Tracking energy signatures of British homes from 2020 to 2025

1. INTRODUCTION

Domestic buildings were responsible for 26% of UK energy demand in 2024 (DESNZ 2024), and energy for heating homes in winter is a significant element of peak demand (Watson et al. 2019). From an energy policy perspective, it is valuable to be able to characterise trends in energy demand across the whole building stock in order to track performance and emissions from the sector. The ‘in-use’ energy demand varies as a result of weather and changes in the building fabric, heating system and appliance efficiency, but also differs as a result of occupant behaviour.

An energy signature characterises energy use in a building by linking it to environmental parameters, typically relating energy demand to outdoor temperature (Bauwens et al. 2021), thereby normalising for changes in weather. In a temperate country such as the UK, the majority of energy demand is for space heating, with this demand being greatest in the coldest winter months. Power temperature gradient (PTG) is a well-established energy signature technique (CIBSE 2006; Fels 1986) based on a simple model of the relationship between energy demand and outdoor temperature. A piecewise linear model is fitted to the data with two regions: above a ‘balance’ temperature, T_b, consumption does not vary with outdoor temperature. At lower outdoor temperatures, the energy consumption increases linearly as the external temperature decreases. The equations for the energy signature based on energy demand and external temperature are:

where P_tot is the total power consumption, T_ext is the external temperature, T_b is the balance temperature, P_b is the baseline power consumption, and HPLC is the heating power loss coefficient (also referred to as the PTG). An example of fitting this equation to energy and temperature data is shown in Figure 1, which plots synthetic data randomly generated to fit typical patterns for a UK home.

Figure 1:

This paper describes energy signatures derived from the Smart Energy Research Lab (SERL) Observatory dataset of daily energy from 13,000 British homes (Elam et al. 2024) over a five-year period between April 2020 and March 2025.

This paper aims to do the following:

• To outline how the PTG model relates to the physical and behavioural aspects of energy demand.

• To discuss the assumptions, simplifications and limitations of the method and to provide guidance for those wishing to apply it to large datasets of daily energy use.

• To present and discuss changes over time in the PTG’s model parameters for a large group of British homes over five years.

• To outline the relevance of this energy signature measure for energy policy and the assessment of energy efficiency retrofits.

The paper is structured as follows. Section 2 reviews the literature on the use of the PTG method and similar models for assessing domestic energy consumption. Section 3 explains the derivation of the PTG energy signature equation from first principles. The assumptions that must hold for the model to map to the physical properties of the building are outlined and ways in which the behaviour of the occupants can affect the energy signature are described. Section 4 describes the SERL Observatory dataset and the data processing carried out to track the PTG parameters over time. The method for fitting the PTG model and two variants of the method (with fixed balance temperature and including solar gains) are described. Section 5 provides an assessment of variations of the PTG method applied to data for a single year and discusses the variation in model fit by building characteristics. Section 6 presents findings about the change in model parameters over a five-year period for a group of 3,899 homes (for which it was possible to derive energy signature parameters for the full period). Section 7 summarises the strengths and limitations of the method. Section 8 concludes by highlighting the benefits of the PTG method for policymakers and others in providing indicator parameters to track performance over time.

2. LITERATURE REVIEW

An important milestone in the history of the PTG energy signature method was the article by Fels (1986), which sets out a clear definition of this method (described as the Princeton scorekeeping method—PRISM). This paper (and the special issue that it introduces) describes several use cases for both small groups of homes receiving retrofits and for large sets of homes (drawing on utility aggregated data). This review traces the evolution of the method over time and describes how variants of the PTG model have been used with different levels of aggregation and over different time periods.

Fels explains that PRISM was developed to provide a standardised way of quantifying energy savings that result from housing retrofits and to assist companies which offer energy efficiency improvements. The method is based on two readily available data sources: weather data for temperature external to the building, and energy consumption from utility companies.

The PRISM model uses a least-squares iterative method to optimise both the value of the balance temperature and the gradient of the line at temperatures below the balance temperature when monthly gas consumption is plotted against outdoor temperature. In order to assess the impact of interventions, the model is used in predictive mode, to provide a counterfactual ‘no intervention’ baseline against which the actual performance following the intervention is assessed. Fels points out that the three variables describing the line fitted to the energy data have a physical foundation, which is further discussed in Section 3.

