Building and Environment 127 (2018) 221–238

Contents lists available at ScienceDirect

Building and Environment

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

Experimental study on occupants' interaction with windows and lights in T

Mediterranean offices during the non-heating season
Federica Naspia, Marco Arnesanob, Lorenzo Zampettib, Francesca Stazic,∗, Gian Marco Revelb,
Marco D'Orazioa
a
Department of Construction, Civil Engineering and Architecture (DICEA), Università Politecnica delle Marche, Via Brecce Bianche, Ancona, 60131, Italy
b
Department of Industrial Engineering and Mathematical Sciences (DIISM), Università Politecnica delle Marche, Via Brecce Bianche, Ancona, 60131, Italy
c
Department of Materials, Environmental Sciences and Urban Planning (SIMAU), Università Politecnica delle Marche, Via Brecce Bianche, Ancona, 60131, Italy

A R T I C L E I N F O A B S T R A C T

Keywords: The modelling of human behaviour is an important challenge for the building sector, as the actions of users have
Occupant behaviour significant impacts on both energy consumption and comfort assessment. In the search for a comprehensive
Behavioural modelling understanding of the behaviour of occupants, many researchers have directed their efforts towards determining
Offices typical patterns and developing models to predict human-building interactions. This study investigates the be-
Windows
haviour of building users during the summer season in offices in Mediterranean climate. Studies focusing on this
Lights
Mediterranean climate
area are still lacking, despite their importance for cooling loads. A survey is conducted using a dedicated sensor
network to monitor environmental variables, and to determine the presence of people and their interactions with
windows and lights in three offices. The driving factors for the actions of users are assessed and behavioural
models are proposed. The results indicate that interactions with windows and lights are driven by both time-
related events and environmental factors, confirming previous findings. A comparison of the proposed models
with others developed for different climate zones suggests that interactions with windows are affected by the
geographic area, while light switching behaviour seems to be very similar for the different case studies. A
simplified approach for the consideration of different user-device interactions is also proposed. This novel
method, developed to evaluate the interactivity between users and building systems, is based on a coefficient of
interactivity, CI. Both the behavioural models and the simplified approach could be introduced into future si-
mulations to improve predictions of energy use in buildings.

1. Introduction
The building sector is responsible for approximately 40% of the
global energy demand [1,2], with the highest energy consumption at-
tributable to commercial buildings, particularly offices and university
buildings [3]. Reducing energy waste and greenhouse gas emissions is a
challenge of primary importance for the building sector, especially
considering the unsatisfying results that have been achieved one year
after the Paris Agreement [4]. To improve energy efficiency, re-
searchers and designers have focused on low energy, passive, and
adaptive buildings [5], which should be able to combine the reduction
in energy consumption with the comfort preferences of users. In these
buildings, human behaviour is crucial. The presence of occupants and
their interactions with building devices (e.g. windows or lights) can
have significant impacts on heating, cooling, and ventilation require-
ments, as well as on the energy demand from lighting and appliances
[6].
Human-building interactions have acquired increasing relevance in
recent years to account for the established gap between simulated and
real energy consumption. This discrepancy typically averages nearly
30% [7], but higher values have been observed [8]. Understanding the
behaviours of users in buildings and converting these into mathematical
models are essential for accurate simulation results and to improve
building design and management, as well as to conserve energy during
building operation.
In recent years, the capabilities of building performance simulation
(BPS) programs have improved considerably (e.g. computing power,
level of detail in modelling, and availability of libraries). However,
inclusion of human components is still lacking. Fig. 1 shows the three
main approaches utilized by BPS programs to simulate the actions of
occupants. The first approach, which is still used most frequently, is the
traditional method. This consists of standard schedules that over-

∗
Corresponding author.
E-mail addresses: f.naspi@staff.univpm.it (F. Naspi), m.arnesano@univpm.it (M. Arnesano), l.zampetti@staff.univpm.it (L. Zampetti), f.stazi@univpm.it (F. Stazi),
g.m.revel@univpm.it (G.M. Revel), m.dorazio@staff.univpm.it (M. D'Orazio).

https://doi.org/10.1016/j.buildenv.2017.11.009
Received 13 September 2017; Received in revised form 2 November 2017; Accepted 5 November 2017
Available online 06 November 2017
0360-1323/ © 2017 Elsevier Ltd. All rights reserved.
F. Naspi et al.
simplify actual human-building interactions and usually result in in-
correct evaluations of the behaviour of users [9].
To better consider the impact of users on design scenarios, the en-
ergy-related behaviours of different occupants have been included in
the models. Users working in the same environmental conditions can
behave very differently as a result of personal preferences, background,
and habits, as well as their responsiveness to energy issues. Some re-
searchers have proposed employing typical ‘behaviour styles’ (e.g. en-
ergy saver, average occupant, and energy waster) to simulate different
levels of energy consciousness [10–12], while others have utilized the
concept of ‘user type’ (i.e. active, medium, and passive user) to re-
produce variability in use of building controls [13–15].
The different behaviour styles can involve device adjustments (e.g.
windows and lights), set-point regulation (both heating and cooling),
and interaction with plug loads. The difference between the three be-
haviour styles arises from the willingness of the occupants to reduce
energy use and to modulate their thermal preferences (e.g. the cooling
set-point is set at 26 °C for the energy saver behaviour style, 24 °C for
the average occupant, and 22 °C for the energy waster [10]). Simula-
tions performed incorporating these behavioural styles revealed that an
energy waster behaviour style can consume more than twice as much
energy as an energy saver style, providing a concrete validation of
human impacts on energy consumption [10,11].
The classification of active, medium, and passive users is de-
termined by the frequency with which occupants interact with devices.
The user type categorization can be assessed by responses to ques-
tionnaires [16,17], by the number of recorded actions [14,18,19], in
relation to preferred set-point temperatures [20], or according to nu-
merical indicators [13]. The user classification method has also been
implemented in simulations, and indicates wide variations in window
and thermostat control patterns [14,18–20].
These approaches improve upon the results from standard sche-
dules. However, they still categorize the behaviour of occupants into
pre-determined classes and, in general, represent boundary behaviours
[10]. Further improvements have been made with the development of
stochastic behavioural models. These are derived from real data, and
should more accurately predict the dynamic interaction between the
users and the building.
222
Building and Environment 127 (2018) 221–238
Fig. 1. The inclusion of the human component in
BPS according to different approaches, from the
lower to the higher level of accuracy.
Such behavioural models are typically developed by incorporating a
large amount of observational data (coming from multiple case studies),
which allows the assessment of statistical correlations between drivers
(i.e. stimuli) and actions.
As users have different adaptation possibilities of depending on the
building type, the adjustments analysed will reflect this. In dwellings,
people can interact freely with their surroundings, while office workers
typically have fewer opportunities to adapt the indoor environment to
their personal comfort preferences. Building management can also in-
fluence behaviour. In naturally ventilated buildings, users can often
modify window and door positions to improve room temperature and
air exchange. In contrast, in mechanically ventilated buildings, users
must passively accept the system operation. Moreover, personal ad-
justments are minimal because metabolic rate and clothing level remain
fairly constant [21]. To determine human interactions, comfort needs to
be investigated in order to account for the perceptions of occupants in
relation to the specific built environment [22,23]. Additionally, the
support of advanced data acquisition tools and techniques [24–26] can
be useful for improving the energy performance of buildings.
Offices have been widely studied and, operation of windows and
lights are the most frequently investigated patterns [27]. Operation of
windows produces consequences for both the thermal comfort of users
and building energy consumption [28], while lighting adjustments af-
fect building electricity consumption, thermal energy loads, and the
visual comfort of occupants [29].
In studies of the stimuli that trigger window adjustments, re-
searchers have found that opening windows can be dependent on CO2
concentration [30], indoor temperature [13], outdoor temperature
[31], or both indoor and outdoor temperatures [32]. Similarly, window
closing actions are related to indoor [33] and outdoor temperature
[13], or to their combination [32].
Window status is affected not only by environmental conditions, but
also by time-related events, habits, and daily routines. A typical pattern
for window use consists of opening the window upon arrival (even for
small time intervals), maintaining the position unchanged during in-
termediate periods, and closing the window upon departure [34]. In
offices, opening actions occur most frequently at the first entrance into
a room and when re-entering the room after lunch [35,36], while
F. Naspi et al. Building and Environment 127 (2018) 221–238

