Applied Energy 88 (2011) 612621

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Methodology for characterising domestic electrical demand by usage categories

R.A.R. Kilpatrick *, P.F.G. Banll, D.P. Jenkins
School of the Built Environment, Heriot-Watt University, Edinburgh EH14 4AS, UK

a r t i c l e

i n f o

a b s t r a c t
Electricity consumption in the United Kingdom is continually growing with demand from the domestic sector a potential/major contribution to this increase in consumption. Although demand is increasing, little information exists on the domestic components that contribute to an increase in domestic energy consumption. Thus, a greater understanding on what is contributing to the increase in domestic energy usage is a pre-requisite to understand how it can be reduced in the future or, if not reduced, contained at its current level. This article discusses a separation lter designed for disaggregating domestic electrical demand data into different appliance categories. The lter is applied to a real time domestic electrical dataset spanning 1 year, and trends in standby, cold, heating element spikes and residual demand are identied. Several reasons to account for each of the trends are discussed. Additionally, the lter is applied to synthetic data both to conrm the accuracy of the separation lter and to nely adjust the lter for future application. The results indicate an increase in occupancy-related demand consumption during the winter months and an increase in cold consumption during the summer months. Furthermore, the results demonstrate that in contrast to changes observed in occupancy-related demand and cold consumption, there is little variation in standby and heating element spike consumption throughout the year. Finally, the potential advantage of incorporating a tailored separation lter into domestic smart meters is discussed. 2010 Elsevier Ltd. All rights reserved.

Article history: Received 15 October 2009 Received in revised form 11 June 2010 Accepted 2 August 2010 Available online 17 September 2010 Keywords: Domestic Electrical demand Power Appliances Energy Demand proles

1. Introduction Domestic energy accounts for a signicant proportion of total energy consumption both in industrialised and developing countries; estimates suggest that domestic consumption accounts for 31% of UK energy consumption [1]. Homes are where people spend a sizeable portion of their lives, where they eat, sleep and relax. It is therefore not surprising that domestic energy consumption results in a large impact on the national grid of industrialised and developing countries. According to Wood and Newborough [2], the UKs peak power requirement increased from 37.7 GW in 1968 to 50 GW in 2000. Wood argued that domestic energy use was the main driving factor for this rise in power usage. Furthermore, Balaras et al. estimated that building stock in the EU accounted for 40% of total energy consumption, with 63% of that value being associated with residential use [3]. In the UK, electricity consumption of a household can vary from 2000 to 6000 kW h/year [2]. The average domestic electrical energy consumption for a single household in Scotland suggested is to be 4792 kW h/year [1] and the average household consumption in Great Britain is 4628 kW h/year [1]. The UK is one of the top four most energy consuming countries within the EU, in terms of residential and tertiary energy use [3].
* Corresponding author. Tel.: +44 (0) 131 451 4637. E-mail address: rark4@hw.ac.uk (R.A.R. Kilpatrick). 0306-2619/$ - see front matter 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.apenergy.2010.08.002

Several studies have been undertaken to investigate domestic electricity use within the UK ([4,5] for example) but little information is given on how total domestic energy use is affected by the seasons or occupancy. In addition, interpretation of domestic usage data is complicated by the fact that several methods have been used to analyse energy demand data. Methods used have ranged from computer simulation and analysis programs to basic numeric calculation. To gain a better insight into electricity demand, it is essential to break energy usage into its contributing components. Few investigations have shown annual trends for domestic buildings, or the annual trends within each of the appliance categories. This lack of basic information makes it difcult to advocate practical measures to reduce energy usage at the individual domestic level which could ultimately have a positive impact on national energy usage. This study had two primary aims; the rst was to design and develop a separation method that was capable of dividing a total energy demand prole into smaller individual energy categories. The second aim was to: (a) use the designed method on real-time data, (b) study the different energy categories, (c) determine how they changed throughout the year, with possible explanations to account for the observed changes. Although the described methodology will be applicable to other industrialised and developing countries this study will focus on the UK. A universal methodology

R.A.R. Kilpatrick et al. / Applied Energy 88 (2011) 612621

613

is important as it must be able to accommodate specic differences in factors that contribute to domestic energy consumption. Houses in other countries (particularly those with domestic air-conditioning) will have different electrical demand characteristics than those in the UK, where air-conditioning is not normally available in domestic dwellings. In addition, houses with electrical demands that have a weaker relationship with external temperature (i.e. buildings without electrical heating or electrical cooling) are likely to be more suitable for the approach chosen in this study.

