1. Introduction
Our urban and residential areas have become far more technological and digitally revolutionised [
1]. Long-standing amenities have now become enhanced, driven by data analytics. Examples include the smart grid for optimised energy management integrated with information communication technologies [
2]; multifaceted intelligent control systems for traffic flow control [
3]; emergency management and public security systems [
4]; electronic health records for both hospital management systems and efficient patient processing [
5]; and Internet of Things (IoT) sensor integration for multifaceted data collection opportunities [
6]. Other less mainstream services, such as e-grocery, referring to the online ordering of produce being directly delivered to the customer [
7] (the benefits of which were particularly evident as a supply solution during the Covid-19 pandemic) and drone-based delivery for food products and parcels (direct from the supplier to the customer), are capable of improving delivery times and transforming the way distribution networks operate [
8]. Both of which are regularly documented in the media and pose urban logistical challenges for the future.
The technological changes are the product of many complex social paradigms. One of which being the need to reduce our carbon footprint, creating cleaner sources of energy or energy efficient services. As countries begin to adapt to climate change and consider the impact they have on the environment (for example the UK government’s ambition to have a net zero emission by 2050 [
9]), smart-city technologies provide an ideal solution for creating energy-efficient and sustainable cultures [
10]. Smart systems have resulted in more opportunities than ever before for data scientists and technologists to create intelligent services which have significant advantages for reducing carbon emissions. The smart meter, as part of the smart grid, is a particular example [
11]. Evidence suggests that residential buildings are responsible for contributing significantly towards global energy consumption [
12]. As an integral (and well known) part of the smart grid infrastructure, smart meters now have a core role in reducing costs for the consumer and utility company and tremendous environmental benefits. Therefore, research into the recognition of human behavioural patterns in residential settings using only smart meter data has gained significant traction over the last five years [
13,
14], which could lead to the development of automation services to support large-scale reductions in emissions.
The datasets produced by smart meters are granular, and behaviour profiling is achieved typically by one of three techniques: (1) aggregated load monitoring, (2) disaggregation using high-frequency sampling and (3) a hybrid approach, combined with IoT sensors within a wider smart home environment context.
However, as residential homes are core contributors to the carbon emission levels, it is the end-user that ultimately should make changes to their own carbon footprint. Whilst real-time information services can be provided (e.g., using In-Home Displays (IHD)) to make citizens more situationally aware of their consumption, there are limitations for the provision of autonomous services on a global/mass scale. This is because of two considerations: (1) the big data capabilities to provide autonomous transformation on a mass scale are still insufficient, bearing in mind that one smart meter produces 175 MBs of data a year on average [
15]; (2) not everyone has a smart meter and, excluding governmental policy changes, some may never have one. Smart meters have a tendency to polarise opinion due to the considerations surrounding privacy issues and the attractiveness of the data for cyber-criminals [
16]; therefore, some home owners are averse to having a smart meter installed. However, this is not unique to the smart metering technology. There are considerable privacy concerns relating to wider smart home technologies in general, particularly with the increasing ubiquity in wireless technologies for home automation (and the sensitivity surrounding remote patient monitoring applications [
17]).
Therefore, this paper considers the perspective of the individual home by presenting an investigation into their view of the smart meter installation within the smart cities’ context. Considering the polarised view of the smart meter installation, coupled with the popularity of implementing green technology initiatives, the public opinion concerning the smart meter installation is complex. Yet if the greener society and energy revolution goals are to be met, a greater expansion of smart urban initiatives, such as the smart metering infrastructure, is required. Whilst related research shows differing opinions on the potential reduction in energy consumption by means of a smart meter installation (2–15% [
11,
18]); there is nonetheless a unanimous acknowledgement of a decrease in consumption, which is ultimately beneficial.
The quantitative investigation in this paper aims to contribute to the discussion by means of the following research empirical hypotheses; (1) H1: Certain end-user characteristics (such as age, social class, employment status and education) have a statistically significant impact on the consideration of using smart meters for reducing the energy bill; (2) H2: Age, social class, employment status and education can be statistically attributed to adopting smart meters due to concerns about the environment. (3) H3: Users within a particular age group, social class, education level or employment status are statistically interested in changing their home behaviour by using a smart meter; (4) H4: Certain end-user characteristics are linked to how individuals feel about using smart meters to save energy. The corresponding Null hypotheses would be the inverse of the H1–H4 statements and are labelled as H01–H04. For example, H01 would be there is no link between the users’ characteristics and their interest in using a smart meter to reduce their energy bills.