Fels defines degree-days as the difference between the outdoor temperature and the calculated balance temperature for that building, (T_b – T_ext). A practical simplification of the PTG method which has been widely used by energy professionals (CIBSE 2006) is to work with a fixed value for T_b (usually 15.5°C in the UK). This is frequently referred to as ‘degree-day correction’. In these cases, degree-days are calculated based on this fixed reference point, not using a variable T_b derived from the data (CIBSE 2006). Kennard et al. (2022) describe how building and national energy demands are often normalised by locally measured degree-days using a nationally defined base temperature. This ‘base temperature’ varies significantly between countries (Azevedo et al. 2015).

Hammarsten (1987: 98) defines the energy signature of a building as ‘a set of parameters that describe its energy performance’ and points out the value of energy signature methods to provide predictions of future energy demand. Hammerstein also states:

Hammarsten recommends a similar model to that described by Fels (1986), but includes internal temperature and a term for solar gain. Many researchers concur that the accuracy of the model is improved by using the indoor–outdoor temperature difference as the explanatory variable (Allinson et al. 2022; Chambers & Oreszczyn 2019). However, suitable internal temperature data are rarely available unless sensors have been supplied for a research project (many homes are partially heated so temperature should ideally be monitored in each room).

Much investigation of energy demand in buildings in the decades since Fels (1986) and Hammarsten (1987) has used similar methods. An example of the use of energy signature analysis reported by Belussi & Danza (2012) is typical of historical studies in that data for a relatively short period for a single or small group of buildings are analysed: in this case, a single apartment in Italy. The analysis, based on weekly gas readings for short periods in winter before and after a refurbishment, only considers the sloping part of the PTG curve. The results show a ‘convex’ line of power use against external temperature, leading to the suggestion that the heating system lacked the capacity to maintain the internal temperature at low outdoor temperatures.

Several papers investigating energy demand in UK homes have suggested different models to characterise energy demand. In their investigation of quarterly UK energy data over a 10-year period, Summerfield et al. (2010) use an equation which is quadratic in temperature and also includes a price term. The introduction of a price term in this model recognises an effect not considered by Fels (1986): that energy demand is likely to be impacted by costs, with occupants making more efforts to save energy when prices rise.

An example of the use of similar models to track changes in overall energy demand for the building stock over time is the analysis by Elwell et al. (2015). They investigate quarterly energy demand for the UK domestic building stock over eight years to assess the uptake of more efficient condensing boilers (these were mandated as replacement boilers from 2005). Four models were fitted to the data, all of which include a constant multiplied by a (T_b – T_ext) term, but all also add separate terms for linear dependency of energy use on gas price, with some models also including extra terms for hot water. Elwell et al. report a specific heat loss of 215 ± 10 W/°C and balance temperatures from 14.77 ± 0.3°C in 1998 to 13.77 ± 0.5°C in 2013.

Summerfield et al. (2015) apply the PRISM technique and refer to the gradient of the sloped section of the curve as the PTG. The method is applied to daily energy and temperature data from 567 UK dwellings, and the difference in PTG depending on dwelling type, age and number of bedrooms was investigated using multivariable regression. A line is fitted to energy readings when the external temperature is between 0 and 15°C, effectively assuming that T_b ≥ 15°C.

Both Summerfield et al. (2015) and Elwell et al. (2015) acknowledge the behavioural aspect of the energy signature. The residents’ choices of how to operate the heating system (thermostat setting, operating hours and thermostatic radiator valve settings) and how to ventilate their home can all impact the internal temperature and hence energy signature (Gram-Hanssen 2011; Huebner et al. 2015; Nicol et al. 2012). Hollick et al. (2024) use the PTG energy signature method to investigate changes in energy demand in around 1,000 British homes during the COVID-19 pandemic and highlight changes in both the HPLC, base load and balance temperature parameters during the lockdown periods. The energy and temperature data are optimised to the PTG model using Bayesian techniques: this has the advantage of allowing additional quantification of uncertainty compared with a frequentist approach, but is computationally more demanding.