closing actions are strongly connected to the duration of absence, and 2. Methods
thus are most frequent at departures [37]. During intermediate periods,
window positions tend to remain unchanged, as occupants will adjust This study consists of an experimental monitoring campaign that
their thermal sensations and indoor air quality (IAQ) perceptions to the collected data on environmental variables, user presence, and adaptive
indoor environment [38,39]. behaviour. The data was used to develop behavioural models, which
The triggering parameters for lighting switch-on events are related were compared to previously published models, and to study occupant
to illuminance levels, and these events increase with decreases in both interactions. The phases of this survey are given below:
indoor and outdoor illuminance. There are several different methods to
define illuminance (e.g. room, outdoor, and perceived illuminance 1. Monitoring: acquiring indoor environmental parameters, occupancy
level), with the work plane illuminance most frequently adopted patterns, and user adaptive actions on windows and lights;
[40–42], as it is easy to record and accurately reflects the user position 2. Data processing: evaluating the frequency of actions in relation to
inside the room (e.g. sitting at a desk working at a computer). the time of day to assess differences in user behaviour;
In offices, interactions with electric lights are also driven by the 3. Analysis: employing logistic regression analysis to estimate the in-
time of the day and individual habits. Many studies have demonstrated fluence of monitored parameters on device status and using good-
that lights are turned on and off most frequently at arrival [43] and ness-of-fit estimators to assess the level of statistical significance of
departure [44], respectively. Even if the switch-on probability is also the correlations;
influenced by illuminance levels [40], workers often switch on lights 4. Interpretation: comparing the resulting models to those presented in
upon arrival to signal their presence inside the room [45]. The prob- previous studies carried out in different climate zones to assess si-
ability of switch-off is primarily connected to the duration of absence, milarities, and developing a ‘coefficient of interactivity’, CI, to quan-
and lights are turned off independent of the illuminance levels when tify the likelihood of occupants to modify the environment.
users definitively leave their office [41]. It has been noted that actions
during intermediate periods are very limited; as a consequence, lights The first three steps in this approach have been commonly em-
are usually on during working hours [46,47]. ployed by many other researchers [31,53,57–60]. The collection of
In some circumstances, occupant habits and routines can act to environmental and behavioural data and their subsequent processing
support and promote the optimal management of building devices. are essential steps in the development of behavioural models. Logistic
Many studies have highlighted that traditional behaviours can enhance regression analyses have been used by several studies to infer the
the thermal environment, increasing comfort levels and reducing en- probability of an action occurring.
ergy consumption [48]. In particular, a night ventilation strategy can
positively affect the indoor temperature during the summer in a Med- 2.1. Case study
iterranean climate [49]. Ventilation during the night reduces indoor air
temperature peaks, but its effectiveness can vary depending on outdoor The survey was performed in three multi-occupied offices located in
environmental conditions and building thermal mass [50]. the same building (Università Politecnica delle Marche) in the city of
The identification of behavioural patterns and the development of Ancona (Latitude: 43°35′15” N; longitude: 13°31′01” E; altitude:
models to predict occupant behaviour inside buildings are two sig-
nificant areas of focus in the building sector [51]. Office buildings have
been the target of many surveys analysing human behaviours and for
the development of behavioural models. Recent studies [27,52] re-
viewing drivers and models, have also highlighted the effect of the lo-
cation of case studies and the actions considered. Most of the buildings
monitored have been located in continental Europe (e.g. the UK,
Switzerland) or in the East Asia (e.g. China and Japan), while the
Mediterranean region has been scarcely examined [44]. Moreover, very
few studies have presented monitoring campaigns focused on more
than one adaptive action [53], and multi-occupied spaces have also
received limited attention [54].
For the standardisation of behavioural modelling [55,56], it is of
primary importance to have a comprehensive knowledge of occupant
behaviours, which will require filling in the gaps that have been iden-
tified.
To address the identified lacks, the objective of this study is an
evaluation of human-building interactions during the non-heating
season in multi-occupied offices located in the Mediterranean region.
The proposed models are compared to previous findings for different
climatic areas in order to assess the influence of the geographic zone. A
simple method is also presented to classify user interactions with de-
vices and to evaluate behavioural differences in similar spaces.
The identification of behavioural relationships and their in-
corporation into simulations will enhance the simulation results, and
provide improved ability to bridge the gap between real and predicted
energy consumption. Moreover, the inclusion of occupant perspectives
will elucidate decision-making processes and can help designers eval-
uate the best energy retrofit strategy.
Fig. 2. External view of the building and identification of the surveyed rooms.