This work is uniquely poised as the focus is principally on gas smart meter data. Gas smart meters are primarily used for automatic billing applications. They have known uses in an industry setting for the detection of anomalies (e.g., leaks or violations), remote reading and preventative maintenance. However, the majority of similar research in this domain has a core focus on the electricity metering infrastructure [
11,
19] or water smart meters [
20]. To achieve this investigation, the Commission for Energy Regulation’s (CER) smart meter gas dataset is used, which is comprised of readings taken from 1026 gas smart meters in Ireland between 2010 and 2011 [
21]. The dataset is accompanied by a granular pre- and post-survey consisting of both qualitative and quantitative data relating to the demographics and usage profile of the bill payer. This paper is structured as follows.
Section 2 provides a background discussion on related work.
Section 3 outlines the mixed methodology used in this research.
Section 4 discusses the results achieved, and the paper is concluded in
Section 5.
4. Results
In this section, each of the hypotheses outlined in
Section 1 are evaluated in order. The data is pre-filtered to check for missing values, with the final resulting dataset comprised of 1365 respondents, reduced down from 1576. The expectation in the evaluation is to uncover subtle influences. This is because the data used is from the pre-trial conducted in 2011–2012. During this time, the smart meter technology was less known and an emerging technology. Furthermore, whilst energy/carbon reduction has been debated for many years, it has gained prominence and momentum in more mainstream media outlets in recent years; in line with government-led policies such as the aforementioned UK government’s aim to reach net zero emissions by 2050.
4.1. Hypothesis 1 Evaluation—Reducing the Bill
Hypothesis 1 involved experimenting with a multiple linear regression model, with Question43323 (I am interested in changing the way I use gas if it reduces the bill) set as the outcome variable. The results are outlined in
Table 2, which provides the estimate standard deviation, error,
t-value and
p-value.
Two predictor variables have a statistically significant impact on the respondents’ attitude towards reducing their energy bill. Social class (Question401) and age group (Question300) have t-values of 2.217 and 3.162, with p-values of 0.268 and 0.0016, respectively (where a t-value above 2 and p-value below 0.05 imply statistical significance). Education has a negative impact, meaning that respondents with a higher education qualification are more likely to have a higher interest in reducing their bill (as the grading system for education in the survey is the inverse of the Likert scale for interest in reducing billing). As age and social class both have positive effects, the assumption is that as age and social class increase, the participants are more likely to have an interest in the cost benefits of the smart metering infrastructure.
However, overall, the above model explains 3.65% of the variance, where the multiple R
2 value is 0.03651 and the
p-value for the
f-test is 2.681e−
10, well below 0.05. Survey data, when compared with census data, tends to produce lower R
2 values; however, the
p-value is well below 0.05, demonstrating that the social class and age group are statistically significant considerations when reducing the bill is the outcome variable.
Figure 3 displays (a) the co-efficient plot and (b) a visualisation of the fitted values. As (b) displays, the residuals do not have a local mean of 0, indicating that functional form has been violated. However, as the residuals are downward sloping, heteroscedasticity is checked using the Breusch–Pagan (
bp-test) test, where a
p-value of 4.487e−
09 is generated. The values are therefore checked using the coef-test (linear regression with robust standard errors). This confirms the results and produces
p-values of 0.001602 for Question300 and 0.026791 for Question401, confirming the results.
Figure 4a displays the component-plus-residuals plot, where the blue line represents where the linear relationship should be. The purple line refers to the estimated linear relationship. Question5418 (when compared with the other three) appears to be the least accurate. This would be in line with the survey data overview presented in
Table 2.
Figure 4b displays the influence index plot, where outlier values are depicted.
