytime hot water refers to systems that heat water whenever required, regardless of tariff. Off-peak systems switch on the water heater between 10 p.m. and 8 a.m. and 11 a.m. and 4 p.m. Night-rate hot water is switched on between 12 a.m. and 7 a.m. As expected, for all three types of electric hot-water system the relationship with outdoor temperature was negatively linear, with an increase in degreeday (air temperature) causing a decrease in energy consumption (Fig. 9). The energy consumption of night-rate hot-water systems was affected by outdoor weather most directly (R2 = 0.48) (Fig. 9C), off-peak and anytime hot-water systems showing diminished weather sensitivity with R2 = 0.37 and 0.38, respectively (Fig. 9A and B). Energy consumption of all three types of domestic hot-water systems was affected more by outdoor weather during summer than in winter.

temperature above 18 C in Sydney, householders start using their coolers more intensively, and with an average daily temperature below 18 C, householders start making more use of their air-conditioners in heating mode. The non-zero energy consumption minimum in Fig. 1a indicates that some households in our sample were cooling their houses on days cooler than a mean of 17.75 C and heating their houses on days warmer than that base temperature. 4.2. Space heating and cooling Probit analysis of the likelihood that air-conditioners were switched on at anytime during the day, as a function of the degreeday index revealed that, during the heating season on the coldest of Sydney days, only 65% of households were switching on their air-conditioners in heating mode. However, during the cooling season, on the warmest of Sydney days, about 90% of households switched on their airconditioners. Those houses not using their air-conditioners during extreme days may not have been occupied. Unfortunately occupancy could not readily be diagnosed from the raw data collected in the RES. The database contained appliance energy audit data but nothing on occupant behaviour or demographics. The relatively large proportion of households choosing not to use their air-conditioners to heat on Sydneys coldest days may also indicate that they were using alternative forms of heating (i.e. not reverse cycle airconditioning). Many Sydney households have several space heating devices of various energy types and use them interchangeably and/or simultaneously, making it difcult to automatically monitor total heating demand [20]. For example, 55% of households monitored in the Residential Energy Study had at least one gas appliance, which could quite possibly include a portable, unued natural gas heater of the type that is still quite popular in New South Wales despite the indoor air quality problems posed by their combustion products. The relationship between outdoor weather and airconditioner energy consumption was consistently stronger in summer than in winter (R2 = 0.56 for summer, compared to 0.35 for winter). Reverse cycle air-conditioner load proles indicate that peak energy consumption occurs in the late afternoon during summer and in the evening during winter. Outdoor temperature minima occur in the early morning and maxima in the mid afternoon. Consequently, during the heating season at the coldest time of the day, occupants are sleeping and, as a result, energy consumption is minimal, the mild climate experienced in Sydney allows for these behaviours, where heating appliances are switched on only when occupant are home and awake. The lack of heating during this coldest time of the day undoubtedly weakened the statistical relationship between daily energy consumption and heating degreedays, reinforcing the importance of time-of-day in the prediction of space heating energy consumption. This point was conrmed by the single appliance case study. Probit analysis of the proportion of time the

4. Discussion 4.1. An appropriate degreeday base temperature for Sydney The severity of climate can be characterised concisely in terms of degreedays. The degreeday base temperature is generally regarded as the outdoor temperature at which neither articial heating nor cooling is required. Heating degreedays, or degreehours, calculated with respect to a base temperature of 18 C are widely used in Australia [20]. For cooling degreedays however, the base temperature is not so unanimously agreed. During the heating season the total heat loss coefcient of a building does not change as windows are closed and air exchange rate is fairly constant. However, during the cooling season heat gains can be partially regulated and the onset of articial cooling can be postponed by increasing ventilation rates, e.g., opening windows and doors to the exterior [28]. Often different base temperatures are used for cooling degreedays, depending on the building type and ventilation rate. This base temperature represents the transition point between comfort achieved purely through natural ventilation and comfort achieved by compressor-based cooling. Presumably, houses with poor or limited cross-ventilation potential make the transition from ventilation to refrigerated cooling at lower outdoor temperatures than houses with good climatic design. Unfortunately the design features and ventilation capabilities were not recorded for the houses in the present study, so a constant cooling degreeday base temperature of 18 C was applied across the entire sample. The results of reverse-cycle air-conditioner energy consumption versus degreedays in Fig. 1 indicate the parabolic minimum occurred at 0.25 C degreedays, so a degreeday base temperature of 18 C seems to be conrmed by these data from the Sydney context. The relationships for both seasons demonstrated that, at an average