223
F. Naspi et al. Building and Environment 127 (2018) 221–238

Table 1 2.2. Measurements

Main features of the surveyed spaces.
A monitoring system was set up to record both occupant actions and
Room 1 2 3
environmental parameters. The monitoring began on 3rd May and
Net floor area (m2) 20 20 15 ended on 31st October 2016. The timeframes for the analyses in this
Internal Height (m) 3 3 3 study are defined as from 3rd May to 29th July for the non-heating
Heated volume (m3) 60 60 45
summer period, and from 9th September to 28th October 2016 for the
Ratio S/V 0.33 0.33 0.33
Orientation North North East non-heating autumn period. Table 2a lists the probes in the sensor
Number of persons 3 3 2 network, along with their respective performance ranges.
Glazed surface (m2) 6 6 6 The sampling time for the acquired data series is 10 min. The
Opening surface (m2) 3 3 3 number of occupants and window openings are presented as cumulative
quantities recorded during each 10 min interval. Moreover, each op-
erable window is equipped with two contact sensors to determine
140 m). Fig. 2 shows an external view of the building, and the dashed
whether the window is closed or open. Fig. 3 shows a view of one of the
rectangles indicate the monitored spaces. Built in the 1970s, the uni-
surveyed rooms and identifies elements of the monitoring system.
versity building has a concrete load-bearing structure and is covered by
Outdoor environmental data are obtained from a weather station
double glazed ribbon windows in aluminium frames. The external en-
operated by Marche Civil Defence, located approximately 2 km from
velope is a cavity wall with concrete panels on the façade (approxi-
the case study building. The outdoor air temperature is the only para-
mately 40 cm). Indoor partitions consist of chipboard covered with a
meter obtained from the weather station.
plastic material (8 cm), while the floor slabs are concrete with inter-
Spot measurements of thermal comfort were performed during the
posed insulation. A plasterboard ceiling (about 1 m) contains the
monitoring campaign to provide a more complete assessment of the
electrical air systems.
thermal environment. One recording occurred in the summer (between
In each monitored space, only two casement windows are operable,
20th and 29th July 2016), and one was taken in the mid-season (be-
while the other two are fixed. Table 1 lists the main features of the three
tween 24th October and 7th November 2016). The predicted mean vote
rooms. The offices are occupied by two or three people between 27 and
(PMV) and the Percentage of Person Dissatisfied (PPD) were analysed
34 years of age (mean: 30). Occupancy and number of people vary
by a micro-climate station that acquired the necessary comfort para-
throughout the week and month, as well as between the different
meters using probes positioned according to the guidelines given in ISO
rooms, depending on the activities of the occupants. However, the of-
7726 [61]. The PMV and PPD indexes were evaluated in accordance
fices are typically occupied from Monday to Friday and from 9:00 a.m.
with ISO 7730 [62]. The metrological characteristics of the probes used
to 6:30 p.m., with slight variations in entrance and exit times depending
for these analyses are summarized in Table 2b. During the measurement
on daily schedules.
campaigns, the subjective responses of users were also investigated, and
the resulting mean thermal vote (MTV) was compared with the ob-
tained PMV. The measurement station was located in the centre of the
Table 2 room, near the occupant work stations. Air temperature, humidity,
Main features of the adopted probes. velocity, and mean radiant temperature were acquired at 10 min in-
tervals.
a)Sensor network
The post-processing of acquired data allows for the application of
Parameter Sensor Accuracy Range Number typical profiles for each case using dedicated software. The clothes
worn by occupants were assessed through surveys. The methods out-
Air temperature [°C] Thermistor ± 0.4 °C 0÷70 °C 3 (1 per
lined in ISO 9920 [63], which provides a procedure for the evaluation
(SHT75) room)
Relative Humidity Capacitive ± 1.8% 0÷100% 3 (1 per of clothing insulation and exhaustive tables of values for the thermal
[%] (SHT75) room) insulation of clothing materials (Icl) [m2K/W], were also used to assign
CO2 [ppm] NDIR ± 50 ppm 0÷2000 ppm 3 (1 per clothing levels. A metabolic rate (M) of 1.2 met has been assigned based
room) on the typical activities performed in the office rooms.
Light [lux] Photodiode Si ± 3% 0.02÷20 klx 3 (1 per
room)
To compare objective results with subjective sensations, all subjects
Occupancy [num. PIR n/a 12 m 3 (1 per involved in the survey were required to fill out monthly questionnaires.
occupants] room) The questionnaires were anonymous and were used to acquire personal
Windows Openings Magnetic n/a n/a 12 (4 per information (e.g. age, gender, and typical occupancy profile), recurring
[num.] room)
behaviours (e.g. action frequency and sequence), and thermal sa-
tisfaction.
b)Micro-climate station

Parameter Sensor Accuracy Range

2.3. Numerical methods

Air temperature Pt100 ± 0.1 °C −30C + 100 °C The models proposed in this study are driven by a combination of
[°C] environmental parameters and time-related events, as suggested in the
Mean radiant Globe ± 0.2 °C −30C + 120 °C
literature discussed in the Introduction.
temp. [°C] thermometer
Air velocity [m/ Omni ± 0.02 m/s (0.05–1 m/s) 0.05–5 m/s To perform the analyses, only data collected during occupied per-
s] directional ± 0.1 m/s (1–5 m/s) iods has been considered, as the presence of people inside the room is a
hot-wire precondition for the occurrence of any actions. An action is not only
Relative Capacity ± 2.5% 5 -98%
dependent on occupancy, but also on the device status (e.g. a window
Humidity
[%]
can be opened only if it is currently closed). Therefore, the probability
of an action can be assessed only in relation to these premises. After

224
F. Naspi et al. Building and Environment 127 (2018) 221–238

Fig. 3. Plan view of the three rooms with identification of probes'

position (on the top) and photo view of one of the surveyed rooms.