1.1. Literature review Load prole analysis in necessary to understand why energy usage is increasing and where it is being used in the domestic setting. Total energy consumption gures are useful for billing data, but it does not give an insight into how electricity is used, nor how consumer behaviour inuences electrical use. To better understand electrical usage (and to analyse load proles), total electrical demands can be broken down into various categories of usage. Wood and Newborough [2] believed that domestic energy usage could be described in terms of predictable, moderately predictable and non-predictable usage. Firth et al. [4] believed energy usage could be broken down into standby, continuous, cold and active. It can be argued that both Wood et al. and Firth et al. definitions of energy use are quite similar. Continuous power is the result of appliances being left on throughout the day. Such appliances include alarm clocks and home alarm systems. Standby is very similar to continuous, but is generally associated with the low power mode of the appliance. Thus they consume electricity at a value between zero and the rated power of the appliance. Appliances in the standby category include TVs, Hi-Fis and other home entertainment systems. Cold power includes any appliance that uses a compressor refrigeration system, such as fridges and freezers. Active power is considered to result from appliances being turned on for a period of time. This active component is thought to represent the remainder of the electrical use and can be associated with appliances in operation such as cookers, showers, kettles, TVs and lights. In this study, we have considered, standby and continuous power as one standby component. Several methods exist for analysing and disaggregating demand data. One method is a software based system discussed by Farinaccio and Zmeureanu [6]. This approach is a top-down method that uses previous data or assumptions to derive energy proles for individual types of energy demand. The Farinaccio method has several stages to separate out energy data. It uses pre-existing knowledge, gained during a training phase, to create a hypothesis about the appliance or prole. The change in electric demand due to the appliance being turned on or off, the Herin initial signal, can be used to recognise appliances within the total load prole. This method effectively uses predetermined appliance signatures and attempts to identify them with signatures in the proles. A more statistical approach was discussed in Aydinalp et al. [7], and involved taking into account the dwelling, number of occupants, number of children and whether the dwelling is owned or rented. A calculative approach, using numerical calculations to determine the individual components of energy usage was described by Firth et al. [4]. This method involves multiplying the power usage with the time intervals to obtain the energy use. The total power consumed, PTOTAL, is determined by the summation of cold, active and continuous and standby (C&S) power components. If PTOTAL is dened, it is then possible to calculate the other power values contributing to PTOTAL. The rst calculation is determining the C&S power component. Firth assumes that the minimum value on the total load prole will represent the standby/ continuous component.

The cold appliance consumption is more complicated to dene, due to the nature of the cycling pattern in energy consumption. In this study we have assumed a constant minimum value, a constant maximum value, and that the pattern does not change throughout the day. Firth et al. [4] suggest that taking data from 1 am to 4 am will ensure an indicative cold power activity prole, as it is assumed that there will be no other activity than the refrigeration contributing to energy usage. It should be noted that this also assumes that there is no home security lighting or courtesy lighting that would affect the outcome of the monitoring period. The last step is to determine the active portion of the energy consumption. This portion accounts for the active appliances and the active standby appliances (TVs or DVD players). If the standby/continuous and the cold power components have been calculated, they can then be subtracted from the total power consumption to give an indication of active power. The denitions of power, energy and duration for each of the different power categories used in Firth et al.s investigation can be found in Table 1. Further research into domestic electricity consumption was carried out by Sidler [8]. As part of the ECODROME project, 20 households were studied over a 2 year period. Each plugged load and lighting system was monitored with plug in meters that sent data automatically to a storage PC. The system was designed to be a install and forget system that requires no input or attention from the house occupants. The study provided detailed electricity demand data of each households plugged load and lighting, that was able to create a total electricity demand prole. Constructing an algorithm that can disaggregate loads would have a clear application in the area of smart metering of domestic energy usage. There has been an increase in legislation across Europe, regarding the importance of smart metering [9] and how it should be used. In particular, it is often suggested that smart meters will be more effective if they provide the user with some graphical interface showing how much energy they are using [10]. The methodology described in this study aims not just to present the energy used by the occupant, but also to provide a visual method of displaying the type of energy being consumed by the occupant. It is suggested that information of this type indicating, for example, whether standby loads are too high or energy use during occupancy is excessive, will provide the user with an extra level of information on which they can modify their behaviour in relation to activities that consume energy. The methodology discussed in this paper aims to extend observations made in previous research. The methodology discussed in Farinaccio et al. aimed at breaking down recorded energy data into individual appliance usage, based on their unique energy signature. There is one important point highlighted in this work: the initial signal, used to identify appliances, can be confused by two or more appliances that have a similar initial signal. Equally the system can be confused with two appliances being turned on at the same time, a problem that is evident with any load recognition software [6]. It appears that this method is designed to separate demand data into the contributing appliances, rather than divide the demand data into standby, cold and active. The methodology and results discussed in Firth et al. [4] appeared to be more focused on producing an understanding of annual energy consumption. A calculative approach was used to determine annual active, standby and cold energy consumption, and could be used to determine daily standby, active and active energy consumption. It does not appear that this method could be applied to daily demand data to produce separate energy use proles. The methodology discussed in Sidler [8] was a different approach to the previously discussed research. Individual appliances were monitored to create demand proles. This method was based on a bottom-up approach rather than a top-down approach. The method discussed in [8] provided accurate demand