172

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

appliance was switched on peaked at an outdoor temperature of 7 C, but the proportion of hours the appliance was switched on decreased sharply as temperature decreased below 7 C, presumably because the householders were asleep. Further reinforcing this interpretation is the observation that, during the cooling season (summer) when airconditioner load peaks in the afternoon (coinciding with the warmest time of the day), house occupants were more likely to be awake and therefore more likely to respond to the heat by turning on their air-conditioners, thus explaining why the statistical relationship between daily energy consumption and degreedays was stronger in summer than in winter. One of the key outputs from the probit regression tool is a 50% threshold temperature which can be interpreted as the temperature at which the largest proportion of households change their appliance from off to on. These results, along with load proles, and the direct relationships produced between energy consumption and outdoor weather may potentially be useful for power utility operations in predicting system spikes and peaks in the Sydney market. At the time of writing a commercially available service, called eWeather Online [29] provides eight-day weather (from CSIROs Division of Atmospheric Research) and electricity demand forecasts (from the National Electricity Market Management Companys (NEMMCO), Short-Term Projected Assessment of System Adequacy (STPASA) for wholesale electricity sellers and buyers. The information is used to forecast possible spikes in electricity demand, across all sectors [30]. The present paper presents a detailed analysis of the weather sensitivity of electricity end-use in the residential sector. Since the residential sector in general and space heating and cooling appliances in particular have a large impact on electricity peak loads [2], the weather sensitivity results reported here, along with appliance penetration rates, are potentially useful in relating systems peaks to individual end-uses in the residential sector. 4.3. Refrigerators/freezers Outdoor temperature has a stronger inuence on refrigerator energy consumption in summer than winter (R2 = 0.31 for summer, compared to only 0.06 for winter). One possible explanation may be the fact that the householders were heating the rooms in which the refrigerators were located during winter months. Modifying the interior temperature breaks, or at least weakens the relationship between outdoor temperature and refrigerator energy consumption, as demonstrated in the single-household case study in Fig. 8. At the time of the Residential Energy Study (1993/1994) Sydneys peak in electricity consumption typically occurred in winter [1], suggesting that households were choosing to articially heat more readily than cool, further explaining refrigerator/freezer energy consumptions relative independence from outdoor temperature. Current refrigerator test procedures for use in Minimum Energy Performance Standards (MEPS) and energy

efciency labelling schemes are performed using a single reference ambient temperature inside a climate chamber. This ambient temperature differs world wide, from 30 C in Korea, Japan and Chinese Taipei, to 32 C in USA, Australia and New Zealand. The International Organisation for Standardisation (ISO) species 25 C for temperate climates and 32 C for tropical climates [31]. In reality refrigerators operate under a range of ambient temperatures, and energy consumption and performance will vary under these non-steady-state conditions, as demonstrated in the results of this study. The aim of energy labelling is to encourage customers to purchase the appliance that uses least energy during actual use. A test procedure undertaken at a single temperature will not demonstrate the appliances performance under realistic ambient conditions. For example, the French Domestic Measurement End-use Campaign [16] discovered that in situ energy consumption in refrigerators is lower than the energy consumption measured under laboratory test conditions. The linear relationship between an average of ninety-three refrigerators daily energy consumption and degreedays in the present study could potentially inform the design of a more realistic and representative refrigerator test protocol for sub-tropical Australian conditions. Daily freezer energy consumption also increases linearly with an increase in degreedays. The relationship between energy consumption and degreedays is moderately strong with two thirds of the day-to-day variation in energy consumption being explained by concurrent degreedays. This relationship between freezer energy consumption and degreedays is stronger than that for refrigerators (R2 = 0.42 for refrigerators, 0.67 for freezers), possibly due to freezers not being accessed by the householders as often as refrigerators, resulting in fewer door openings (heat gains) and changes to food load. 4.4. Domestic hot water Domestic hot-water energy consumption showed a stronger relationship to outdoor temperature in summer than in winter (for anytime hot-water systems, R2 = 0.30 in summer and 0.05 in winter). Depending on the mode of the hot-water system between 30 and 48% of the day-to-day variance in energy consumption was explained by changes in outdoor weather. In the case of outdoor hot-water storage systems, an increase in outdoor temperature will decrease the amount of heat lost to the ambient environment. Some of the unexplained variance in hot-water energy consumption in this study could be due to the variety of installation locationsespecially indoors versus outdoors. Unfortunately, this hypothesis could not be investigated any further because the necessary information on installation location was not recorded in the original RES database [1]. Other factors, apart from weather, affecting hot-water energy consumption include the nature of other hot-water consuming devices such as dishwashers and washing machines, the

