Energy and Buildings 117 (2016) 332–340

Contents lists available at ScienceDirect

Energy and Buildings

journal homepage: www.elsevier.com/locate/enbuild

A preliminary research on the derivation of typical occupant behavior

based on large-scale questionnaire surveys
Xiaohang Feng, Da Yan ∗ , Chuang Wang, Hongsan Sun
School of Architecture, Tsinghua University, China

a r t i c l e i n f o a b s t r a c t

Article history: Occupant behavior is a major contributing factor to building energy consumption. Large discrepancies
Received 1 July 2015 might occur when occupants with diverse types in terms of behavior patterns reside in an identical
Received in revised form building. Moreover, it is vital to derive several kinds of typical occupant behavior that represent the
17 September 2015
assemblage of occupants in a simplified way as the reference for building designers and energy pol-
Accepted 21 September 2015
icy makers. This paper presents a preliminary research on occupant behavior categorization through
Available online 24 October 2015
the building energy simulation, where some typical occupant air-conditioning behavior patterns use in
Chengdu, China have been derived from a large-scale questionnaire survey based on the energy con-
Keywords:
Occupant behavior
sumption levels. These typical occupant behavior patterns could be used as the context for an overview
Data mining of building energy in a community, as well as the evaluation of building energy-saving technology where
Building energy simulation occupants are involved. Further efforts are required in the algorithm of dividing occupant behavior into
Questionnaire survey typical patterns, the validation of the classification results, and the analysis of other behaviors.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction To present occupant behavior in buildings in a quantitative way

Occupant behavior is now recognized as a non-negligible fac-
tor to building energy consumption. It influences energy use in
multiple ways, such as occupancy-based casual gain, operations on
appliances, set points of air-conditioning, etc. Al-Mumin et al. [1]
surveyed the occupancy patterns and operation schedules of elec-
trical appliances in 30 residences in Kuwaiti and used them as input
data of a thermal simulation program. The surveyed occupants left
all lights on even rooms were unoccupied, with the AC thermo-
stat set at 22◦ C. This pattern of occupant behavior led to a 21%
of annual electricity consumption compared to the default sched-
ule in the simulation program. In order to study the uncertainty
in heating energy consumption caused by occupant behavior, Cle-
venger and Haymaker [2] studied the different settings regarding
behavior schedule and set points in American codes. The simulation
revealed that a 150% discrepancy of energy consumption was found
between these different settings. Li et al. [3] measured the cool-
ing energy consumption in a residential building during a summer
in Beijing, China. The energy consumption of different apartments
varied widely, caused by the variance in the durations when air-
conditioning was on. Due to its significance, occupant behavior has
drawn much attention in building simulation field.
∗ Corresponding author.
E-mail address: yanda@tsinghua.edu.cn (D. Yan).
so as to integrate them into simulation programs, various mod-
els have been developed. For example, Nicol and Humphreys [4]
studied the use of windows, lighting, heating, etc. with surveys to
see how the use of each control varies with outdoor temperature.
Andersen et al. [5] measured occupants’ window opening behavior,
as well as indoor and outdoor environmental conditions, based on
which a logit model was developed, describing the probability of
an opening or closing event took place. Yasue et al. [6] surveyed the
air-conditioner and window use in residential buildings in Japan,
and proposed the sigmoid function to describe the probability of
operating air-conditioners and windows. These models are gener-
ally developed from intensive field surveys or measurements, and
correlate behavior with some extrinsic factors like temperatures or
daily events.
When it comes to industry such as building design or code
formulation, the diversity of occupant behavior becomes more sig-
nificant than the case studies where the models are developed.
Occupant behavior is necessarily simplified to an “average” or
several “typical” patterns instead of applying such complicated
models. The different composition of occupant behavior patterns
in a building may have great influence on the energy consumption.
Since determining the parameters in these models often requires
intensive measurements, it is temporally and financially consum-
ing to describe each occupant in a large population applying the
approach in which models are developed. The researches men-
tioned above in occupant behavior have focused on the quantitative