614

R.A.R. Kilpatrick et al. / Applied Energy 88 (2011) 612621

Table 1 Components of power in relation to power consumed, energy consumed and duration. Component Standby Cold Heating element Residual Power consumed (kW) Low Low High Medium Energy consumed (kW/h) Low Low Low High Duration (min) Long Long Short Long

total electrical demand of space heating or air-conditioning used in cooler or hotter climates. A basic overview of the separation methodology is shown in Fig. 1. 2.1. Standby removal The rst stage in the separation technique is removing the standby component. The standby component is recognised as having a value equal to that of the lowest point on the load prole. Standby Power use typically describes the power consumption of appliances when they are switched off or not providing their primary services [11]. In order to remove standby power from the demand data, the assumption is made that the standby power consumption remains constant throughout the prole. It is this constant value that is subtracted from the total demand prole. A simple algorithm can be written to identify the lowest value and subsequently remove it from the prole. Clearly, if the assumption that the minimum value is equal to the standby component is not correct, then this will have an impact on the following separation of the cold and active components. 2.2. Cold removal

data, but required a large number of sensors and could only process data that has been collected from the monitored house. Conversely, the disaggregation methodology discussed in this paper uses total demand data and breaks it down into appliance category usage. The discussed methodology may not provide detailed individual appliance usage, but it can be applied to older electrical demand data.

2. Methodology The method used in this investigation utilised several stages to separate standby, cold and active components from a total demand prole. This section discusses each of these stages in turn. It should be noted that the developed disaggregating methodology was based on UK domestic electricity data. Ideally, this discussed methodology can be applied to a variety of electric demand data, though further research would be needed into determine the inuences on

The second stage of separation is removing the cold component. Cold power is characterised by a cycling on/off pattern and is easily recognisable in the load prole, shown in Fig. 2. Following the removal of the standby component, the cold pattern should start from zero (C0), increase to a set point (C1) then decrease back to zero after a predetermined time (TN). This pattern then repeats

Fig. 1. Data ow diagram, showing each lter and output.

R.A.R. Kilpatrick et al. / Applied Energy 88 (2011) 612621

615

after another preset time (TF). The assumption that the pattern does not change throughout the period of the load prole was made. A similar method for removing the standby component is applied to the cold element removal. First, key parts of the pattern are identied and values recorded. These key points are as follows:     C0 (W) C1 (W) TN (min) TF (min)

An initial approach was to use a lter to remove any value in the prole that had a value equal to, or less than the on value, C1. In addition, the hypothesis is made that, if the load prole point was above the C1 value, then C1 value is subtracted from the load prole point. This approach has two disadvantages. The rst disadvantage is that the approach will remove any other pattern that is less than the C1 value, including power spikes and other valuable data. The second disadvantage is that the lter will also remove non-cold energy from the large spikes (those that last for 30 min or longer) hence resulting in misleading data. A more practical approach is to construct a cold prole spanning an entire day, then to subtract this prole from the total load prole. The cold prole, as dened in Section 1, can be constructed using the information gained from the previous analysis stage. In theory, if the cold pattern remains constant throughout the year, then a standardised cold prole can be constructed and applied to each single day. This concept would be viable for achieving generic cold appliance lter. Thus it is important to establish the variability in the cold power throughout the day. 2.3. Active removal The remaining component of the load prole is the active portion, as dened as Section 1. To further understand energy use, this

active portion can be broken down into two further categories; heating element spikes and residual demand. The heating element spikes are a result of an appliance that utilises a heating element being turned on, such as toasters, kettles, and electric ovens or hobs. These appliances cause large spikes (several kilowatts) and can last for several minutes. The residual demand is the remaining area of active portion once the heating element spikes have been removed. This component, usually the result of lighting and home entertainment, is very important as it can represent a large percentage of the Active energy consumption. The rst stage of separation of the active component is removing the heating element spikes. This was achieved by designing and implementing a ve stage lter. The multi-stage lter is based on the assumption that the majority of heating spikes only last for 5 min or under. The lter analyses the active portion of the demand data and subsequently removes any spikes that coincide with a predetermined criterion or condition. A spike factor (S.F) is dened, that compares power at any given time, Pt, with a point some other time, Pt+i, where i is the duration between these two points (normally 15 min). The S.F can be adjusted to increase or decrease the spike removal, with a high S.F value only removing the larger power spikes. The spike factor is represented in the equation below:

S:F

P t i Pt

The selected time intervals were i = 5, 4, 3, 2, and 1 min(s). With the heating elements removed, the remaining data can be classed as the residual demand. No further separation was carried out on this component as it tends to be a relatively smooth prole resulting from several different appliance signatures (such as TVs, computers and lighting, as well as the electric pump associated with a boiler). The residual demand prole itself tends to be a good indicator of occupancy in the dwelling, so energy use in this prole can be considered to be occupancy-related.