As smart meter data become more widely available, there has been increasing interest in how to characterise the physical properties of homes remotely, taking advantage of granular energy data. In the UK, this is often discussed under the acronym SMETER (smart energy thermal efficiency ratings) (Allinson et al. 2022). The objective is usually to obtain a rating based on heat transfer that reflects the energy performance of the building fabric. Variations in the way the heating and ventilation are operated can be accounted for if internal temperature measurements are available from the building. For example, Chambers & Oreszczyn (2019: 443) describe a method that adapts the PTG technique to estimate fabric properties ‘independently of occupant thermal behaviour’. Where internal temperature measurements are available, these are included in the calculations, and where they are not, the internal temperature was assumed to have a linear relationship with external temperature below T_b. In order to minimise the impact of variations in solar radiation:

• Days when the temperature was below T_b but there was high solar gain were removed in order to minimise impact of solar gains on the HPLC slope calculation.

• To calculate P_b, days with low solar gain were selected.

The results from this method applied to 654 UK dwellings showed a good match to those based on the nationally representative Cambridge Housing Model results based on extensive building surveys.

The GHG SMETER Project report (Hollick et al. 2025) describes the application of several SMETER methods to assess the efficacy of energy efficiency improvements funded through grant schemes. For the particular sample of homes analysed, two methods not requiring internal temperatures were found to have a ‘plausibility rate’ of 54% of homes with credible heat transfer coefficients (HTCs). Hollick et al. point out that there was no ‘ground truth’ assessment against which to compare the result, so an assessment of whether the calculated value was plausible was specific to each method, e.g. on statistical goodness-of-fit tests with empirically determined thresholds.

SMETER methods aim to allow the quantification of changes in fabric performance following energy efficiency interventions. In-use energy signature techniques, such as the PTG method described here, enable a different assessment: the change in demand that results from a combination of changes in both fabric and occupant behaviour can be investigated. This is of value for policy and research which considers the real-world impact of interventions in occupied homes.

3. THE PTG MODEL

The energy demand during the heating season can be stated as:

where P_tot is the total power consumption, P_h is the heating power consumption and P_b is the baseline power consumption for non-heating appliances.

The flow of heat through the building envelope (walls, windows, floors and roof), Φ_h, which needs to be balanced by heating in order to maintain an internal temperature, T_int, when the outside temperature is T_ext, can be expressed as a function of the HTC and temperature difference:

(This equation assumes that heat loss due to infiltration and ventilation is also proportional to the temperature difference between outside and inside.)

The relationship between heat consumption and heating power is:

where η_hs is the efficiency of the heating system and F is the ‘free heat’ provided by solar gains and the occupants, and heat gain from electrical appliances.

The balance temperature, T_b, is defined as the lowest external temperature at which P_h = 0 (i.e.. the external temperature below which the heating would be turned on) and can be derived from equations (3) and (4):

The HPLC, combining heating system and fabric efficiency, is defined as (Chambers & Oreszczyn 2019):

Based on the assumption that the mean internal temperature and free energy gains are constant over the calculation period, combing equations (3) and (4) to derive P_h followed by substitution into equation (2) gives:

Substituting this into equation (5) gives the heating season equation introduced in Section 1:

The above equations do not include a separate term for solar gains. If the heating provided from solar gains, defined as a constant g (sometimes defined as the ‘solar aperture’) multiplied by (variable) solar insolation I, is included in the energy balance, equation (2) becomes:

In equation (4), solar gains are assumed to be constant. The free heat F in equation (4) is made up of the solar heat gains gI and other gains (from occupants, etc.) F_o. Taking into account variable solar heat gains, equation (4) then becomes:

If the solar balance temperature, T_b(sol), is defined as the lowest temperature at which P_h = 0 when solar gains are 0, then from equations (3) and (8):

This balance temperature is the temperature at the breakpoint in the piecewise linear model. Unlike T_b in equation (5), it is not the external temperature below which the heating is required, but the temperature at which it would have been switched on if there were absolutely no solar gains.

This can be used to derive a version of equation (1), which includes variation in solar gains:

Since F includes the solar gain while F_o does not, F > F_o and T_b(sol) > T_b.

4. METHODS

This analysis is based on data from the Smart Energy Research Lab (SERL) Observatory (SERL 2025). This UK research council-funded project has been collecting smart meter data from British households since 2019. Daily and half-hourly gas and electricity data for more than 13,000 households are brought together with Energy Performance Certificate (EPC), weather and location data. Each household that joined the SERL Observatory completed a survey which includes information on built form, number of occupants, heating type and other information. A further survey, sent out in January 2023, included a question on household income and received over 5000 responses (Huebner et al. 2023). Webborn et al. (2021) discuss the representativeness of the SERL Observatory in detail. In summary, there is good regional coverage of Great Britain, but owner-occupiers are over-represented (80.3% compared with 64.6% reported in the English Housing Survey 2019/20) and renters are correspondingly under-represented.