http://dx.doi.org/10.1016/j.enbuild.2015.09.055
0378-7788/© 2015 Elsevier B.V. All rights reserved.
 X. Feng et al. / Energy and Buildings 117 (2016) 332–340 333

modeling but are not capable of describing the diversity of a large

population, which still has a margin from the industrial application
of occupant behavior in simulation where the representativeness
of diverse occupants is required.
With the development of data collection and analysis technol-
ogy, there have been more and more researches on date mining of
occupant behavior recently to study representative behavior pro-
files. Yu et al. [7] applied the data mining technique, cluster analysis,
in attempt to classify occupants into several categories in terms
of context factors (such as climatic conditions) and their behav-
iors. This study introduced the cluster analysis into data mining in
building energy simulation, aiming at predicting the energy con-
sumption level when context factors and occupant behavior related
factors were known. D’Oca and Hong [8] applied cluster analysis
to categorize window opening and closing behavior in 16 offices
according to the influencing factors, durations of the window state,
the number of window position changes and the most frequent
window degree of opening. Furthermore, the association rules were
applied to produce two typical window opening office use profiles.
Another study adopted data mining to analyze the space heating
performance in residential buildings [9]. Six patterns of the room
temperature demand were revealed, associated with the heating
Fig. 1. Technical approach to categorize occupant behavior patterns.
energy consumption to predict the performance of heating system.
These studies have introduced new data mining techniques
consumption could be represented by a “typical” one. Fig. 1 shows
into occupant behavior simulation field, providing new perspec-
the technical approach of our study.
tives in behavior pattern categorization. Nevertheless, they are
The questionnaire survey was distributed to get the large sam-
limited to looking into the cases where the measurement was
ple data regarding occupants’ air-conditioning behavior patterns.
undertaken. Since the patterns are derived from the specific data
The questionnaire was designed in line with the occupant behavior
set, heavily depending on measured cases, they are not con-
modeling framework proposed by Wang [10], who had proposed a
vincible to be extrapolated to other occupants or buildings. It
new approach for the quantitative description from an action-based
is meaningful to conclude behavior patterns and their distri-
view. The questionnaire was used to find the influencing factors
bution in large population in terms of energy consumption to
for occupants to operate various appliances like air-conditioning,
be used in energy simulation and technology evaluation. With
lighting, windows, etc, while it was not capable of capturing the
air-conditioning behavior as an example, this exploratory study
quantitative relationship between the factors and the behavior.
proposes a novel approach to categorize occupant behavior into
To overcome this shortcoming, a series of field measurements
several typical patterns, and to derive the distribution of these
on indoor environment and behaviors were conducted to build
typical patterns, from a broad survey, in attempt to narrow the
the quantitative behavior model. The patterns were described as
gap between behavior modeling and industrial application where
conditions under which occupants would turn on/off their air-
representativeness of occupant behavior is required. The typical
conditioners, e.g. “turn on when feeling hot”, or “turn on when
patterns could be used to analyze the energy variation with differ-
sleeping”. Patterns could be combined in the questionnaire sur-
ent distributions of occupants in a community.
vey, which led to a larger amount of patterns than the case when
only a single pattern was allowed. After the questionnaire sur-
vey, the patterns were input into the simulation program DeST
2. Method [11], in which a novel occupant presence and behavior module
had been developed [12], to simulate the energy consumption with
The purpose in this study is to conclude the typical air- different behavior patterns. From the distribution of the simu-
conditioning behavior patterns that differentiate energy usage lated energy consumption, some typical air-conditioning behavior
levels. Starting with this, energy consumption is labeled as the patterns were concluded based on different levels of the air-
index to tell apart behavior patterns. Patterns with similar energy conditioning energy consumption.

Fig. 2. Daily outdoor temperature of the typical meteorological year in Chengdu.