Fig. 2. Example cold prole, showing key values used for data analysis (time axis in minutes:seconds).

616

R.A.R. Kilpatrick et al. / Applied Energy 88 (2011) 612621

3. Analysis To investigate the accuracy of the obtained results a comparison is needed. Ideally, data with known variables, i.e. known equipment usage, duration of usage and energy consumption corresponding to usage, are required as a comparison. The problem with using total electrical demand data is that it is very difcult to positively identify appliances from the prole. This can result in separation errors, causing inaccurate Standby, Cold and Active values. One solution is to use composite data, i.e. an electrical demand prole made up of known individual appliance proles, as an alternative to real-time data. The advantage of using composite data is that the load proles are constructed from the summation of known data. Although the summation of the data is articial, (albeit based on typical occupancies), the appliance signatures used in the prole are real-time appliance signatures [12]. The user sums the real-time appliance signatures over an entire day to construct the synthetic prole, as shown in Fig. 3. The end result is a load prole graph with known values for standby, cold and active portions. The constructed synthetic load prole was subjected to the separation lters described in Section 3. To determine how successful the separation was, the results from the lters were compared to the known variables, as discussed previously. It should be noted that one disadvantage of using the synthetic prole is that the standby portion of energy use is not a separate component. Instead it is part of each real time signature. This does not affect the separation technique, but will affect the comparison results. To overcome this, the standby component is removed, as per the methodology, but is then added to the residual demand. This coincided with how the synthetic data is combined. 3.1. Cold comparison Fig. 4 compares the actual cold prole from the synthetic (original) data with the cold prole predicted by the cold prole lter of Section 2.2. The two proles, though similar, are not an identical match. The proles appear to be matched up to 0700 h but after that the proles become unsynchronised. The likely reason for this is that the synthetic prole may change during the day resulting in a non-symmetric prole. If this is the case, the designed separation lter will not successfully remove the entire cold prole. To successfully remove the cold prole a ne tuning system needs to

be incorporated into the separation method. Each time the Cold lter is applied, the resulting active prole has to be checked to ensure the cold prole has been successfully removed. 3.2. Heating element spike comparison Fig. 5 compares the known heating element spikes (from kitchen use) from the synthetic data with those predicted by the ltering process of Section 2.3. The graph demonstrates that the majority of heating element spikes are removed by the lter. There are several spikes that are either not removed or not completely removed by the lter. Examples of these problems are shown at 0730 h, 0815 h, 1700 h and 2130 h. Initially the lter removed only 20% of the spikes, with the spike factor set at 2. With the spike factor adjusted to 1.15, the lter removed approximately 80% of the spikes. The problem with adjusting the spike factor to this value is that the lter then also removes unwanted features such as parts of the cold appliance signature. This is noticeable during the hours from 1000 h to 1530 h. However this only represents a small portion of the total demand prole and is considered an acceptable error margin. 3.3. Residual demand comparison Fig. 6 shows the residual demand for both the synthetic data and that following the use of the lter. The synthetic residual demand data is the summation of lighting power and home entertainment power. The ltered prole has a similar shape to the synthetic data, but there are several discrepancies with the prole matching. Several spikes do not match the synthetic prole. This could simply reect the fact that the spike lter has not successfully removed all the heating element spikes. In theory the spike factor could be adjusted to remove these additional spikes. Another problem is that from 0900 h to 1500 h the ltered prole appears to resemble a cold prole pattern. This could be due to errors in removing the cold prole in the earlier lter stage. Incorrect data may have been removed accidentally by the Cold lter resulting in the unmatched proles. 3.4. Synthetic test results The synthetic test is useful in determining how successful the discussed methodology is in separating load prole data. It can

Fig. 3. Example of synthetic data representing one 24 h period.

R.A.R. Kilpatrick et al. / Applied Energy 88 (2011) 612621

617

Fig. 4. Comparison of synthetic data cold prole (original), and synthetic ltered cold prole (lter).

