Cluster analysis of occupancy schedules in residential buildings in the United States
Introduction
Occupant behavior in buildings has been an emerging area of research in recent years, the study of which has significant benefits to the efficient design of building spaces, heating, ventilating and air conditioning (HVAC) systems, and lighting systems, as well as to the improvement of thermal comfort controls [1]. Six key factors are considered to influence energy consumption in the building sector, including the (a) building envelope, (b) equipment, its (c) operation and maintenance, (d) indoor comfort criteria, (e) occupant behavior and (f) weather conditions [2]. Among these six parameters, (e) occupant behavior, (d) indoor comfort criteria and (c) operations and maintenance generally have a comparatively greater impact on energy performance. Occupants’ behavior and lifestyle choices (e) significantly influence the building indoor environment and overall energy consumption [3], [4]. Lo et al. [5] found that accurate occupant detection can reduce the total energy use in an open space office building by up to 30%. Among climate conditions, housing type and occupant behavior, occupant behavior was found to have the most influence on cooling energy consumption in residential buildings [6]. Variation in occupants’ behavior have also been found to result in an up to 87% change in the air change rate in a residential building which uses both air conditioning and operable windows [7]. A study of 25 households in a residential building in Beijing, China found that HVAC electricity consumption varies from 0 to 14 kWh/m2, due in large part to the variations in how occupants adjust their HVAC system setpoints and controls [8]. An et al. (2018) studied the energy consumption of air conditioning in different rooms in a residential building sector in China, obtaining up to different profiles were obtained based on the occupant’s utilization of the air conditioning system [9]. In another study, Xia et al. (2019) completed cluster analysis of the energy consumption used by the air condition system in residential building in China and showed three different consumption patterns of the air conditioning system, which varied depending on occupant preferences [10]. It was also shown that, compared to commercial buildings, occupant behavior has a more significant impact on residential building energy consumption [11], [12], [13]. This demonstrates the significance of occupancy prediction in residential buildings, as much of energy consumption is driven by occupants’ use of energy-consuming devices. Therefore, in order to evaluate the overall energy consumption in a residential building, accurate prediction of occupancy schedules is needed.
The development of occupancy schedules has been an ongoing topic in recent literature; however, most studies focus on assessing the occupancy profile for a single or small subset of buildings [14], [15]. Occupancy prediction based on the data of a single building can predict the occupancy profile for that particular building, however it may not necessarily accurately represent other residential buildings’ occupancy characteristics. Therefore, there is a significant need to evaluate the characteristics of the overall occupancy schedules for the U.S. population, for use in improving how occupancy is represented in building energy models. Currently in building energy simulation models, static occupancy schedules are used, which provide a “typical” occupancy schedule in the United States over a year long period. However, the drawback of this is that if a typical schedule is used throughout the entire simulation period, the daily and hourly variations in occupancy are not represented. In a typical energy model which uses such schedules, there are no situations where no one present in the home (i.e. an occupancy fraction of zero). This becomes problematic when estimating energy and/or demand savings from the use of occupant-centric controls. In addition, there have been very few studies that characterize the different types of occupancy profiles for the U.S. population as a whole. This study seeks to address these two challenges.
Several studies in recent literature have focused on estimating the occupancy patterns in buildings. Different approaches have been considered to evaluate occupancy schedules. Statistical methods, data-driven methods and survey-based analyses are among the most commonly used methods. To evaluate the stochastic nature of occupancy prediction, Markov chain methods are among the most commonly used in recent literature. One of the pioneer studies in this field by Page et al. uses a discrete time Markov chain model to evaluate stochastic occupancy profiles [16]. This model works well for single zone, single occupancy scenarios, for shorter spans of absence from the studied building(s). Richardson et al. also studied first order Markov Chain models to evaluate occupancy scenarios in a building space [17]. Adamopoulou et al. developed Markov and semi Markov models based on the data obtained from camera and motion sensors, to simulate the occupancy for different spaces during different times across a one-day period [18]. Two stochastic models were proposed by Chen et al. for multi-occupant, single zone and multi-occupant, multi-zone scenarios [19]. For the multi-occupant single zone scenario, an inhomogeneous Markov chain method was used for occupancy prediction. Beyond the use of Markov chain models, other methods such as recursive algorithms have also been used for daily occupancy forecasting. However, as mentioned, most studies have focused on a single building or set of buildings, which is difficult to use to generalize for the overall U.S. population [20].