Table 3 data processing, the environmental variables were discretised into k

Summary of recorded parameters in relation to occupied periods. bins according to the Sturges' rule [64] (i.e. k= (log2 n) +1, where n is
the sample size), and the conditional probability for each bin was cal-
Variable Max Min Mean Median St. Dev.
culated.
OUTDOOR Temperature (°C) 34.6 4.3 19.3 18.9 5.9 The models were developed using a logistic regression analysis,
which is a frequently used approach for dichotomous dependent vari-
ROOM 1 Indoor temperature (°C) 31.4 17.8 24.7 24.6 2.8
ables. The resulting logit probability distribution can be described by
Indoor RH (%) 75.8 29.9 52.3 51.6 8.4
Work-plane illuminance 1655 6 383 353 230.7 the equation:
(lux)
exp(α + β1 x1+⋯+βn x n )
CO2 concentration (ppm) 1301 438 619 548 176.6 p (x1, …, x n ) =
1 + exp(α + β1 x1+⋯+βn x n ) (1)
ROOM 2 Indoor temperature (°C) 32.0 18.9 26.2 26.5 2.9
Indoor RH (%) 70.9 30 50 50.2 7.9 where the vector x1, …,xn is the one of the predictors (i.e. the en-
Work plane illuminance 1999 0 450 403 235.9 vironmental parameters), α is the intercept, and β1, …,βn are coeffi-
(lux)
cients estimated by regression analysis using a maximum likelihood
CO2 concentration (ppm) 1181 380 542 455 178.5
estimation. In this study, a logistic analysis is used to determine the
ROOM 3 Indoor temperature (°C) 34.1 18.2 25.7 25.6 3.1 probability of opening and closing windows and switching on or off
Indoor RH (%) 76.9 29.0 49.6 49.3 8.7 lights for varying environmental triggers.
Work plane illuminance 1684 2 322 268 254.3 Currently, numerous studies [13,16] have assessed the appro-
(lux)
CO2 concentration (ppm) 1084 363 515 458 146.4
priateness of such an approach for estimating the probability of an
adaptive behaviour in response to one or more predictor variables.
ROOMS Indoor temperature (°C) 34.1 17.8 25.4 25.4 3.0 The goodness-of-fit (GOF) of the statistical models has been eval-
TOTAL Indoor RH (%) 75.8 29.0 50.8 50.6 8.5 uated using three different estimators. These assess how well the
Work plane illuminance 1999 0 380 339 246
functions fit the observed data and, as a consequence, evaluate the
(lux)
CO2 concentration (ppm) 1301 363 562 495 173 strength of the relationships. The three GOF estimators calculated for
each correlation in this survey are: the area under the receiver oper-
ating characteristic (ROC) curve (AUC), the McFadden's R2 index, and

225
F. Naspi et al. Building and Environment 127 (2018) 221–238

Table 4 of approximately 50% in all cases. The mean CO2 concentration

PMV and PPD indices for the different periods.
Period Mean PMV
July 1.0
October/November 0.1
the Negelkerke's R2 index [65]. These are all dimensionless estimators,
and better relationships are indicated by higher values.
The ROC curve graphically illustrates the prediction ability of a
binary classifier (e.g. logistic regression). To obtain the curve, the true
positive rate (TPR) and the false positive rate (FPR) are calculated and
combined into a single metric. The AUC, which is the area under the
resulting curve, can have values between 0.5 (no correlation) and 1
(exact prediction). However, values higher than 0.7 are typically con-
sidered acceptable correlations [13]. The pseudo-R2 indexes (i.e.
McFadden's R2 and Negelkerke's R2) are indicators used to define the
predictive capability of models with binary or multinomial outcomes,
similar to the coefficient of determination (R2) for linear regression
models [66]. Many researchers have previously calculated these esti-
mators to assess the probability of logistic regression models [67–69].
3. Results
The following section is split into three sub-sections to provide a
clearer explanation of the survey results. Sub-section 3.1 presents an
overview of the thermal environment during the monitoring period and
an analysis of the frequency of actions at different times of day. Sub-
sections 3.2 and 3.3 present the user interactions with the windows and
the lights, respectively.
3.1. Analysis of thermal environmental and user behaviour
A summary of the indoor and outdoor parameters monitored during
the survey is reported in Table 3. This gives the variation ranges for
each parameter, as well as statistical analyses including the mean,
median, and standard deviation values. These values refer only to
periods in which the rooms were occupied, as occupancy is the main
precondition for the occurrence of actions.
Room 3 had the highest indoor temperatures owing to its orienta-
tion (i.e. east-facing), which provides direct solar radiation in the early
morning. Despite this, the room with the highest mean indoor tem-
perature is Room 2 (26.2 °C). This could suggest a more conservative
use of the environment in Room 2, with less frequent window opening.
Indoor humidity is very similar between the rooms, with a mean value
2 day. According to the literature [37], closing events seem to occur
1.5
1 demonstrating a similar frequency between departure and intermediate
PMV
(562 ppm) is generally acceptable. However, at some points the con-
centrations exceeded the limit of 1000 ppm recommended by the
Mean PPD
ASHRAE. It should be noted that the mean work plane illuminance for
26.5% the three rooms (380 lx) is less than the level (500 lx) recommended by
5.9% EN 12464 [70] for normal office tasks performed at a desk.
Indoor thermal conditions have also been analysed according to the
PMV comfort model and the subjective perceptions of users. The two
spot measurements described in Section 2.2 allowed for estimation of
the PMV and PPD indices. The comfort figures reported in Table 4 re-
present mean values, and refer exclusively to the occupancy data. These
indicate a warm mean sensation in the summer season and a neutral
sensation during the autumn.
For a better understanding of the variation due to occupancy, Fig. 4
also reports data for two typical summer days. The solid and dashed
lines refer to an occupied and empty day, respectively. It can be ob-
served from the fluctuations in the ‘occupied’ trend that occupant ad-
justments and interactions with the surroundings modify the environ-
ment to match their preferred thermal sensation.
Fig. 5 shows the trend in thermal vote obtained from analysis of the
questionnaires, for both the whole sample and sub-divided between the
different rooms. Only June and July present votes deviating from
neutrality (i.e. zero), when a slightly warm sensation prevails, while the
votes for the other months are approximately zero. In general, the three
rooms have similar trends, but the exposure in Room 3 produces higher
indoor temperatures (due to direct solar radiation) and its occupants
more often experience overheating. In contrast, occupants of Room 2
seem to feel cooler sensations. A comparison between the objective
measurements (reported in Table 4) and the subjective sensations
(presented in Fig. 5) indicates a good correspondence, confirming that
occupants are generally thermally satisfied.
The occurrence of actions has been analysed for different times of
day: arrival, intermediate periods, and departure. Herkel et al. [36]
considered an action to have occurred at arrival if it happens within
15 min of the first entry into the room. Similarly, departure included
the 15 min before leaving the office. The remaining time is evaluated as
an intermediate period. Since the monitoring system used in this survey
acquired data at 10 min intervals, a period of 20 min (i.e. 2 time steps)
was used to define arrival and departure.
Fig. 6 presents the absolute frequencies of window opening and
closing separately for each of the three rooms as well as overall. During
the non-heating season, 63% of openings occur at arrival on average
(61% in Room 1, 59% in Room 2, and 68% in Room 3). Actions during
intermediate periods are less frequent, suggesting that people are less
inclined to act because they have adapted to the indoor conditions.
These results are consistent with previous findings [35,36,39], which
have also highlighted the difference in interactions at varying times of
primarily at departure (approximately 60% of events). Such behaviour
is more evident in Rooms 2 (69%) and 3 (63%) than Room 1 (49%),

periods.
0.5
The same evaluation was carried out for turn-on and turn-off events,
0 and the absolute frequencies for light switching are presented in Fig. 7.
Contrary to the literature [45–47,71], light-switching at arrival is not
-0.5 obviously more frequent than during intermediate periods. The per-
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
10.00
11.00
12.00
13.00
14.00
15.00
16.00
17.00
18.00
19.00
20.00
21.00
22.00
23.00

centage difference of roughly 16% is too small to indicate that users are
driven specifically by first entering the room. On the contrary, these
Time [h]
results seem to indicate that users turn the lights on in response to il-
Occupied (28/7/2016) Empty (24/7/2016) luminance levels. Turn-off events during intermediate periods are
negligible (2%); they are almost all triggered by departure. According
Fig. 4. PMV trend in an occupied and non-occupied summer day. to the literature [41,44], once lights are switched on, their status