The initial processing step is to create a dataset combining daily electricity and gas consumption with daily mean temperature for the home. Days with incomplete energy data are not included. For analysis of change over time, the data are broken down into 12-month periods (1 April–31 March) starting on 1 April 2020. Homes known (from survey responses) to use unmetered energy (e.g. oil heating) and homes known to have gas heating but with no gas meter readings available are not included in the dataset.

The single-year analysis was carried out on data for the year starting on 1 April 2023. This year was chosen as earlier years were affected by the COVID pandemic and steep rises in energy prices following the invasion of Ukraine in February 2022, so were more likely to have sharp variations in energy demand during the year (Zapata-Webborn et al. 2023).

The Python scipy.optimize.curve_fit function was used to fit a regression model to optimise the equation parameters for each individual home in each annual period. The code can be used to fit either gas data only, or total metered energy (gas plus electricity). Three model options were available (Table 1). Upper and lower limits were set for the parameters (Table 2).

Many different parameters can be used to assess confidence in model fit. The parameter reported here is R² (the square of Pearson’s coefficient) for the difference between actual data and model results for both sections of the piecewise PTG model.

The chosen model was fitted twice for each dataset. Based on the results of the first model run (labelled ‘pre-filter’), days when T_ext < T_b – 1.5°C and P_tot < P_b were identified. The plot of data for many homes includes a ‘cloud’ of points around T_b (Figure 5) and the selection of 1.5°C is a pragmatic decision to remove the days that most clearly seem to be unoccupied while leaving data points close to T_b as inputs for the optimisation algorithm. The data for these days, interpreted as ‘unoccupied’ since there was no heating on a day with mean temperature below the balance point (and allowing for error in the T_b estimate), was then removed before the second model run, designated ‘post-filter’. Homes for which R² < 0.1 in the first run were removed before the second run since there was very low confidence in model fit. The results in the following sections are labelled to indicate the model variant using the format model: fuel: run, e.g. ‘PTG_fixed_T_b: total: post-filter’.

Medians were calculated for parameters from each run. In order to comply with statistical data control procedures (ONS 2022), the median results reported are the mean of the 10 results closest to the median, which, given the large number of homes in each sample, differs very little from the actual median. Some example plots of data for individual homes are included below. In order to preserve anonymity, two steps were taken:

• A very small random increment was added to each energy data point. The value of this followed a normal distribution with a mean of zero and a standard deviation (SD) of 1% of the maximum energy demand. The same procedure was followed for the external temperature with an SD of 0.1°C.

• The power results were then divided by the maximum value, so that the chart is plotted with a power scale from 0 to 1.

5. ASSESSING THE PTG METHOD

This section describes results for fitting the PTG model to data from homes in the SERL Observatory for the year from 1 April 2023 to 31 March 2024.

5.1. COMPARING PRE- AND POST-OCCUPANCY FILTER MODEL RUNS

From the approximately 13,500 homes recruited to the SERL Observatory, 11,205 provided some energy data for the period. A total of 2,325 homes (21%) were not analysed as a small number did not have sufficient data points and the remainder had indicated in a survey that they used either unmetered heating fuels (such as oil) or the main heating fuel was not recorded, or the home had gas heating but no gas smart meter, leaving data from 8,800 homes for energy signature derivation. Total energy parameters were obtained for 6,935 homes from the post-filter run; in other words, the model did not converge or R² < 0.1 for 22% of the homes investigated.

An example from an individual home (Figure 2) illustrates a case (deliberately chosen from the most extreme examples) where the regression algorithm provides a poor fit to the PTG model. Daily total energy demand in winter is extremely variable at all external temperatures. This leads the algorithm to fit a curve with an implausible balance temperature of around 7°C, illustrating how some homes do not match the model assumption of linear dependence of power demand on external temperature during the heating season.

Figure 2:

The results for the initial, pre-filter run of the PTG_variable_T_b model and the second post-filter run for the same homes with days that appear to be unoccupied removed were compared. Table 3 shows the statistics for the results. As would be expected, the mean R² for the post-filter run is higher than for the pre-filter run, since the unoccupied days would not be expected to conform to a linear relationship between power used and external temperature during winter. The median HPLC and median P_b show little change, but the median balance temperature increases by 0.6°C. Figure 3 shows the distribution of T_b values from the two runs; Figure 4 shows the distribution of HPLC values for the post-filter run. The HPLC pre-filter results followed a very similar, non-normal distribution.