Different data mining algorithms are also used to explore occupancy scenarios. Clustering techniques can be used to identify distinct features among occupancy schedules, allowing for the classification of people into different categories based on their typical schedules. Clustering-based algorithms have also been used in recent literature for electricity consumption data to estimate occupancy patterns in buildings [21]. Buttitta et al. used cluster analysis to identify groups of households with similar types of occupancy profiles for five regions in the UK [22]. Zhao et al. used C4.5 Decision Tree and Support Vector Machine (SVM) methods based on data from Bluetooth-connected devices to predict occupancy schedules [23]. In other research, neural network methods, including Artificial Neural Networks, Convolutional Neural Networks, and Recurrent Neural Networks are also used to develop representative occupancy profiles [24], [25], [26]. However, most studies focus on commercial buildings, with evaluation of performance limited to the specific location of testing only.
Several survey-based studies have also focused on occupancy schedule prediction in residential buildings. Balvedi et al. developed a 24-hour and 48-hour schedule for weekdays and weekends, respectively, based on interviews with occupants [27]. Similarly, Carpino et al. conducted a questionnaire-based study across 80 families in Italy for 2 weeks in 2017 to evaluate the percentage of time people spent in their home [28]. Occupancy profiles and their interaction with energy-consuming appliances for residential buildings has also been studied in China based on the completion of a questionnaire and subsequent interviews of occupants [29]. These survey-based studies provide insights on occupant behavior and their interactions with building systems. However, conducting similar studies to represent the overall U.S. population would be expensive and time intensive.
To study occupant behavior for the U.S. population, it is important to use representative data. However, it is challenging to create a dataset to represent the typical characteristics of occupant behaviors [30]. Time Use Survey data can fulfil this data requirement, as it statistically represent the overall population of a country. Several countries, including China, Japan, Belgium, the U.S., etc… publish their time use survey data which contains detail about their respective populations’ characteristics and their activities throughout the day [31], [32]. Aerts et al. used clustering methods with the Belgian Time Use survey for 2005 data to identify the typical occupancy profiles in residential buildings [33]. Occupancy profiles and their interaction with the building energy systems were evaluated using the UK Time Use Survey data [34]. UK Time Use Survey data was also used to predict the location and activities of occupants using time-inhomogeneous first order Markov Chain model [35]. Similarly, French Time Use survey was used to characterize occupants based on their activity profiles [12]. Spanish Time Use survey data from 2009 to 2010 was used to predict the occupancy and pattern of energy consumption in the residential buildings [36]. ATUS data from 2006 was used to characterize different activity patterns of occupants by [37]. However, the variation in the occupant profiles for households with different numbers of members is not evaluated in these recent studies. In addition, additional study of data with more recent datasets is also needed for the improved characterization of occupancy profiles.
In summary, occupant behavior is among the most important parameters in predicting energy consumption in residential buildings. Detailed study is required to analyze characteristics of occupancy schedules to develop a better understanding of the variations in occupancy schedules. In order to address the previously discussed challenges, this paper uses a clustering algorithm to analyze the different type of typical schedules occupants follow in residential building in the United States. To achieve this, American Time Use Survey (ATUS) data from 12 years, 2006 to 2017 [38] was used. Factors influencing occupancy schedules, including age of the occupants and weekday or weekend, are also considered in this analysis. The results of this work provide unique insights using data across a longer time span as compared to previous studies, and the most recent data available. In addition, unlike previous analyses, this study also analyzes the characteristics of each of the profiles. This provides a clearer picture of occupancy, for use in assessing the potential energy and demand savings from occupancy-centric control in residential buildings. This paper is organized in four sections. First the dataset characteristics are discussed, followed by data processing and the clustering methodology used for overall data analysis. Next the results are discussed, comparing variations in results according to age, and day of the week. Different characteristics of the cluster profiles are also analyzed in this study. This is followed by the conclusions, limitations and ongoing and future work. The results of this study can be used to create an occupancy simulator for the residential buildings in the United States to predict schedules based on known occupant characteristics.