334 X. Feng et al. / Energy and Buildings 117 (2016) 332–340

Table 1 temperature (◦ C), presenting the lowest temperature when there

Options in the questionnaire for air-conditioning modes in the living-room.
is a probability for occupants to switch on the air-conditioning; l –
Switching on modes Switching off modes the scale parameter (◦ C), to nondimensionalize the temperature; k
[a] Never on [a] Never off until the end of summer – the shape parameter, presenting the sensitivity to the environ-
[b] Always on during summer [b] Off when leaving the living-room ment;  – the time step used in measurement and simulation,
[c] On as long as entering the [c] Off when leaving home typically 5 or 10 min; u, j, k are constant parameters that could be
living-room decided by data fitting after measurements.
[d] On when feeling hot [d] Off before sleeping
The event triggered patterns are modeled as a probability when
[e] On regularly at (time) [e] Off when feeling cold
[f] On if there is a guest [f] Off regularly at (time) the related event happens. For instance, “turn on as long as entering
[g] Others, please fill in [g] Others, please fill in the living-room” is described as

p if  = 0
Table 2 Pon = (2)
Options in the questionnaire for air-conditioning modes in the bedroom. 0 if  =
/ 0
Switching on modes Switching off modes
where Pon – the probability for occupants to switch on the air-
[a] Never on [a] Never off until the end of summer
[b] Always on during summer [b] Off when leaving the living-room
conditioning;  – the current time in simulation;  0 – the moment
[c] On as long as entering the bedroom [c] Off before sleeping when the related event happens.
[d] On when feeling hot [d] Off when getting up in the morning
[e] On when sleeping [e] Off when feeling cold
[f] On regularly at (time) [f] Off regularly at (time)
2.3. Occupant behavior simulation module in DeST
[g] Others, please fill in [g] Others, please fill in
The occupant behavior models have been implemented in DeST
at present. The structure of the occupant behavior simulation mod-
2.1. Questionnaire survey ule is shown in Fig. 3.
The models regarding the occupancy, the operations of the light-
A broad questionnaire survey to investigate occupants’ air- ing, the cooling/heating and the window have been implemented.
conditioning behavior in residential buildings was distributed in Building models are built in traditional DeST environment, with
Chengdu, China in the summer of the year 2013. Chengdu is a an additional dialog to input patterns and parameters describing
city located in the hot summer and cold winter zone of China. Its occupants’ presence and behaviors. The attributes of the building
daily average outdoor temperature as well as the daily highest and are stored in a database, from which the room layout is known. In
lowest [13] are shown in Fig. 2. the occupancy and lighting module, the schedules are the outputs
The questionnaire contained demographic questions such as the representing the occupants in the room, the lighting energy con-
number of residents, the education level, and the basic information sumption, and the casual gain of occupants and lighting. Since the
of the buildings, as well as a list of options describing under what calculation of occupancy and lighting is not strongly coupled with
occasion they would switch on or off the air-conditioning in their the thermal simulation of the building, they are simulated prior to
living-rooms and bedrooms. Tables 1 and 2 show the options for the thermal simulation, providing the core program of DeST with
air-conditioning modes in the living-room and bedroom listed in a SQLite database storing the schedules of the occupant number
the questionnaire, respectively: and the lighting energy. The cooling/heating and window behav-
The questionnaire survey was distributed in 287 districts by the iors are more complicated than the occupancy and lighting, as they
local survey team. The surveyed districts were chosen randomly are strongly coupled with the thermal condition in the room. In
to cover the urban area of Chengdu. The total amount of the valid this simulation module, they are discretized into 5 or 10 min as
responses to the questionnaire survey amount was 1426, among a time step. In each time step, the operations on cooling/heating
which 424 respondents answered switching on modes, and 431 and windows are dependent on their previous state, the previous
respondents answered switching off modes in living-rooms, while environmental conditions and the events that have just taken place.
568 respondents answered switching modes, and 568 respon- This module outputs the current on/off states of the cooling/heating
dents answered closing modes in bedrooms. Multiple choices were and the window, which are used as inputs for calculation in the next
allowed in our survey, e.g. occupants might choose “[d] on when time step. Finally, the occupant behavior module outputs the state
feeling hot” and “[e] on when sleeping” at the same time. schedules of the appliances as well as the cooling/heating energy
consumption.
2.2. Air-conditioning behavior model