4 3.5 3 2.5 2 1.5 1 0.5 0 00:00

Heating Element(Original) Heating Element(Filter)

Power (kW)

02:00

04:00

06:00

08:00

10:00 12:00 Time

14:00

16:00

18:00

20:00

22:00

Fig. 5. Comparison of synthetic heating element spikes (original) and heating element spikes from ltered synthetic data (lter).

be seen from Figs. 46 that the developed lters are reasonably but not totally effective in separating out the different power components. The results highlight that the disaggregation methodology is suitable; however the set-up parameters do require adjustment to achieve optimum results. The correct selection of the spike factor, and careful design of the cold proles, improves the accuracy of the separation technique. Additionally, the correct set-up parameters could be adjusted when real-time data is introduced. 4. Results and validation The separation method is applied to total electric demand data, with one-minutely temporal precision, for an entire year (data taken from Peacock and Newborough [13]) for a single dwelling in 1994, see Fig. 7 for an example of monthly power consumption. Little information is known about the building, i.e. size, age, occupancy, but this is not required for the separation (with the lters actually being designed for use with such a blind dataset). Due to time constraints, it is not possible to apply the lters to all 365 days. Instead, 8 days from each month (each Wednesday and Saturday in the week) are chosen, totalling 96 days and the results

scaled up for the entire year. It is important to select a weekend day because the energy prole can be signicantly different to a weekday. Generally, more energy is consumed at the weekend due to the occupants staying in the house for longer periods. The total annual consumption of the house is 5900 kW h, whereas the total calculated from the 96 design days (and extrapolated for the whole year) is 6060 kW h conrming that the chosen design days were suitable for extrapolation. The result of applying the separation algorithms for all design days is shown in Fig. 8, giving monthly breakdown by energy use category. For the results section, we will dene spring months as March, April and May, and summer months as June, July and August. Autumn months are also dened as September, October and November, and lastly winter months as December, January and February. 4.1. Standby results Fig. 8 shows that the average standby energy consumption varies slightly throughout the year (with a maximum of 5.1 kW h/day in August to a minimum of 2.7 kW h/day in February). If both May

618

R.A.R. Kilpatrick et al. / Applied Energy 88 (2011) 612621

Fig. 6. Comparison of synthetic residual demand (original) and residual demand from ltered synthetic data (ltered).

Fig. 7. Example of power demand for the studied dwelling over a 1 month period (July 1994).

and August months are removed, there is little change between the summer months and the winter months. In December the consumption was 4.12 kW h/day compared with Julys consumption which was 4.18 kW h/day. Previous research [8,11,14,15], investigated standby power consumption, with varying detail. From such work, it can be established that the average standby power consumption in the UK is 32 W, resulting in 277 kW h/year or alternatively 8% of total domestic electricity demand [14]. The average EU yearly standby energy consumption gure is 439 kW h/year [8], which equates to over 50 W of standby. A worse case scenario could see the standby at 81 W resulting in a consumption of 647 kW h/year [15]. The maximum standby power, minimum standby power and yearly standby energy consumption gures of the studied building were 212 W, 90 W and 1383 kW h/year, respectively. When compared with the EU average it would appear that the studied house has abnormally high standby power, due either to poor energy management or to having a large amount of appliances.

The marginal rise in standby energy during the summer months could be due to a variety of causes. The main reason could be the occupants watching more television during the winter months than the summer months. The standby power consumption of a TV will decrease when it is turned on because it is now identied as consuming active power. The other reason for the increase in Standby power during the summer months may be due to more appliances being used in the summer than in winter, perhaps related to outdoor activities. Furthermore, the standby energy can also contain the standby power associated with the refrigeration appliances which consume more energy in the summer due to higher external temperatures. 4.2. Cold results Fig. 8 also shows the cold prole energy consumption. The results reveal that there is a difference in consumption between the summer months and winter months (with the maximum

R.A.R. Kilpatrick et al. / Applied Energy 88 (2011) 612621

619

Fig. 8. Standby, cold, heating element and residual demand results for a single dwelling over a 12 month period.