Table 3:

Statistics for PTG_variable_T_b model parameters for pre- and post-filter runs, 2023–24.

	MEDIAN HPLC (W/K)	STANDARD DEVIATION HPLC (W/K)	MEDIAN Tb (°C)	STANDARD DEVIATION Tb (°C)	MEDIAN Pb (W)	STANDARD DEVIATION Pb (W)	MEAN R2	STANDARD DEVIATION R2
Pre-filter	202	122	14.6	2.24	477	474	0.64	0.18
Post-filter	198	119	15.2	2.31	468	479	0.70	0.15

[i] Note: Results are for 6,935 homes with both runs; and for total energy.

Figure 3:

Figure 4

The results are sensitive to the offset included in the filter; the reported analysis filtered out days with low energy use at temperatures at least 1.5°C lower than the pre-filter-calculated balance point. When a different filter was applied, where all days with low energy use below the balance point (i.e.. no offset) were removed, the results for the post-filter run when compared with those shown in Table 3 gave a median HPLC that was 5% lower, a median baseload power 2% lower and a median balance temperature that was 0.6°C higher.

Figure 5 shows an example for a single home of a shift to a lower balance temperature when unoccupied days are removed. The effect of removing unoccupied days (shown in orange) in this case is that the calculated balance temperature decreases, as does the calculated HPLC.

Figure 5:

5.5 COMPARING THE RESULTS FOR DAILY, WEEKLY AND MONTHLY MEANS

In order to investigate the impact of different sampling periods on the PTG model parameters, the results for the total energy second model run using daily data were compared with results from the model using weekly and monthly mean power and external temperature for the same homes. Table 7 shows that there were significant differences in the PTG model parameters obtained for different aggregation periods. For daily and weekly data, a minimum of 30 data points was required for each home, while 12 monthly points were required. The impact of only a small number of points to which to fit the model is evident in the significant differences in the resulting parameters compared with weekly and daily data.

Table 7:

2023–24 statistics for daily, weekly and monthly aggregated data, PTG_variable_T_b: total: post-filter for 6,212 homes.

SAMPLING PERIOD	MEDIAN HPLC (W/K)	STANDARD DEVIATION HPLC (W/K)	MEDIAN Tb (°C)	STANDARD DEVIATION Tb (°C)	MEDIAN Pb (W)	STANDARD DEVIATION Pb (W)	MEAN R2	STANDARD DEVIATION R2
Daily	197	117	15.5	1.94	454	348	0.71	0.13
Weekly	227	133	14.3	2.08	471	356	0.80	0.15
Monthly	295	176	12.8	2.29	537	391	0.86	0.06

[i] Note: HPLC = heating power loss coefficient.

Particularly marked is the much lower median T_b for monthly data. Figure 6 shows the distribution of the balance temperatures calculated for each aggregation period. Figure 7 shows the HPLC distribution. Figure 8 shows an example for the lines fitted for different periods for a single home. The benefit of having more points close to the balance temperature inflection point in order to determine this accurately is clear. Since several months are in the ‘shoulder season’ when there will be a mix of heating and non-heating days, and there are few months when average external temperatures are very low, the spread of points on either side of the break point is much lower when only monthly data are available.

Figure 6:

Figure 7:

Figure 8:

5.6 VARIATION IN MODEL PARAMETERS BY BUILDING CHARACTERISTIC

Statistics for model parameters for homes grouped by built form are shown in Table 8. This follows a typical UK building typology: detached houses are single-family homes not attached to any other; semi-detached (the most common type in the sample) are attached to one other home; and terrace homes (sometime referred to as row houses) are attached to homes on either side (end-of-terrace homes are included in this category). Flats (apartments) are the least common building type in the sample.

Table 8

Statistics for model parameters grouped by building characteristics, 2023–24, PTG_variable_T_b: total: post-filter.

BUILT FORM	N	MEDIAN HPLC (W/K)	MEDIAN Tb (°C)	MEDIAN Pb (W)	MEAN R2
Detached	1,877	262	15.5	572	0.73
Semi-detached	2,281	201	15.3	483	0.71
Terrace	1,834	177	15.3	446	0.69
Flat	725	90	15.6	297	0.60

[i] Note: HPLC = heating power loss coefficient.