2.4. Data
The modeling of air-conditioning behavior is presented in Wang
[10] and Ren et al. [14]. Factors influencing occupants’ switch-
The patterns were obtained from the large-scale questionnaire
ing on/off air-conditioning are divided into environment triggered
survey, while the parameters in the corresponding models were
and event triggered. Whether the operation of the air-conditioning
supposed to be decided by measurements. The limitation in the
takes place is presented as a probability correlated with the indoor
parameter determination was that the current measured cases
temperature or the daily event, varying among the patterns.
could not cover all the patterns listed in the questionnaire. For
The pattern “turn on when feeling hot” is considered to be envi-
the patterns that were observed in measurements, the param-
ronment triggered (by the indoor temperature), described as
⎧  t − u k
eters in Ren et al. [9] were adopted. For those that were not
⎪
⎪ observed, a probability was assumed for occupants to take action
⎨ −
l

based on the description of the patterns. In the following part, the
1−e t ≥ u, when occupied
Pon = (1) air-conditioning behavior in the living room is studied, while the
⎪
⎪
⎩ bedroom could be studied following this technical approach.
0 t<u
The amount of each combination of switching on/off modes in
where Pon – the probability for occupants to switch on the air- the questionnaire was counted. The proportions of the mode com-
conditioning; t – the indoor temperature (◦ C); u – the threshold binations are shown in Fig. 4.
 X. Feng et al. / Energy and Buildings 117 (2016) 332–340 335

Fig. 3. Occupant behavior simulation module in DeST.

Fig. 4. Survey results of the air-conditioning behavior in the living room (patterns in Table 1).

Fig. 5. Survey results of the combined air-conditioning behavior in the living room (patterns in Table 1).

The label in the x axis is in correspondence with that in Table 1. A
combination of the switching on and off modes would increase the
total amount of options. From the concern that these two modes
may be related, the following simulations set up both switching
on and off modes in each case. Fig. 5 provides an overview on the
selected modes by the respondents.
It could be seen that the most occupants, accounting for nearly
25%, would switch on their air-conditioning when they feel hot,
while switch it off when they feel cold. However, there are a large
amount of behaviors that does not abide by the indoor temperature
instead of being driven by events.
In the previous measurements, several probability curves were
developed for different patterns[14]. For instance, for the switch-
ing on mode “On when feeling hot”, the curve like Eq. (1) with
the parameters u = 27.75, l = 15.87, k = 2.22 was adopted. For the
switching off mode “Off when feeling cold”, the curve like Eq. (1)
was adopted, with t and u reversed, and the parameters u = 30.25,
l = 152.88, k = 1.30.
2.5. Simulation
A 3-floor residential building model located in Chengdu, China
was built in DeST. Fig. 6 shows the layout of the 2nd floor, consisting
of two households. The envelope was designed to comply with the
current design standard for residential buildings in hot summer
and cold winter zone in China [15].
There are commonly multiple occupants in a household. Since
the purpose of the simulation was to tell apart the difference of
energy consumption among various behavior patterns, only one
occupant was assumed to be in the household in the simulation to
present the effect of occupants’ control on air-conditioning clearly.
This simplification might underestimate the impact of occupant
336 X. Feng et al. / Energy and Buildings 117 (2016) 332–340

Fig. 6. Layout of the 2nd floor of the building model in Chengdu.

Fig. 7. Occupancy in the living-room and bedroom during 3 days.

behavior on energy consumption. However, as it was applied to all
the cases, the comparison of energy consumption among the pat-
terns would still be reliable. The movement of the occupant was
generated following the model developed by Wang et al. [16]. The
occupant was set to frequently stay in the living-room and the bed-
room, with the long-run proportions of 0.4 and 0.45 respectively.
Fig. 7 shows the simulated occupancy during 3 days in both the
living-room and the bedroom.
The lighting could be operated by occupants, which is also
modeled under the current framework. Lighting behavior is
modeled with the pattern “on when feeling dark, off when
leaving or sleeping” for all cases to make the different air-
conditioning behavior patterns comparable. Fig. 8 shows the
simulated lighting state during 3 days in the living-room.
The window is another factor that could have a strong impact
on indoor environment and cooling/heating energy, which have
been described by many models in literature [17–19,5,10]. As is
mentioned before, the air-conditioning and window could have
coupling effect on indoor environment and energy use. The var-
ious window behaviors may conceal the influence of different
air-conditioning patterns. To highlight the air-conditioning behav-
ior, the window behavior was simplified to a unique pattern, i.e.,

Fig. 8. Lighting state in the living-room during 3 days.