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174

173

nature (insulation status) of the distribution pipes, and occupant behaviour. The weather sensitivity of hot-water energy consumption may further be increased in summer due to occupants thermal comfort preferences. For example an increase in outdoor temperature may prompt occupants to prefer cooler showers that require less hot water than is the case during cooler weather.

radiant temperature equals air temperature (i.e. shade condition), it is feasible to explicitly calculate the impacts of shortand long-wave radiation on the human heat balance and to equate these impacts to the temperature of an isothermal enclosure that would be required to exert the same net radiation impact [32], i.e. mean radiant temperature. This avenue of further research would enable a quantitative assessment of the impacts of solar radiation variations on energy end-use.

5. Conclusions Acknowledgements Statistical models were established between outdoor weather and energy consumption for the following appliances: room air-conditioners, room heaters, refrigerators, stand-alone freezers and domestic hot-water systems. All appliances exhibited some form of weather sensitivity. The relationship between outdoor weather and individual appliance energy consumption was consistently found to be stronger in the cooling season than the heating season for all appliances included in the analysis. The thermal comfort index SET was found to be the most useful of predictor of space-cooling energy consumption, indicating that outdoor wind speed and relative humidity, as well as air temperature affect occupants thermal comfort during summer which, in turn, determines space-cooling demand. These relationships provide actual in-use energy consumption rather than oversimplied laboratory consumption, and this in turn may be used to design more realistic and representative appliance test protocols (often used in the implementation of MEPS). Probit regression was found to be a useful statistical technique in predicting the degreeday values at which households tend to heat and cool. Probit models of space heating and cooling appliance usage patterns can predict the probability of the appliances being switched on under various outdoor weather conditions. These relationships, along with load proles and the direct relationships between energy consumption and outdoor weather, have the potential to assist in the prediction of system spikes and peaks. By examining the mean daily temperature associated with minimum heating and cooling energy consumption for Sydney, a degreeday base temperature of 18 C was found to be the appropriate base temperature for the calculation of both heating and cooling degreedays. This report is the rst empirical conformation that 18 C is the most appropriate degreeday base temperature for Sydney, all-year-round. The inclusion of the following parameters in the original RES would have been useful in giving a more comprehensive picture of electricity consumption in Sydney: information about the type, location and age of the appliance being monitored, details of the building envelope, e.g., thermal performance, solar aspect, proportion of building shaded by vegetation, and socio-demographic details of occupants. Suggestions for future research arising from the present project include a more thorough application of the ET and SET indices of thermal climate to modelling outdoor thermal conditions. In particular, instead of assuming mean The authors would like to thank Tony Marker and Shane Holt at the Australian Greenhouse Ofce for presenting us with this research problem and facilitating our access to the NSW residential energy survey dataset. Lloyd Harrington of Energy Efcient Strategies is thanked for his critiques on the various drafts throughout this project. Observed weather data used in this project were purchased from the Australian Bureau of Meteorology.

References
[1] M. Camilleri, N. Isaacs, A. Pollard, J. Jowlett, Energy Used in Australian AppliancesAnalysis of 1993/1994 RES Appliance Energy Use Data, BRANZ, 2000, 59 pp. [2] R. Bartels, D.G. Fiebig, Residential end-use electricity demand: results from a designed experiment, The Energy Journal 21 (2000) 5181. [3] J.F. Busch, A tale of two populations: thermal comfort in airconditioned and naturally ventilated ofces in Thailand, Energy and Buildings 18 (1992) 235249. [4] N.H. Wong, H. Feriadi, P.Y. Lim, K.W. Thom, C. Sekhar, K.W. Cheong, Thermal comfort evaluation of naturally ventilated public housing in Singapore, Building and Environment 37 (12) (2002) 12671277. [5] M. Camilleri, N. Isaacs, A. Pollard, A. Stoeklien, J. Tries, T. Jamieson, Energy Use in New Zealand Households. Report on Aspects of Year Four of the Household Energy End-use Project (HEEP), BRANZ, Judgeford, 2000, 35 pp. [6] RECS, Residential Energy Consumption Survey, Energy Information Administration, USA, 1997, from http://www.eia.doe.gov/emeu/recs/ contents.html. [7] L. Lutzenhiser, A question of control: alternative patterns of aircondtioner usage, Energy and Buildings 18 (1992) 193200. [8] G.S. Brager, R.J. de Dear, Historical & cultural inuences on comfort expectations, The Cross Cultural Transfer of Environmental Building Information, International Research Workshop, University of British Columbia, 2002. [9] G. Prins, On condis and coolth, Energy and Buildings 18 (1992) 251258. [10] AGO, National Appliance and Equipment Energy Efciency Program, Future Directions 20022004, Australian Greenhouse Ofce, Canberra, 2001, 15 pp. [11] V.I. Ugursal, A.S. Fung, Impact of appliance efciency and fuel substitution on residential end-use energy consumption in Canada, Energy and Buildings 24 (1996) 137. [12] M.F. Fels, K. Kissock, M.A. Marean, C. Reynolds, Advancing the art of PRISM analysis, Home Energy Magazine Online, vol. July August, 1995.