4.2. Hypothesis 2 Evaluation—Enviromental Impact
This experiment is concerned with how users consider changing the environment as a factor in their decision to adopt a smart meter. Question4333 (“am interested in changing the way I use gas if it helps the environment”) is the outcome variable with a Likert scale of 1 (strongly agree) to 5 (strongly disagree). The evaluation results are presented in
Table 3. None of the predictor variables appear to have a statistical significance for the outcome variable; with Question300 having the lowest
p-value (0.0916), which is well above 0.05. Therefore, it could be inferred from this experiment that whilst changing the environment might be of interest to the end-users (as outlined in
Section 3,
Figure 2), there appears to be little individual relationship between the end-users’ age, employment status, social class or education. It is not possible to disprove H
02.
However, as a collective, the predictor variables produced a 2.806
f-test and 0.02455
p-value, with an R
2 of 0.0082. The co-efficient is plotted in
Figure 5a, with the fitted values displayed in
Figure 5b (given the low impact, the component, residuals and diagnostics results are not plotted for this experiment).
4.3. Hypothesis 3 Evaluation—Behaviour Change
Question4334 asks if end-users are interested in changing their behaviour around the home. According to
Figure 2 (in
Section 3), there is a strong level of interest, but less so than when compared with having an interest in reducing the bill or helping the environment. For example, 44.9% select 1 (strongly agree), as displayed in the frequency distribution
Table 4. However, 117 participants also display no interest (i.e., 5 in the Likert scale) in changing their home routine.
In this experiment, the relationship between age, class, education, employment status and the end-users’ feelings towards changing their behaviour is examined. Using multiple linear regression, a
p-value of 8.974e−
07 displays that there is statistical significance, confirmed by the
f-test, which is 8.5. However, the R
2 value of 2.44% shows that the variance is relatively low. The results are presented in
Table 5, where, similarly to the H
1 experiment, Age (Question300 (
p = 0.0242)), but also Employment Status (Question310 (
p = 0.0086)) have the most statistically significant impact on the respondents’ attitude towards changing their home behaviour.
Figure 6a displays the Co-efficient Plot. Due to the downward sloping trend, heteroscedasticity is checked again using the
bp-test, which produces a
p-value of 0.1353. Therefore, the results are checked with the coef-test, which produces 0.008577 for Question310 and 0.024216 for Question300, confirming the conclusions achieved with the multiple linear regression. The results are plotted in
Figure 6, with the co-efficient plot displayed in 6a and the fitted values in 6b.
Figure 6a shows the linear relationship and estimated linear relationship for H
3.
Figure 6b has relatively few outliers present (
d-values), meaning there appears to be a minimal level of influential data points, which may have distorted the results.
Record 1232, in
Figure 7b, appears to have the highest outlier value. Therefore, conducting an investigation into the survey responses of the individual may offer insight into the premise for the high outlier score. Observation 1232 in the dataset corresponds to User 2235, who is in the 26–35 age group, unemployed (actively seeking work) and refused to provide information relating to their social class. The individual also left education at the third level (the lowest level). The individual responded with 3 in the Likert scale for Questions 4333 and 4334, showing a lukewarm interest in reducing their bill and changing the environment. The user also had no interest in either changing their home behaviour or saving energy, responding with 5, strongly disagree, to both. This may be an indication for the high outlier score, as very few individuals both strongly disagreed with having an interest in saving energy and changing their home behaviour.
4.4. Hypothesis 4 Evaluation —Save Energy
Concerned specifically with saving energy, Question5011c displayed a highly positive response from the participants, particularly when compared with changing home behaviour. The general trend displayed in
Figure 2 (
Section 3) is that the respondents are overwhelmingly positive. 961 people (which is 70.4%) of the participants selected 1 (strongly agree) in being interested in reducing energy. The multiple linear regression experiment produced an R
2 score of 1.64%, with a
t-test score of 5.6 and a
p-value of 0.0002, showing there is statistical significance between the users’ characteristics and their interest in saving energy; however, the results show that there are no specific characteristics which can be linked to the users’ interest in saving energy. None of the
t-test scores are above two. Similarly, none of the
p-values are below 0.05. The authors conclude from the results for H
4 (
Table 6) that there is no statistical significance between age, social class, employment status or education when displaying an interest in saving energy. The regression results are visualised in
Figure 8.