 X. Feng et al. / Energy and Buildings 117 (2016) 332–340 337

Table 3 air-conditioning in pattern A, P(B) the probability in pattern B, the

Models and parameters for switching on AC in the simulation case.
probability for occupants to switch on the air-conditioning in the
Switching on modes Switching off modes combined pattern of A and B could be calculated by
[a] Never on pon = 0
[b] Always on during pon = 1 P (A ∪ B) = P(A) + P(B) − P(A) · P(B) (5)
summer
[c] On as long as entering pon = 0.9
Event triggered modes like “on as long as entering the living-
the living-room
room” were not presented in our measurement. In these cases, a
[d] On when feeling hot 
Pon =
 k specific probability was assumed based on the term. In a case where
t−u
− 
1−e
l
t ≥ u, when occupied
“on as long as entering the living-room”, there is a high probability
0 t<u for occupants to turn on the air-conditioning, so the probability
[e] On regularly at (time) pon = 0.9, whent= 14 : 00 was assumed 0.9. While guest visiting is not so common that the
[f] On if there is a guest pon = 0.002 probability was assumed 0.01. The option “[g] Others, please fill
in ” was intended to make sure that the provided options cover
Table 4 most of the cases where occupants would switch on/off the air-
Models and parameters for switching off AC in the simulation case. conditioning. Few respondents chose this option during our survey,
which indicated that most patterns were covered. Since the small
Switching on modes Switching off modes
amount of modes containing option [g], they were removed from
[a] Never off until the poff = 0
simulation cases. For the switching off modes, since options “[b] Off
end of summer
[b] Off when leaving poff =
when leaving the living-room” and “[c] Off when leaving home”
the living-room
tleave
21.91 are not discriminated in the current model, they are considered
−
2.79
1−e
tleave,
when
21.91
leaving the room identically as “Off when leaving the room”.
−
2.79
With the boundary conditions set up, the energy consumption of
[c] Off when leaving poff = 1 − e , when leaving home
home
the patterns derived from the questionnaire survey was simulated.
[d] Off before sleeping poff = 0.8, when going sleeping The simulation results was referred as the basis for typical behavior
poff = pattern categorization. According to Lang et al. [15], the cooling
[e] Off when feeling
cold
  1.30
30.25 − t period for design in hot summer and cold winter zone is from June
− 
152.88 15th to August 31st, which was used as the simulation period in
1−e t ≤ u, when occupied
0 t>u the simulation. The simulation time step was 10 min, in line with
[f] Off regularly at poff = 0.8, whent = 21 : 00 the time step adopted in measurements.
(time)

3. Analysis
the window was assumed to be closed during the simulation. An
air infiltration of 1 ACH [15] was set to represent the window state.
The simulation generated the schedules for the occupancy, the
The analysis was limited to the living-room, as the bedroom
indoor temperature, the on/off state of the air-conditioning, as well
could be studied in the same way. The set-point of the air-
as the cooling load, which were stored in a.csv file. The simulation
conditioning was 26◦ C, while the relative humidity between 0.4 and
results during 3 days for the pattern “[d] On when feeling hot, [c]
0.6, which meant that once the air-conditioning was on, the room
Off when leaving the room” are shown in Fig. 9. It could be seen that
temperature was set to be 26◦ C if the current temperature was
when the room temperature is high, the occupant would turn on
higher than 26◦ C, and the relative humidity remains between 0.4
the air-conditioning to cool the room, maintaining the room tem-
and 0.6 when air-conditioning was on. After the building context,
perature at 26 ◦ C, and when the occupant is about to leave the room,
the occupant and the behavior patterns were specified, the scenar-
the air-conditioning would be switched off. In the case where the
ios with both models and parameters were set up to be simulated,
occupant operates the air-conditioning as “[c] on as long as enter-
which are shown in Tables 3 and 4.
ing the living-room, [d] Off before sleeping”, the occupant tends to
When multiple options were chosen in the questionnaire, the
keep the air-conditioning on for longer time, which could be seen
total probability for occupants to operate the air-conditioning,
from Fig. 10.
which meant at least one event in these options would occur,
Following the same procedure, the air-conditioning use and
should be calculated in terms of the stochastic processes. Let A1 ,
cooling energy consumption in different listed patterns were simu-
A2 , . . ., An denote events, the probability that at least one of them
lated, except for those containing option [g]. The results are shown
would occur is defined as [20]
in Fig. 11, with an ascending order of the energy consumption.
n