226
F. Naspi et al. Building and Environment 127 (2018) 221–238

2.0 Fig. 5. Mean thermal vote of the subjects' involved in the survey.

1.5
Mean thermal vote

1.0

0.5

0.0

-0.5

-1.0

-1.5

-2.0
May June July Sept. Oct.
Month
Room 1 Room 2 Room 3 Mean

2=Warm; 1=Slightly warm; 0=Neutral; -1=Slightly cool; 2=Cool

300 300
Opening absolute frequency

Closing absolute frequency

250 250

200 200

150 150

100 100

50 50

0 0
ROOM 1 ROOM 2 ROOM 3 TOTAL ROOM 1 ROOM 2 ROOM 3 TOTAL
Arrival Intermediate Total Departure Intermediate Total

Fig. 6. Absolute frequency for the windows a) opening and b) closing.

typically remains unchanged until users leave the room. 3.2.1. Openings at arrival
In summary, the analysis reported in Figs. 6 and 7 suggests a de- As it was determined that interaction with windows is more fre-
pendence on the period of day for window interactions, but not for quent at arrival than during intermediate periods, logistic regressions
light-switching behaviour. For this reason, only the behavioural models have been performed considering only this time period. In accordance
presented in the following section for window adjustments have been with the literature, four possible driving variables are considered: CO2
developed to incorporate the effects of different periods (i.e. arrivals concentration, indoor temperature, outdoor temperature, and the dif-
and intermediate periods). ference between indoor and outdoor temperature (i.e. ΔT). The logistic
regression coefficients, p-values, and GOF estimators for each correla-
3.2. Modelling window opening and closing tion are listed in Table 5.
The p-values clearly indicate that CO2 concentration and ΔT are not
In this section, trigger parameters for window opening and closing triggers (p-value > 0.001) for window opening. This conclusion is also
actions are investigated and logistic models driven by the identified supported by the GOF estimators, which are very weak. Conversely,
variables are proposed. indoor and outdoor temperatures are clearly recognisable as driving

Fig. 7. Absolute frequency for lights a) switching-on and b) switching-off.

227
F. Naspi et al. Building and Environment 127 (2018) 221–238

Table 5
Logistic regression parameters and goodness-of-fit estimators for window opening models at arrival.

Room Correlations α Z β1 Z p-value AUC McFadden's R2 Neglekerke R2

1 Open-T.in. −11.65 ± 4.03 −5.66 0.49 ± 0.17 5.65 0.000 0.83 0.29 0.44
Open-CO2 5.41 ± 3.34 3.17 −0.01 ± 0.01 −3.14 0.002 0.73 0.08 0.14
Open-T.out. −6.88 ± 2.46 −5.47 0.33 ± 0.12 5.52 0.000 0.83 0.28 0.43
Open- ΔT 0.94 ± 0.6 3.09 −0.32 ± 0.17 −3.78 0.002 0.72 0.10 0.17

2 Open-T.in. −24.47 ± 9.03 −5.311 0.923 ± 0.347 5.22 0.000 0.93 0.55 0.69
Open-CO2 9.25 ± 5.19 3.49 −0.02 ± 0.01 −3.73 0.000 0.79 0.17 0.26
Open-T.out. −12.85 ± 4.62 −5.45 0.52 ± 0.19 5.31 0.000 0.91 0.46 0.61
Open- ΔT −0.03 ± 0.60 −0.09 −0.31 ± 0.18 −3.37 0.001 0.71 0.10 0.16

3 Open-T.in. −8.73 ± 3.25 −5.27 0.34 ± 0.13 5.24 0.000 0.8 0.22 0.35
Open-CO2 7.11 ± 3.71 3.75 −0.02 ± 0.004 −3.77 0.000 0.72 0.12 0.20
Open-T.out. −7.07 ± 3.05 −4.55 0.32 ± 0.14 4.41 0.000 0.84 0.28 0.42
Open- ΔT 0.57 ± 0.76 1.48 −0.13 ± 0.13 −1.97 0.140 0.63 0.02 0.04

Total Open-T.in. −11.10 ± 3.17 −6.86 0.43 ± 0.129 6.54 0.000 0.81 0.24 0.36
Open-CO2 1.99 ± 1.54 2.53 −0.005 ± 0.003 −3.17 0.002 0.61 0.03 0.05
Open-T.out. −7.67 ± 1.62 −9.28 0.34 ± 0.073 9.13 0.000 0.85 0.30 0.45
Open- ΔT 0.17 ± 0.41 0.78 −0.24 ± 0.11 −4.26 0.433 0.68 0.06 0.11

factors, the p-values are less than 0.001 in all cases, and the GOF es-
timators suggest a good statistical correlation. Even if both of these
parameters are significant, the outdoor temperature provides better
results than the indoor. This suggests that outdoor temperature could be
the main stimulus for window opening during arrival. The window
opening models for indoor and outdoor temperatures are reported in
Fig. 8a and b, respectively. The solid lines depict the regression models
for the observational data (cross points). Monotonically increasing
functions relate the opening probability to both indoor and outdoor
temperatures. The T50 (i.e. the temperature at which half of the occu-
pants open a window) is 26 °C indoor and 23 °C outdoor. The graphs
showing the models for each room with both variables are provided in
Appendix A (Fig. A1). These models confirm a more conservative be-
haviour with less frequent window operations in the occupants of Room
2.
3.2.2. Openings during intermediate periods
Window opening events during intermediate periods have been
correlated to indoor temperature and CO2 concentration because when
the windows are closed, the occupants should be influenced only by
indoor conditions. Table 6 reports the regression parameters and GOF
estimators. The p-values are always less than 0.001 for the indoor
temperature, while they are higher for CO2 concentration in both
Rooms 1 and 2. The GOF estimators also underline the weakness of
these correlations, particularly in comparison to those for the indoor
temperature. As a result, only the logistic regression model for the in-
door temperature is shown in Fig. 9. The function has an increasing
trend, although it is much flatter than those obtained for arrival. In fact,
the maximum opening probability barely exceeds 20%. The models for
the individual rooms are provided in Appendix A (Fig. A2).
3.2.3. Closings during intermediate periods
The analysis of closing actions considers only events occurring
during intermediate periods, while the events occurring at the end of
the working day are excluded because they are not environmentally
triggered. Indoor and outdoor temperature are the variables considered
in this evaluation [16,32,34,36]. Table 7 summarizes the regression
parameters and GOF estimators.
The parameters show good statistical correlation, and suggest both
are driving factors for window closings. The GOF estimators indicate
that indoor temperature is the main trigger in Rooms 2 and 3, while the
outdoor temperature is the principal driver in Room 1. The difference in
the GOF is quite small, suggesting that both variables influence occu-
pant behaviour. The overall window closing models are presented in