4.5. Discussion
With smart meters polarising opinion, due to the aforementioned privacy concerns, the perspective of the end-user is important within the climate change discussion. The privacy factor could have a separate impact on the end-user perspective of the smart metering installation. However, investigating this consideration falls out of the remit of the survey data available using the CER dataset. Typically, applications relating to data mining algorithms (such as machine learning or deep learning) are used to invade the privacy of consumers in sophisticated manners [
36]. However, the questions available in the survey data do not cater for the end-users’ knowledge of advanced techniques such as time pattern analysis, consumer profiling or appliance/occupancy detection. Therefore, in the experimentation, the authors aimed to further the conversation into the end-users’ perspective when considering whether to adopt a smart meter, and how certain attributes may affect their decision towards (1) reducing the energy bill, (2) helping the environment, (3) changing their home behaviour and (4) saving energy.
In the related work discussed in
Section 2, Fredericks et al. focus in one level of education in their investigation; but as demonstrated in the experimentation results in this paper, education has a varied impact depending on the question. For example, in H
1 social class (
p = 0.268) and age group (
p = 0.0016) are statistically significant contributors to the positive opinion towards the smart meter installation when considering a reduction in energy billing, whereas education has a lesser impact. Similarly, in H
4 there is a cross-spectrum interest in reducing energy, whereas no core attributes, such as education, play a role in this interest.
Concerning the end-user perspective on adopting the smart metering technology to change the environment, H2 shows that there is a statistical significance, which supports the background investigation (for example, 0.8% variance in respondents regarding the reduction carbon emissions). As previously discussed, Fahin et al. outline that by understanding the household occupants’ characteristics, it would be possible to implement targeted change through suitable energy saving campaigns. Yet, as demonstrated in H2, it is a challenge to link homeowner characteristics to their opinions on changing the environment. Since none of the predictor variables had a significant impact on the outcome variable, future works might include further predictor variables (such as dwelling type). This makes targeted advertising a challenge without load profiling, using real-time analytics such as clustering or supervised learning. However, in H3, when the focus is not purely on reducing carbon emissions, the results demonstrate that age (p = 0.0242) and employment status (p = 0.0086) have a statistically significant effect on the attitude towards changing home behaviour.
The main limitations of this work are linked to the dataset. While the information within is comprehensive and the survey is granular, some opinions on energy conservation might have changed since the trial took place, during 2011–2012. However, worldwide, the technology is still very much in its infancy, and many are unfamiliar with it and with what can be achieved by means of its installation. For that reason, the opinions reflected in this survey may be apparent for other nations currently in a stage of deployment similar to that of the CER trial in 2012.
5. Conclusions
As part of the industry/city 4.0 evolution, smart city technologies have the potential to both raise awareness concerning local environmental challenges and promote the use of future renewable energy technologies. In this paper, four hypotheses were investigated related to the end-users’ perspective on the installation of smart meters to reduce billing, change the environment, save energy and change home behaviour. The results demonstrated that there is a statistical significance between user characteristics and their opinions on the benefits offered by smart metering technologies. Specifically, (1) (H1) the authors are able to infer that the characteristics of age, social class and education have a statistically significant impact on the consideration of using smart meters for reducing the energy bill; (2) (H2) The results from H2 demonstrate that it is not possible to disprove the null hypothesis (H02), and therefore it is not apparent that age, social class, employment status and education can be statistically attributed to adopting smart meters due to concerns about the environment; (3) (H3) The third experiment demonstrated that users within a particular age group or employment status are statistically interested in changing their home behaviour by using a smart meter. However, employment and social class have no statistical significance on the users’ interest in changing behaviour; and (4) (H4) It is very clear that there is no end-user characteristic in the experimentation that can be attributed to how individuals feel about using smart meters to save energy.
In our future work, we will incorporate data from the post-survey to compare with the opinions held before the smart meter trial took place. Morever, adding additional variables to the multiple regression models related to the dwelling type (e.g., #Question450—dwelling type, #Question452—rent vs own, #Question460—number bedrooms), could offer further insights into linking the end-users’ perspective on the smart metering infrastructure to their attributes.