n


It is shown that there is a large discrepancy of the cooling
P Ai = P (Ai ) − P Ai Aj + P Ai Aj Ak
i=1
(3) energy consumption between different air-conditioning behavior
i=1
n+1
i<j i<j<k
patterns, ranging from 0 to 29.1 kWh/m2 . In the pattern where
−· · · + (−1) P (A1 A2 · · ·An )
air-conditioning is always on and never off, the cooling load
Moreover, if events Ai and Aj are independent is 29.1 kWh/m2 , while for an occupant who only turns on air-
conditioning when feeling hot and turns it off when leaving home or
P(Ai Aj ) = P(Ai ) · P(Aj ) (4)
feeling cold (mode d,ce), the cooling load is only 9.9 kWh/m2 . Fig. 11
The air-conditioning behavior is expressed to be driven by also illustrates how much energy consumption is contributed by
mutually independent influencing factors in each mode. For each behavior pattern, if the proportion of the patterns in amount
instance, there is no dependent relationship between the room is extrapolated to a large population.
temperature and the time when an occupant enters a room. As the Cluster analysis with the centroid method [21] was introduced
influencing factors could be regarded independent, the occurrence to recognize 5 levels of energy consumption from Fig. 11. The cool-
of air-conditioning behavior in each mode is considered indepen- ing energy consumption of different patterns in the same level was
dent. Therefore, the union of events is calculated using Eqs. (3) and supposed to be close. Table 5 shows the 5 categories and their
(4). Let P(A) denote the probability for occupants to switch on the composition.
338 X. Feng et al. / Energy and Buildings 117 (2016) 332–340

Fig. 9. Simulation results of the air-conditioning operation in the living-room with mode d,c.

Fig. 10. Simulation results of the air-conditioning operation in the living-room with mode c,d.

Fig. 11. Energy consumption with each air-conditioning behavior pattern (patterns in Table 1).

Table 5 Table 6
Five categories of cooling behavior patterns and their composition. Classification of patterns in the questionnaire.

Cluster 1 2 3 4 5 Group Representative modes

Centroid (kWh/m2 ) 0.00 5.37 12.90 20.35 28.32 1 [a] Never on

Size 1 3 22 18
The typical patterns would be extracted from the cooling energy
distribution of the behavior patterns. They were supposed to
present both the extreme values and the majority in amount of the
energy consumption, as well as the brevity of description. Cluster 1
is composed of only one pattern, i.e. [a,b]. However, the switching
off pattern “[b] Off when leaving the living-room” is contradictory
with the switching on pattern “[a] Never on”, therefore [a,b] is mod-
ified to [a]. The switching off pattern “Never off” should be provided
in the future survey. Cluster 2 is composed of 3 patterns, from [f,c]
8 2 [f] On if there is a guest, [c] Off when leaving the room
3 [d] On when feeling hot, [e] Off when feeling cold
4 [c] On as long as entering the living-room, [c] Off when leaving
5 [b] Always on, [a] Never off
to [e,ce], represented by [f,c] as it is relatively significant and brief.
Cluster 3 is composed of 25 patterns, from [d,cdef] to [df,e], rep-
resented by [d,e]. Cluster 4 is composed of 18 patterns, from [c,cd]
to [cd,e], represented by [c,c]. Cluster 5 is composed of 8 patterns,
from [bcd,deb] to [bc,a], represented by [b,a]. The categorization
of behavior patterns is shown in Table 6.These patterns could be
 X. Feng et al. / Energy and Buildings 117 (2016) 332–340 339

Fig. 12. Distribution of the amount and energy consumption of the 5 typical behav- Fig. 13. The frequency distributions for different simulation times of mode d,c.
ior patterns in Table 6.