1 1
a b
0.9 0.9
Window opening probability (%)

Window opening probability (%)

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
18 20 22 24 26 28 30 32 34 36 38 11 13,5 16 18,5 21 23,5 26 28,5 31 33,5
Indoor Temperature (°C) Outdoor Temperature (°C)
Fig. 8. Observational data (cross points) and logistic regressions (solid lines) for window opening behaviour at arrival.

228
F. Naspi et al. Building and Environment 127 (2018) 221–238

Table 6
Logistic regression parameters and goodness-of-fit estimators for window opening models for intermediate period.

Room Correlations α Z β1 Z p-value AUC McFadden's R2 Neglekerke R2

1 Open-T.in. −10.78 ± 3.84 −5.50 0.32 ± 0.16 3.82 0.000 0.66 0.04 0.04
Open-CO2 −2.56 ± 1.52 −3.29 −0.001 ± 0.002 −1.22 0.222 0.54 0.004 0.01

2 Open-T.in. −25.01 ± 6.34 −7.73 0.87 ± 0.25 6.98 0.000 0.89 0.26 0.29
Open-CO2 3.28 ± 2.01 3.2 −0.01 ± 0.004 −5.83 0.001 0.85 0.20 0.22

3 Open-T.in. −14.31 ± 3.67 −7.63 0.42 ± 0.14 5.91 0.000 0.8 0.13 0.14
Open-CO2 0.12 ± 2.07 0.10 −0.007 ± 0.004 −3.53 0.000 0.73 0.07 0.07

Total Open-T.in. −13.10 ± 2.05 −12.53 0.4 ± 0.08 9.52 0.000 0.75 0.09 0.11
Open-CO2 −0.88 ± 0.92 −1.89 −0.004 ± 0.001 −5.55 0.000 0.68 0.04 0.05

1 The work plane illuminance was the selected driver, as it accurately

reflects the user position while working (i.e. typing on a PC while sit-
0.9
ting at a desk). Table 8 summarises the logistic regression parameters
Window opening proability (%)

0.8 and GOF estimators. Work plane illuminance has a very strong statis-
tical correlation. The AUC are all approximately 0.9, indicating very
0.7 significant correlations. Fig. 11 shows the overall model.
0.6 The functions all follow very similar trends, with a decreasing work
plane illuminance corresponding to an increasing probability of a turn-
0.5 on event. The I50 (i.e. the illuminance level at which half of the occu-
pants will turn the lights on) is approximately 30 lx, which is a very low
0.4
illuminance level. Furthermore, when the work plane illuminance is
0.3 lower than 130 lx, the probability of a turn-on event is completely
negligible. These results are in agreement with previous findings
0.2 [40,42], confirming the effects of the selected variables and the trends
in the proposed models. Fig. A4 in Appendix A shows the models for the
0.1
three rooms.
0
18 19.5 21 22.5 24 25.5 27 28.5 30 31.5 33
4. Discussion
Indoor Temperature (°C)
Fig. 9. Observational data (cross points) and logistic regressions (solid lines) for window This study investigates human-building interactions during the non-
opening during intermediate periods. heating season. It focuses on offices in the Mediterranean area, as the
relevant studies in this climatic zone are lacking [44,52]. The adaptive
actions performed on windows and electric lights have been correlated
Fig. 10a and b. Monotonically decreasing functions relate the closing with the time of day and environmental variables, and behavioural
probability to both parameters. The closing probability at indoor tem- models have been proposed. The aim of this study is to assess whether
peratures higher than approximately 24 °C is negligible, as well as at occupants are driven by daily routine and/or by environmental drivers,
outdoor temperatures greater than 21 °C. The models for the individual and to compare the obtained models to others developed for similar
offices are provided in Appendix A (Fig. A3). environments (i.e. offices) but different climates.
Some of the behavioural patterns observed in the case study are in
3.3. Modelling light-switching line with those reported previously [27,29,34,35], while others differ
[45,46].
Given the results presented in Section 3.1, the light-switching For window operations, the surveyed occupants primarily opened
models have been analysed only in relation to environmental triggers. them at arrival, while most of the closing events correspond with

Table 7
Logistic regression parameters for window closing models.

Room Correlations α Z β1 Z p-value AUC McFadden's R2 Neglekerke R2

1 Close-T.in. 9.22 ± 3.07 5.89 −0.55 ± 0.14 −7.97 0.000 0.82 0.158 0.170
Close - T.out. 3.58 ± 1.64 4.28 −0.34 ± 0.08 −8.18 0.000 0.84 0.164 0.177

2 Close-T.in. 14.13 ± 6.63 4.18 −0.71 ± 0.26 −5.32 0.000 0.83 0.160 0.167
Close - T.out. 5.71 ± 3.12 3.6 −0.42 ± 0.14 −5.93 0.000 0.78 0.199 0.207

3 Close-T.in. 8.82 ± 3.55 4.87 −0.53 ± 0.15 −6.83 0.000 0.82 0.180 0.190
Close - T.out. 0.64 ± 1.73 0.72 −0.22 ± 0.08 −5.28 0.000 0.75 0.092 0.098

Total Close-T.in. 8.41 ± 1.95 8.45 −0.51 ± 0.08 −11.95 0.000 0.83 0.161 0.172
Close - T.out. 2.28 ± 1.01 4.40 −0.28 ± 0.05 −11.47 0.000 0.81 0.137 0.146