consumption of 2.91 kW h/day in September and the minimum consumption of 2.01 kW h/day occurring in March). If September is removed, a slight rise in cold energy consumption can be seen during the summer months. The results show that the average consumption of cold energy during the spring months rises from 2 kW h/day to 2.5 kW h/day, the summer months see a drop in consumption from a peak of 2.8 kW h/day to 2.6 kW h/day. Autumn also sees a drop in cold energy consumption which peaked at 2.9 kW h/day and fell to 2.6 kW h/day. Lastly the cold energy consumed during the winter period remains at a fairly constant 2 kW h/day. This difference in consumption is likely to be due to the refrigeration devices working harder during the summer months. If there is a rise in ambient temperature, the fridge or freezer will require more power, either as a change in peak cycling power or a change in the time period of each cycle. Meier [16] discovered during researching refrigerator energy use that energy consumption of a fridge doubled when kitchen ambient temperature increased from 17 C to 28 C. Another reason for this difference between summer and winter may be due to another Cold appliance being used during the summer months. This appliance could be a wine cooler, beer fridge or an alternative cooling device. 4.3. Heating element spike results In the case of heating element spike consumption, the results illustrate that there is no apparent trend between this energy consumption and the varying seasons. The maximum heating element spike consumption of 3.88 kW h/day occurs in October and the minimum consumption of 2.03 kW h/day occurs in July. If the results are further analysed, and a trend line added, it is possible to discern a slight rise in energy consumption towards the winter months (with summer having an average of 2.29 kW h/day and winter having an average of 2.48 kW h/day). It is difcult to explain this rise in demand purely as a function of appliance usage. As discussed before, the main appliances that use a heating element are kettles, showers, toasters and electric cookers or hobs, as well as appliances that heat water, such as washing machines.

One reason for the rise in this energy use during winter could be that the electric cooker is working harder in winter than in summer. If the ambient temperature was low, the cooker would use more power to maintain the desired temperature. Alternatively, the cooker could be used more during winter because of more holiday meals being cooked (e.g. Christmas and New Year period). Another reason for a winter rise in consumption could be due to the residents boiling a kettle more often i.e. having more cups of tea/coffee during the winter months, or even increased use of electric heaters. However, the seasonal difference appears to be very small. 4.4. Residual demand results Residual demand energy consumption is seen to drop during the summer months and then rise during the winter months. The peak value of 12.26 kW h/day occurs in December and the minimum value of 4.95 kW h/day occurs in March. This rise in winter can be seen further if we look at the individual seasonal residual energy consumption. The spring sees an average consumption of 5.6 kW h/day whereas the summer has an average consumption of 6.3 kW h/day. The rise in energy consumption is apparent between autumn and winter months with averages of 9.08 kW h/ day and 10.24 kW h/day, respectively. From this information, it is suggested that the trend in residual demand inuences the trend in active power consumption. The trend in residual energy consumption seen in Fig. 8 could simply relate to the increase of lighting. As it becomes darker earlier in the winter, lighting will be used earlier in the day and be on for longer when compared with the summer. Lighting can account for 10% [15], 12% [15] and 20% [5,17] of total electricity demand depending on energy category. This equates to 605.9 kW h/year, 727 kW h/year and 1211 kW h/year, respectively in the studied house. If this is compared with the average UK of 715 kW h/year [18], it falls between 10% and 12% lighting ranges. Thus any increase in the use of lighting above normal will have a signicant impact on the total Active consumption for that household. With

620

R.A.R. Kilpatrick et al. / Applied Energy 88 (2011) 612621

the information unavailable, it is difcult to accurately determine if the rise in energy consumption during winter months is solely due to the increased usage of lighting, but it is reasonable to suggest that this is a major factor. If the monitoring techniques used in Sidler [8] were applied instead of the separation lter, the assumption that lighting is a major factor in increasing residual demand could be answered. However, as discussed in Section 2, this would require a longer research time, and the added cost of buying large quantities of monitoring equipment. Another reason for the increase in power during the winter months is the use of home entertainment systems such as televisions, PCs, computer consoles, HI-FIs and DVD players. It is likely that residents will stay inside more during the winter months due to, for example, bad weather conditions, increasing their use of home entertainment systems. This conrms the previous asserted point that the residual demand prole is essentially an occupancy driven prole. The rise in energy use during the winter could again be a result of space heaters being used (albeit smaller systems that do not show up as large power spikes). Based on the available information, it will be assumed that the Active component of the load prole is purely the summation of lighting power and home entertainment. 4.5. Overview of annual consumption The energy consumed in each energy component for the selected dwelling over the year is shown in Table 2. From this diagram it can be assumed that the studied dwelling has relatively high energy consumption. According to Firth et al. [4], the high, medium and low consumption classications can be determined by the standby, cold and active percentages (in relation to total energy consumption). Table 2 demonstrates the three energy consumption groups and the corresponding component percentages. Additionally, the standby, cold and active results from this investigation can also be found in Table 2. From this table, it can be conrmed that the studied house falls within the high energy group. A second conrmation that the studied dwelling is a High Energy Group building is that the total energy consumption of 4841 kW h/year is around 1300 kW h less than that of the total found in this investigation. 5. Discussion The aim of this investigation was to create a usable and generic separation technique that when applied to domestic load data would identify and separate individual appliance signatures that contribute to total load. To accomplish this objective the following specications were dened for the developed lters that would be used on total energy data. The lters would: (a) be able to separate standby, cold and active components of energy use, (b) be applica-