Table 7
used as the typical occupant air-conditioning behavior patterns for The average cooling energy consumption and the standard deviation for each case
further building energy analysis. of mode d,c.
The distribution of air-conditioning behavior patterns could Simulation times 100 50 30 20 10
thus be represented by there five patterns. The amount and the
Average cooling energy 11.53 11.55 11.58 11.57 11.38
proportion of energy consumption of each typical occupant air- consumption
conditioning behavior pattern are calculated and shown in Fig. 12. (kWh/m2 )
Standard deviation 0.37 0.40 0.34 0.37 0.31
(kWh/m2 )
4. Discussion

This work provides a novel approach to conclude typical behav-

5. Conclusion
ior patterns based on the large-scale questionnaire data from the
perspective of building energy consumption, while the defect is that
A questionnaire survey investigating occupants’ air-
the parameters in the various models should still be determined by
conditioning behavior patterns has been distributed in 2013
measurements and that the intensive simulation are required. For
in Chengdu, China. Based on the questionnaire survey and field
classifying a single behavior, such as the air-conditioning behavior,
measurements, the simulation scenarios were designed to study
the proposed method is affordable in time cost.
the difference of energy consumption caused by the variation in
When multiple behaviors where combination explosion could
occupants’ air-conditioning behavior patterns, aiming at conclude
occur, it may not be so efficient to classify the behavior pat-
several typical behavior patterns to represent both the variety and
terns. Furthermore, the patterns derived from the simulated energy
individuality of occupants in terms of energy consumption.
consumption results are determined artificially, with no standard
The simulation results were analyzed, and finally the occupants’
method to group them yet. This preliminary research on the repre-
switching on/off air-conditioning behavior patterns were grouped
sentation of typical occupant behavior is a starting point to classify
into 5 groups, represented by the following 5 patterns: [a] Never
occupants into groups in terms of building energy consumption.
on; [f] On if there is a guest and [c] Off when leaving the room; [d]
Another issue in this study is the randomness introduced by the
On when feeling hot and [e] Off when feeling cold; [c] On as long as
occupant behavior simulation. Repeated simulations may lead to a
entering the living-room and [c] Off when leaving; [b] Always on
range of energy consumption results as the occupants do not take
and [a] Never off.
actions in a deterministic way. The question is how many times
Further efforts could be made to classify other behaviors into
of simulation are sufficient to eliminate the effect of the random-
typical patterns by applying this method. Meanwhile, some prob-
ness. To deal with this concern, some cases were simulated for 100
lems remain to be solved, such as the intensive simulation when
times, with different random seeds for each time. The simulation
combination of behaviors occurs, as well as the algorithm to rec-
results for 100 times of the mode “[d] On when feeling hot, [c] Off
ognize typical behavior patterns from large sample data regarding
when leaving home” were converted to the frequency distribution,
the behavior and energy use.
as is shown in Fig. 13. The x axis denotes the simulated total cooling
energy consumption, while the y axis denotes the frequency that
Acknowledgements
the energy consumption falls in the range of the x axis. The fre-
quency distributions for different simulation times are presented
The research was supported by the Natural Science Foundation
to determine the appropriate simulation times. The average cool-
of China (No. 51521005). It also benefited from the International
ing energy consumption and the standard deviation of the repeated
Energy Agency Energy in Buildings Communities Program (IEA EBC)
simulation were calculated for each case to provide a quantita-
Annex 66: Definition and Simulation of Occupant Behavior in Build-