229
F. Naspi et al. Building and Environment 127 (2018) 221–238

1 1
a b
Window closing probability 0.9 0.9

Window closing probability

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
18 19.5 21 22.5 24 25.5 27 28.5 30 31.5 33 34.5 13.5 15 16.5 18 19.5 21 22.5 24 25.5 27 28.5 30 31.5 33 34.5
Indoor Temperature (°C) Outdoor Temperature (°C)
Fig. 10. Observational data (cross points) and logistic regressions (solid lines) for window closing behaviour.

departures. The less frequent interaction during the intermediate per- 1

iods suggests that people accustom themselves to the indoor environ-
0.9
ment, and are less sensitive to indoor conditions [38]. In contrast, at
arrival, occupants perceive the difference between indoor and outdoor 0.8
conditions (e.g. fresh outdoor air vs. indoor stuffiness) more strongly
Switch-on probability (%) 0.7
[72]. It should be noted that the highest frequency of closing events
occurring at departure could be a consequence of the survey period. In 0.6
the summer, the typically high temperatures rarely trigger closing ac-
tions, and they occur only at the exit of all the users from the room. 0.5
Such results are in agreement with the typical patterns identified by
0.4
previous studies. These have found that the majority of window
openings occur at arrival, after which the window status is unchanged 0.3
until departure, when all open windows are closed [34–39].
Findings concerning light switching behaviours differ in some in- 0.2
stances from those reported previously. In this case study, the users did
0.1
not turn the lights on when entering the room independent of the il-
luminance levels [45–47,71], but were highly triggered by work plane 0
illuminance levels. However, once the lights are turned on, occupants 10 30 50 70 90 110 130 150 170 190 210 230 250
leave their status unchanged until departure. This behaviour is the same Work-plane illuminance (lux)
as has been reported by other researchers [40,41,44,73]. Fig. 11. Observational data (cross points) and logistic regressions (solid lines) for light-
The variables identified as driving factors agree with those re- switching behaviours.
cognised by many previous studies [27,29,74,75]. Indoor and outdoor
temperatures are the key parameters for window adjustments. Outdoor Light switching behaviours are driven by decreases in work plane
temperature seems to be the most influential driver for window opening illuminance. This variable seems to accurately predict interactions with
at arrival, even though the difference in GOF estimators from the indoor lights, and the results confirm the effectiveness of its use [40,42].
temperature results is minimal. This result suggests that at arrival users The identification of triggering parameters for the adaptive actions
are still influenced by the perception of outdoor conditions. of users allowed the development of behavioural models to predict
Indoor temperature is clearly the main stimulus for opening actions interactions with windows and lights. To evaluate similarities with
during intermediate periods. Conversely, CO2 concentration does not previous studies, the proposed models are compared to those developed
trigger window openings because people can scarcely perceive its in- in office buildings in different climates.
crease [32,76,77]. Fig. 12 shows several behavioural models for the window opening
Closing events are functions of both indoor and outdoor tempera- probability in relation to outdoor temperature (Fig. 12a) [16,78] and
tures, but the former parameter seems to be the most influential [34]. indoor temperature (Fig. 12b) [13,37]. The models for arrival and

Table 8
Logistic regression parameters for switching on models.

Room Correlations α Z β1 Z p-value AUC McFadden's R2 Neglekerke R2

1 Turn on-Lux −0.45 ± 0.30 −1.47 −0.02 ± 0.01 −7.57 0.000 0.90 0.258 0.305
2 Turn on-Lux 1.00 ± 1.02 1.93 −0.03 ± 0.01 −6.72 0.000 0.96 0.539 0.549
3 Turn on-Lux 1.55 ± 0.64 4.72 −0.06 ± 0.01 −10.86 0.000 0.93 0.422 0.459
Total Turn on-Lux −0.2 ± 0.39 −0.99 −0.03 ± 0.004 −12.14 0.000 0.92 0.335 0.355

230
F. Naspi et al. Building and Environment 127 (2018) 221–238

Fig. 12. Comparison between different window opening models in relation to a) outdoor and b) indoor temperature.

Table 9
Evaluation of interactivity with windows and lights in each room in relation to the time
the room was occupied.

Room 1 Room 2 Room 3

Lights ON 60 19 89
Lights OFF 61 17 89
Win OPEN 122 41 95
Win CLOSE 126 68 93
Total actions 369 174 366
Occupied period (h) 709.5 491 651.5
Interactivity Lights ON 0.08 0.04 0.14
Interactivity Lights OFF 0.09 0.03 0.14
Interactivity Win OPEN 0.17 0.08 0.15
Interactivity Win CLOSE 0.18 0.14 0.14
CI 0.52 0.35 0.56
Fig. 13. Comparison between different light switching models, driven by work plane il-
luminance.

offices for 3 summer months (from mid-June to mid-September). De-

intermediate periods reported in the previous sections have been in-
tegrated with the total models (black solid lines) to correspond with the
other models for comparison.
The trends for the outdoor temperature display some similarities, as
openings increase when the temperature also increases. However, the
results show clear differences between the models. The UK models show
an increase in interactions at around 10 °C, while the Italian models
have restricted these to a smaller interval (between 10 and 30 °C), and
have greater slopes than the others. Such differences, however, could be
a consequence of differences in sample size. The model proposed by
Rijal [16] used data from 10 naturally ventilated (NV) buildings, while
Nicol [78] used data from 11 NV buildings across Europe (SCATs da-
tabase).
Similar conclusions can also be made for the models driven by in-
door temperature. In this case, the curves also follow a similar trend,
with the probability increasing with increasing indoor temperature.
Occupants in the case study seem to act consistently on windows only
for temperatures greater than 23 °C during the arrival periods, or 26 °C
in the overall model, while in the other regions, these behaviours begin
at temperatures as low as 15 °C. Both of the models in this study have
been developed using data from a single building, while the wide
survey performed by Haldi and Robinson [13] lasted for 7 years and
included 14 cellular offices. Such large differences in sample size could
be a cause of the discrepancy between the models. The data used by
Yun and Steemers [37] has a more similar context, as they monitored 6
spite similar boundary conditions, the UK and the Italian models differ
markedly.
The overall models have trends very similar to the arrival models,
but they are shifted towards higher temperatures. These results confirm
that occupants at arrival are more sensitive to environmental conditions
and, they adjust windows more frequently as a consequence [35,36].
These comparisons suggest that the climate zone, and thus the
temperatures occupants are used to, plays a significant role in window
adjustments. This also confirms the importance of acquiring data from
many different environmental contexts to obtain an overall under-
standing of user behaviour.
Fig. 13 compares the light-switching probability obtained in this
study (black solid line) with those proposed by Hunt [41] in the UK
(grey solid line) and Wang [42] in China (grey dashed line).
The trends of these three functions are very similar: the switching
probability increases noticeably when the work plane illuminance
reaches approximately 100–130 lx, while at higher illuminance levels it
decreases significantly. The Italian and Chinese models have similar
trends for values greater than 80 lx. The former has a threshold value
(i.e. the value above which the electric lights are always off) around
130 lx, while the latter is approximately 220 lx. However, inside this
interval, the probability is less than 10% for both models. At illumi-
nance levels between 0 and 80 lx, the Italian model resembles the UK
model, and the slopes increase significantly.
Moreover, the functions all indicate that users act on lights at