When the fit of the model data is considered, R² is best for detached homes and lowest for flats. One hypothesis is that this could be due to the impact of variable heat flows to or from adjacent dwellings through party walls (and floors and ceilings in the case of flats), causing more variation for buildings connected to neighbours.

Median HPLC is highest for detached homes, as would be expected with their generally higher floor areas. Figure 9 shows HPLC normalised by floor area for different built forms: detached and semi-detached homes have the highest value normalised by floor area, and purpose-built flats the lowest. This is to be expected as detached and semi-detached have the highest ratio of external surface to floor area, so a higher rate of heat loss would be expected. The numbers for this chart are lower than those in Table 8 since floor area information was only available for a proportion of the homes.

Figure 9:

Normalised HPLC tends to reduce as build dates approach the present, as can be seen in Figure 8 (the numbers are lower than those in Table 8 as floor area information was not available for all homes) This result is expected as building standards have improved over time. UK Building Regulations did not include energy conservation elements until the 1970s, and the insulation levels required have been progressively tightened since then.

6. CHANGES IN ENERGY SIGNATURES OVER A FIVE-YEAR PERIOD

Parameters were calculated annually for homes without unmetered heat for which complete data were available over the five-year period from 2020 to 2025 (3,899 homes). Only homes with energy data for at least 50 days in summer (1 June–31 August) and winter (1 November–28 February) in all years are included in the multi-year analysis.

HPLC reduces from 2020/21 to 2022/23 and then increases slightly in the following two years. This shows that the initial downward trend cannot simply be caused by an overall improvement in fabric efficiency over time. Figure 10 shows the ‘energy price cap’ over the same period. This is the maximum British energy suppliers may charge for typical annual (domestic) demand. For most of the period in question, actual supply prices were at or close to the cap. It seems likely that the noticeable drop in HPLC to 2022 was related to efforts to save energy at a time when energy bills were rising sharply, a phenomenon also reported by Huebner et al. (2023) and Zapata-Webborn et al. (2024).

Figure 10:

Figure 11 shows the mean results for HPLC, T_b and P_b for both the PTG_variable_T_b and PTG_Solar models plotted over time. HPLC is lower and T_b is higher for the model including solar gains for all five years (as discussed in Section 5.4 above). The calculation including solar gains allows differences in solar conditions from one year to the next to be taken into account, but in fact the trends in all parameters, apart from balance temperature, are the same for both variations of the model.

Figure 11:

Trends in baseload power P_b are also downward for the first three years, but means are more consistent from 2022/23 onwards. While it is likely that new and replacement appliances are more efficient than older ones, which would contribute to a downward trend in baseload power, this may be counterbalanced by more frequent usage or larger numbers of appliances.

Changes in T_b are likely the result of two effects working in opposite directions. If the year-round power P_b reduces, there are fewer free gains from equipment to contribute to heating, so T_b will increase. If the mean internal temperature reduces (because the residents run the heating for shorter times or at lower setpoints) while HPLC and P_b remain the same, then T_b will reduce. It is possible that the balance between these effects was different in 2021/22, a year of multiple COVID lockdowns which affected patterns of home occupancy.

6.1 CHANGE IN PARAMETERS OVER TIME BY INCOME LEVEL

Figure 12 shows HPLC normalised by floor area, grouped by household income and plotted over five years. The midpoints are based on banded survey answers (e.g. £85,000 is the midpoint of the £70,000–100,000 annual household income category). Households that indicated their income was more than £100,000 per annum are given a nominal midpoint of £110,000. The mean HPLCs for each group follow the general trend seen in Figure 10, with all income levels dropping to a lower HPLC in 2022/23, suggesting that energy-saving behaviours at all income levels were influenced by the steep rise in energy prices that year. The highest income groups have higher heat loss coefficients than lower income groups. This is likely associated with the increasing proportion of detached homes (with typically higher HPLC) as household income increases. A high proportion of the two lowest income groups live in social housing, which has higher energy efficiency standards than the rest of the building stock (and so lower HPLC) (DLUHC 2023). Social landlords have been incentivised to retrofit older properties with government grants. When building new homes social landlords have been more likely to exceed the minimum standard requirements.