Section snippets
Datasets
The American Time Use Survey (ATUS) [37] is an annual survey conducted by the United States Bureau of Labor Statistics. The objective of this survey is to collect information for overall U.S. population, on the activities people complete throughout the day. This information is collected through a combination of email, telephone and in-person interviews. Participants are asked to document their activities, and where the activities take place, and whenever there is any change in activities. In
Methodology
Activity data was first extracted from the ATUS dataset. The entire dataset was divided into several groups based on the classifiers that were selected based on previous studies, including the age of the participants and whether the day of study was a weekday or weekend [40]. Age was divided into seven categories, including under 25, 25–34, 35–44, 45–54, 55–64, 65–74 and over 75 years of age, based on categories of age ranges used in the Residential Energy Consumption Survey data [41]. The data
Activity variation
Total duration of different activities within each age group were evaluated and compared (Fig. 1). The most common activities in Fig. 1 include ‘personal care including sleeping’, ‘household activities’, ‘caring and helping household members’, ‘work related activities’, ‘education’, ‘eating and drinking’, ‘socializing, relaxing and leisure’, ‘traveling’ and ‘other activities’. The ‘other activities’ category consists of activities such as ‘government services & civic obligations’, ‘religious
Conclusions
In this study, ATUS data is studied to analyze the occupancy profiles for individuals in the United States. Cluster analysis was completed initially to determine different patterns in occupancy schedules for people in different age group and whether its weekday or weekends. After that, the accuracy of the cluster profiles was evaluated and the variations in the profiles were studied. Different characteristics of the schedules were also analyzed. The overall key findings of this study can be
Credit authorship contribution statement
Debrudra Mitra: Conceptualization, Methodology, Data curation, Formal analysis, Writing - original draft, Writing - review & editing. Yiyi Chu: Conceptualization, Writing - review & editing. Kristen Cetin: Conceptualization, Methodology, Writing - review & editing, Supervision.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
The information, data, or work presented herein was funded in part by the Advanced Research Projects Agency-Energy (ARPA-E), U.S. Department of Energy, under Award Number DE-AR0001288. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. Additionally, the authors would like to thank Nicholas Steinmetz for his professional and constructive help on statistical methods and techniques.
References (55)
- et al.
Occupancy sensing in buildings: a review of data analytics approaches
Energy Build.
(2019) - et al.
IEA EBC annex 53: total energy use in buildings—analysis and evaluation methods
Energy Build.
(2017) - et al.
Occupant behavior modeling for building performance simulation: current state and future challenges
Energy Build.
(2015) - et al.
A systematic procedure to study the influence of occupant behavior on building energy consumption
Energy Build.
(2011) - et al.
Localized air-conditioning with occupancy control in an open office
Energy Build.
(2010) - et al.
Behavioural, physical and socio-economic factors in household cooling energy consumption
Appl. Energy
(2011) - et al.
The effects of human behavior on natural ventilation rate and indoor air environment in summer—a field study in southern Japan
Energy Build.
(1997) - et al.
Clustering and statistical analyses of air-conditioning intensity and use patterns in residential buildings
Energy Build.
(2018) - et al.
A study on occupant behaviour related to air-conditioning usage in residential buildings
Energy Build.
(2019) Computational intelligence modelling based on variables interlinked with behavioral tendencies for energy usage profile – a necessity
Renewable Sustainable Energy Rev.
(2018)