tive view into the difference regarding the distribution, collected
ings.
in Table 7.
From both Fig. 13 and Table 7, it could be seen that the frequency
distribution of the cooling energy consumption stays similar as the References
simulation times decrease from 100, until the times reach 10, when
[1] A. Al-Mumin, O. Khattab, G. Sridhar, Occupants behavior and activity patterns
both the average and the standard deviation begin to deviate from influencing the energy consumption in the Kuwaiti residences, Energy Build.
other cases. To avoid this deviation while keep a practical labor 35 (6) (2003) 549–559.
cost, each case was simulated for 20 times in the previous study, [2] C.M. Clevenger, J. Haymaker, The impact of the building occupant on energy
modeling simulations, in: Joint International Conference on Computing and
and the average value was used for the comparison of the energy Decision Making in Civil and Building Engineering, Citeseer, Montreal,
consumption. Canada, 2006, pp. 1–10.
340 X. Feng et al. / Energy and Buildings 117 (2016) 332–340
[3] Z. Li, Y. Jiang, Q. Wei, Survey on energy consumption of air conditioning in
summer in a residential building in Beijing, J. Heat. Vent. Air Cond. 37 (4)
(2007) 46–51.
[4] J.F. Nicol, M.A. Humphreys, A stochastic approach to thermal
comfort-occupant behavior and energy use in buildings, ASHRAE Trans. 110(2).
[5] R. Andersen, V. Fabi, J. Toftum, S.P. Corgnati, B.W. Olesen, Window opening
behaviour modelled from measurements in Danish dwellings, Build. Environ.
69 (2013) 101–113.
[6] R. Yasue, H. Habara, A. Nakamichi, Y. Shimoda, Modeling the occupant
behavior relating to window and air conditioner operation based on survey
results, in: Proceedings of BS2013:13th Conference of International Building
Performance Simulation Association, Chambery, France, 2013, pp. 1451–
1458.
[7] Z. Yu, B.C. Fung, F. Haghighat, H. Yoshino, E. Morofsky, A systematic procedure
to study the influence of occupant behavior on building energy consumption,
Energy Build. 43 (6) (2011) 1409–1417.
[8] S. D’Oca, T. Hong, A data-mining approach to discover patterns of window
opening and closing behavior in offices, Build. Environ. 82 (2014) 726–739.
[9] X. Ren, D. Yan, T. Hong, Data mining of space heating system performance in
affordable housing, Build. Environ. 89 (2015) 1–13.
[10] C. Wang, Simulation Research on Occupant Energy-Related Behaviors in
Building, Tsinghua University, China, November 2014 (Ph.D. thesis).
[11] D. Yan, J. Xia, W. Tang, F. Song, X. Zhang, Y. Jiang, Destłan integrated building
simulation toolkit part i: Fundamentals, Build. Simul. 1 (2008) 95–110.
[12] X. Feng, D. Yan, T. Hong, Simulation of occupancy in buildings, Energy Build.
87 (2015) 348–359.
[13] A. Xiong, Y. Zhu, B. Wang, Q. Li, F. Song, Q. Zhu, R. Wu, Y. Jiang, Special
Meteorological Data Set for Building Thermal Environment Analysis in China,
China Architecture & Building Press, 2005.
[14] X. Ren, D. Yan, C. Wang, Air-conditioning usage conditional probability model
for residential buildings, Build. Environ. 81 (2014) 172–182.
[15] S. Lang, H. Lin, X. Fu, Y. Feng, Design Standard for Energy Efficiency of
Residential Buildings in Hot Summer and Cold Winter Zone, China
Architecture & Building Press, 2010.
[16] C. Wang, D. Yan, Y. Jiang, A novel approach for building occupancy simulation,
Build. Simul. 4 (2011) 149–167.
[17] H.B. Rijal, P. Tuohy, F. Nicol, M.A. Humphreys, A. Samuel, J. Clarke,
Development of an adaptive window-opening algorithm to predict the
thermal comfort, energy use and overheating in buildings, J. Build. Perform.
Simul. 1 (1) (2008) 17–30.
[18] F. Haldi, D. Robinson, Interactions with window openings by office occupants,
Build. Environ. 44 (12) (2009) 2378–2395.
[19] G.Y. Yun, K. Steemers, Time-dependent occupant behaviour models of
window control in summer, Build. Environ. 43 (9) (2008) 1471–1482.
[20] S.M. Ross, et al., Stochastic Processes, vol. 2, John Wiley & Sons New York,
1996.
[21] T. Pang-Ning, M. Steinbach, V. Kumar, et al., Introduction to Data Mining,
Library of Congress, 2006, pp. 74.