Figure 12:

7. STRENGTHS AND LIMITATIONS OF THE PTG MODEL

Results from the analysis of a large sample of British homes in the SERL Observatory (SERL 2025) suggest that the PTG energy signature method does not produce reliable results for every dwelling; however, mean parameters for large groups of homes provide a consistent indicator of the energy performance of occupied buildings. Results grouped by built form show that the physical configuration of a house influences how well the PTG model fits, with detached houses having the best R². Including a solar term in the model is particularly relevant when examining change over time, as patterns of solar gains may vary significantly from one year to the next.

The technique relies on gathering sufficient energy data over a range of external temperatures, both during and outside the heating season, in order to determine accurately the balance temperature below which heating is switched on. More confidence can be placed in results based on daily data (as long as these include sufficient data during and outside the heating season) than in monthly averaged demand or other data with a longer sampling period. The availability of daily energy data from smart meters means that it is increasingly possible to gather sufficient data to characterise large numbers of homes.

The importance of sufficient data points above and below the balance temperature—in other words from both summer and winter—is clear from the limitations of the results of the model applied to only 12 monthly data points discussed above in Section 5.5. The change in the results when unoccupied days are removed illustrated in Section 5.1 shows the inter-dependency of the three model parameters and how uncertainty in the calculated T_b has an impact on the calculated HPLC.

The mean balance temperatures calculated over a five year period vary between 15.2 and 15.8°C. The very small difference in median parameters between the fixed and variable balance temperature models for a single year confirms that the assumption of a 15.5°C base temperature for UK heating degree-day calculations is a good approximation of the typical outside temperature below which heating is used.

8. CONCLUSIONS

The power temperature gradient (PTG) energy signature method is a valuable technique for assessing in-use energy performance based only on energy and external temperature data. It allows remote assessment of building performance without the need for instrumentation beyond a smart meter. The method provides more information than simply reporting annual energy since it disaggregates to three variable— heating power loss coefficient (HPLC), T_b and P_b—which have simple-to-interpret characteristics. The analysis of data from the SERL Observatory shows how energy signatures can be tracked over time for large numbers of homes which have given consent for the data from their smart energy meter to be included.

The energy signature has significant potential to help governments track progress to achieving their building efficiency targets. The HPLC indicator is a measure of how much power is needed to raise the temperature by 1°C (this is dependent on the efficiency of both the heating system and the building fabric) and helps track and evaluate policy progress related to making the building stock more thermally efficient. It allows results to be compared across years, allowing for differences in climatic conditions. While human influences mean that the in-use HPLC calculated by the models differs from a theoretical heat transfer coefficient (HTC) based on building fabric and heating efficiency, this indicator has practical applications for ranking the performance of homes and investigating changes in actual performance before and after energy efficiency improvements.

Homes that would benefit most from energy efficiency retrofits can be identified and targeted in policy, and the actual (as opposed to modelled) performance post-retrofit can be tracked. The method has the potential to be used in reducing assessor error in estimating the heating efficiency of homes and to help validate building energy models used by policymakers such as the UK’s Standard Assessment Procedure (SAP) for Energy Performance Certificate (EPCs) and Home Energy Model (HEM).

As buildings and heating systems become more efficient, baseload power for appliances makes up an increasing proportion of home energy demand. This is heavily driven by occupancy and the number and efficiency of the appliances. The baseload P_b calculated from the PTG method can be used to evaluate changes in baseload and the impact of appliance policy.

Energy signatures are not only relevant for energy efficiency policy. The building stock response to external temperature (HPLC) is a critical parameter for planning for an all-electric future as electricity demand for heating will have an important impact on the capacity required from the electricity grid in future.

AUTHOR CONTRIBUTIONS

C.H.: data curation, formal analysis, methodology, software, visualisation, writing—original draft preparation; J.F.: methodology, software, writing—review and editing; F.H.: methodology, software, writing—review and editing; S.E.: funding acquisition, data curation; T.O.: supervision, writing—review and editing.

COMPETING INTERESTS

The authors have no competing interests to declare.

DATA ACCESSIBILITY

Smart Energy Research Lab (SERL) Observatory data are available to accredited researchers working on approved projects. For information about how to apply for data access, see www.serl.ac.uk.

ETHICAL APPROVAL

The SERL Observatory was given ethical clearance to collect energy and other data from participants by the UCL Research Ethics Committee (reference number 14793/001). The participants gave their consent for the use of their data in the research.