Table 2 Energy group classication and criteria [4] plus results from this study in bold. Annual consumption (kW h) Total All dwellings [4] Low energy group [4] Medium energy group [4] High energy group [4] Results from this study 3100 (100%) 1770 (100%) 2689 (100%) 4841 (100%) 6059 (100%) Standby and continuous 601 (19%) 297 (17%) 402 (25%) 1104 (23%) 1383 (23%) Cold appliance 620 (20%) 535 (30%) 577 (21%) 747 (15%) 906 (15%) Active appliance 1879 (61%) 938 (53%) 1710 (64%) 2990 (62%) 3770 (62%)

ble for automatic application to any given load and (c) be versatile with regard to their application. Initial testing of the developed lters proved successful in separating out total domestic usage into individual components and achieved the desired outcome of identifying and isolating energy usage at the appliance level. In addition the developed lters were capable of disaggregating electricity usage into standby, cold, heating element spikes and residual demand components with a reasonable degree of accuracy. Application of the lters to the 96 days dataset, and when adjusted for an entire year, indicated that the heating component that contributed to total energy usage, was primarily related to domestic occupancy. Although active appliance usage was also related to occupancy, there did not appear to be a seasonal variation relationship with the heating component as it only varied from 2 kW h/day to 3.8 kW h/day throughout the year. In contrast although residual demand was also consistent with an occupancy prole, seasonal effects were apparent; consumption was greater in winter at 10. kW h/day compared with a summers average of 6.3 kW h/day. It has already been discussed in Section 4.5 that this change in energy use could be related to a greater requirement for lighting by the occupants due to it getting darker outside in winter or alternatively less lighting being required due to longer daylight in the summer months. The lters also highlighted that both the cold and standby components of total energy usage were only marginally affected by seasonal variation, with an increase from 3.4 kW h/ day (average) in winter to 4.28 kW h/day (average) in summer for the standby component of energy usage. The lters also identied a slight rise for cold appliance usage in summer, with an average of 2.7 kW h/day compared with the average winter consumption of 2.1 kW h/day. The standby, cold and active components identied with the developed lters suggest that when compared to the data published by Firth (see Table 2), the houses contributing to the dataset used in this study were in the high energy group. This view is further supported by the fact that the households in this study used in excess of 6000 kW h/year. An important specication of the lters was that they had to be versatile. In addition to being able to identify and separate standby, cold and active components of total energy usage, it was also a requirement that the lters should be able to monitor appliance energy usage at the level of the individual appliance type. Thus the ultimate goal of developing the lters was that they should be able to be applied retrospectively to historical data in addition to being applied in a real life situation in a domestic setting. Thus a key advantage of the developed lters is that they can be applied to actual measured energy consumption data that has been previously separated into appliance usage data, as described by [6,8]. This disaggregated appliance power usage data can be combined by energy use type, to form accurate standby, cold, heating element spike and residual demand data. This information generated once could then be compared to the results from applying the lter to the same data, similarly to the synthetic data test discussed in Section 3. The ultimate aim will be to extend the specications of the lters to achieve the goal to develop automatic separation lters that can be applied to real time energy usage. To achieve this objective further renement of the lters will be required. One development would be a detection-and-mapping system for the cold lter. If a load prole is recorded when there is least activity, ideally between 1 am and 4 am [4], the cold prole could be mapped and recreated. The recreated cold prole could then be removed from the total prole, allowing further separation of the other power components. A further required enhancement to the lters would be the ability to automatically adjust the spike factor frequently observed in the analysed dataset. Instead of the user manually adjusting the spike factor, the lter would modify the spike fac-