174

M. Hart, R. de Dear / Energy and Buildings 36 (2004) 161174 [23] R.M. Aynsley, W. Melbourne, B.J. Vickery, Architectural Aerodynamics, Applied Science Publishers, London, 1977, 254 pp. [24] AGO, Australian Residential Building Sector Greenhouse Gas Emissions 19902010, Final Report 1999, Australian Greenhouse Ofce, Canberra, 1999, 121 pp. [25] SAS Institute, SAS Online Doc., Version 8, The SAS Institute, Cary, 1999. [26] E.R. Ballantyne, R.K. Hill, J.W. Spencer, Probit analysis of thermal sensation assessments, International Journal of Biometeorology 21 (1977) 2943. [27] M.A. Humphreys, The depenence of comfortable temperatures upon indoor and outdoor climates, in: K. Cena, J.A. Clark (Eds.), Bioengineering, Thermal Physiology and Comfort, Elsevier, Amsterdam, 1981, pp. 229250. [28] ASHRAE, ASHRAE Handbook: Fundamentals (SI edition), American Society of Heating, Refrigeration, and Air-Conditioning Engineers, Atlanta, 1993. [29] eWeather Online, vol. 2001, EWN Publishing, 2001, from http://www.erisk.net/ew/index.html. [30] B. Trewin, Specialised forecasts for the energy industry, Bulletin of the Australian Meteorological and Oceanographic Society 14 (2001) 84. [31] L. Harrington, A comparative assessment of refrigerator test methods, Energy Efcient Strategies, Melbourne, 2001, 13 pp. [32] K. Blazejczyk, H. Nilsson, Holmer, Solar loads on man, International Journal of Biometeorology 37 (1993) 125132.

[13] Pacic Gas and Electricity Company, Residential Energy Survey Report, Pacic Gas and Electricity Company, CA, 1994, 96 pp. [14] D.J. Sailor, J.R. Munoz, Sensitivity of electricity and natural gas consumption to climate in the USAmethodology and results for eight states, Energy 22 (1997) 987998. [15] A. Stoeklien, M. Bassett, ALF3a design tool for energy efcient houses, Thermal Simulation Software, BRANZ, Judgeford, 1999. [16] O. Sidler, A domestic electrical end-use measurement campaign in France, Commission of the European Community Programme SAVE, Dijon, 1997, 168 pp. [17] O. Sidler, Demand-side management: experimental study of highly efcient electrical appliances in household use, Commission of the European Community Programme SAVE, Dijon, 1998, 6 pp. [18] O. Sidler, J. Lemoine, B. Lebot, Key ndings from end-use metering campaigns in 400 European households: The EURECO Project, Commission of the European Community Programme SAVE, Dijon, 2002, 12 pp. [19] Bureau of Meteorology, Climatic Survey, Sydney, NSW, Australian Governmental Publishing Service, Canberra, 1991, 105 pp. [20] V. Badescu, E. Zamr, Degree days, degree hours and ambient temperature bin data from monthly-average temperatures (Romania), Energy Conversion and Management 40 (1999) 885900. [21] L. Harrington, Personal Communication, Sydney, 2001. [22] A.P. Gagge, A.P. Fobelets, L.G. Berglund, A standard predictive index of thermal response to the thermal environment, ASHRAE Transactions 92 (1986) 709731.