R.A.R. Kilpatrick et al. / Applied Energy 88 (2011) 612621

621

tor based on the output. Essentially the system would be based on trial and error to determine the optimum value. The last required enhancement to the current lter would be to extend the number of heating element spike removal lters. The current lter system has ve sub-lters allowing the removal of up to 5 min spikes. This limitation of the current lter system means that only spikes appearing within that time frame will be removed. Appliances such as electric cookers, showers and washing machines tend to have heating times longer than 5 min. Thus, the current spike removal lter would be unable to remove these spikes fully, leading to potential inaccurate results. One solution would be to increase the number of sub-lters from ve, to approximately thirty. With this number of sub-lters, projections suggest that the majority of heating element spikes associated with electric cookers/ovens would be removed. With further development and enhancements to the existing lters, the system has the potential to achieve the goal of a fully automatic separation lter system that can be applied to real time energy usage in a domestic setting. Life style changes can be made by the consumer and the impact of these changes is readily measurable. The direct effect of simply switching off TVs, lights and computers, or switching to energy efcient appliances and lighting has on their energy demand can be instantly available. The latest smart meters discussed by the Energy Saving Trust [10], have displays that range from simple LCD numerical displays, to a more complex colour bar graph system, that moves from red to green depending on the energy use (green representing low consumption and red representing high energy consumption). If the same output format shown in Fig. 8 were to be used, customers could see the bars increase or decrease as appliances are being turned on or off. Such a system, although still only providing basic energy consumption data to the customer, does make it possible to identify areas of energy wastage and then take steps towards more efcient usage of energy. In future, with the introduction of energy category recognition software, further domestic energy savings can be obtained. 6. Conclusions and recommendations A method of disaggregating short-time-resolution electricity consumption proles for dwellings has been developed. It allows a user to enter or copy and paste load prole data into a spreadsheet where a series of lters separates it into four components standby, cold, heating element spikes and residual demand. The system has the potential to be used as an automatic lter in domestic smart meters, which could give householders detailed information not only on the total energy used in a period but also the appliances that are contributing to their total energy consumption and costs. This information would allow the impact of changes in behaviour to be immediately visible to the householder. Applied to a real data set collected from a single dwelling over 12 months the separation lter shows clear trends in the four components, with an increase in occupancy-related electricity consumption during the winter and an increase in cold-related

consumption in summer. On the other hand there is little variation in standby and heating element spike consumption over the year. Since demand for electricity will continue rising unless drastic measures are taken, the ability to identify the individual contributors to total demand allows consumers to recognise and then control their consumption. Therefore it is strongly recommended that research effort be devoted to developing fully automatic pattern recognition software that could be integrated into the next generation of smart meters. Acknowledgement The authors would like to thank David Kane for providing the appliance proles as used in the synthetic data test. References
[1] Energy consumption in the United Kingdom. Ofcial Governmental Document, courtesy of the Department of Trade and Industry (National Statistics); 2008. [2] Wood G, Newborough M. Dynamic energy consumption indicators for domestic appliances: environment, behaviour and design. Energy Build 2002;35(8):82141. [3] Balaras CA, Gaglia AG, Georgopoulou E, Mirasgedis S, Saradis Y, Lalas DP. European residential buildings and empirical assessment of the Hellenic building stock, energy consumption, emissions and potential energy savings. Build Environ 2007;42:1298314. [4] Firth S, Lomas K, Wright A, Wall R. Identifying trends in the use of domestic appliances from household electricity consumption measurements. Energy Build 2007;40(5):92636. [5] Mansouri I, Newborough M, Probert D. Energy consumption in UK households: impact of domestic electrical appliances. Appl Energy 1996;54(3):21185. [6] Farinaccio L, Zmeureanu R. Using a pattern recognition approach to disaggregate the total electricity consumption in a house into the major end users. Energy Build 1999;30(3):245529. [7] Aydinalp M, Ugersal VI, Fung AS. Modelling of the appliance, lighting and space cooling energy consumptions in the residential sector using neural networks. Appl Energy 2002;71(2):87110. [8] Sidler O. DSM: major ndings of an end use metering campaign in 400 households of four European countries. In: European Council for an Energy Efcient Economy (ECEEE) summer study proceedings; 2003. [9] Directive 2006/32/EC of the European parliament and the council 5th April 2006. Energy end-use efciency and energy services and repealing Council Directive 93/76/EEC, EU Parliament Directive. [10] The smart way to display: a summary report on consumer preferences for energy display designs. Energy Saving Trust Publication; 2009. <http:// www.energysavingtrust.org.uk/corporate/Global-Data/Publications/Corporatepublications-folder/The-smart-way-to-display-full-report> [accessed 07.06.10]. [11] Meier A, Lin J, Liu J, Li T. Standby power use in Chinese Homes. Energy Build 2004;36:1221216. [12] Kane D, Newborough M. Estimating carbon savings for domestic base-load micro-CHP systems. World Renewable Energy Congress (WREC) IX, Florence; 2006. [13] Peacock A, Newborough M. Impact of micro CHP systems on domestic sector CO2 emissions. Appl Therm Eng 2005;25(1718):265376. [14] International Energy Agency (IEA). Things that go Blip in the night: standby power and how to limit it, courtesy of IEA Publications; 2001. <http:// www.iea.org/publications/free_new_Desc.asp?PUBS_ID=1110> [accessed 07.06.10]. [15] Bonneville E, Rialhe A. Demand side management for residential and commercial end users. Efciency Eco-Des 2006;146(3):28393. [16] Meier A. Refrigerator use in the laboratory and in the eld. Energy Build 1995;22:23343. [17] Stokes M, Rylatt M, Lomas K. A simple model of domestic lighting. Energy Build 2004;36:10316. [18] Richardson I, Thomson M, Ineld D, Delahunty A. Domestic lighting: a highresolution energy demand model. Energy Build 2009;41:78198.