231
F. Naspi et al.
illuminance values much lower than the recommended minimum value
of 500 lx [70], suggesting that the occupants prefer working with
natural light and turn on the electric lights only when it is absolutely
necessary.
Even though the three models concern different environmental
contexts, numbers of surveyed rooms, and monitoring durations, it can
be concluded that they have very similar trends. The climatic area
seems not to not affect interaction with lights by occupants.
5. Application of findings
Introducing human behaviour to BPS is fundamental for achieving
accurate simulation results and their resulting predictions for energy
consumption and comfort assessment. The implementation of beha-
vioural models in simulations and their coupling with co-simulation is
the best solution to reach these goals. Nevertheless, it is not often
possible to edit source code and implement custom equations, while co-
simulation is supported by only a few programs and its implementation
requires very high levels of expertise with the specific program and
programming language.
While trying to develop and implement a comprehensive approach,
the simplest methods for including user behaviour have been adopted.
Researchers who have surveyed more than one room have noted that
users act on building devices with different frequencies. To account for
this, different user types have been proposed. Occupants are typically
subdivided into active, medium, and passive users, according to the
frequency of their interaction with devices.
Following the approach proposed by Haldi and Robinson [13], a
’coefficient of interactivity‘, CI, is proposed to evaluate the mean number
of actions per unit occupancy time (i.e. hours). While the interactivity
between the different rooms has been compared, the actions ‘per oc-
cupant’ have not been evaluated. In fact, data for the exact number of
occupants for each day of monitoring, and for which user performed
each action, were not available.
Figs. 6 and 7 in Section 3.1 suggested that users in Room 1 and
Room 3 were the most active in relation to windows and lights, re-
spectively. In contrast, users in Room 2 seemed to be very passive, as
suggested by the thermal parameters.
Table 9 reports the number of actions per room for the occupied
hours. These confirm the results of the previous analyses. Room 1 is the
most active in regard to windows (CI = 0.17 for openings and
CI = 0.18 for closings), while Room 3 is the most active on the lights
(CI = 0.14 for both turn-on and turn-off behaviours).
This confirms that the users in Room 2 are the most passive, as they
perform the lowest number of actions and obtain the lowest overall CI
(0.35). It should be also noted that even though the occupants in Room
1 performed a higher total number of actions, the users in Room 3 are
the most active (CI = 0.56).
This analysis suggests that the absolute frequency of actions is not
always a good estimation of user activity, as this parameter is related to
time spent inside the office. Moreover, it is not obvious that the occu-
pants active on one device are also active on another. For example, the
users in Room 1 are the most active on windows, but their interactivity
with lights is much lower.
The analysis proposed by this study is a first step for considering
Appendix A. Logistic regression models for individual rooms
232
Building and Environment 127 (2018) 221–238
human behaviour more accurately, with the aim of enhancing beha-
vioural model development and implementation. Further studies on
user types should take into account these findings, and should parti-
cularly introduce different degrees of interactivity for different devices
to evaluate the most wasteful behaviour.
6. Conclusions
This study investigated occupant behaviour in offices located in a
Mediterranean climate during the non-heating season. The acquisition
of environmental parameters, presence of users, and their adaptive
actions on windows and lights allowed determination of the driving
factors for such behaviours and the development of behavioural
models. A dedicated sensor network allowed for acquisition of all data
over a one year period. The results indicate that window opening and
closing actions are related to the time of day and to both indoor and
outdoor temperatures. Analysis of light-switching behaviours suggests
that turn-on events are a function of decreases in work plane illumi-
nance, while turn-off events are related only to departures.
The proposed models have been compared to previously published
models for office buildings in different climate zones, and the differ-
ences and similarities were evaluated. The window opening models all
have similar shapes, but the Italian model has a higher slope than the
others. This conclusion could be a consequence of the specific mon-
itoring period of this study (i.e. during the non-heating season), which
is limited in comparison to the other surveys. A comparison between
the light switching models suggests that they have very similar corre-
lations, and climatic area seems to not influence user adjustments.
The different behaviours observed for the surveyed rooms provided
for the formulation of a coefficient of interactivity, CI, which weights
the frequency of interactions to the occupancy level. This coefficient
suggests that the absolute frequency is not always the best approach for
evaluating user activity, and that occupants do not have the same de-
gree of interaction with different devices.
The findings of this study provide a further step towards under-
standing occupant behaviours, particularly for the assessment of
human-building interactions in offices located in the Mediterranean
climate. Although many results are in line with those presented pre-
viously, the sample size, monitoring period, and geographic location is
rather limited. Thus, a more extensive investigation would be useful to
confirm the reported results.
To address these limitations, future studies will focus on the ex-
tension of the survey location to other office buildings and on the de-
velopment of behavioural models using data from an entire year of
monitoring. Moreover, the proposed models should be tested in BPS
programs to verify their accuracy and to improve the skills of simula-
tors.
BPS programs are looking for methods to accurately include human
behaviour in the simulation environment. The implementation of be-
havioural approaches is necessary to improve simulation results and
bridge the gap between real and simulated energy consumption. The
improvement of simulations will also aid every aspect of the design
process. In allowing a comparison between different retrofit strategies,
designers will be guided towards the optimal solution for reducing
energy waste and improving user comfort.
 F. Naspi et al.

233
Fig. A1. Observational data (dot points) and logistic regressions (solid lines) for window opening at arrival for each room.
Building and Environment 127 (2018) 221–238
 F. Naspi et al.

234
Fig. A2. Observational data (dot points) and logistic regressions (solid lines) for window opening during intermediate periods for each room.
Building and Environment 127 (2018) 221–238
 F. Naspi et al.

235
Fig. A3. Observational data (dot points) and logistic regressions (solid lines) for window closing during intermediate periods for each room.
Building and Environment 127 (2018) 221–238
 F. Naspi et al.

236
Fig. A4. Observational data (dot points) and logistic regressions (solid lines) for light-switching for each room.
Building and Environment 127 (2018) 221–238
F. Naspi et al.
References
[1] C.H. Pout, F. MacKenzie, R. Bettle, Carbon dioxide Emissions from Non-domestic
Buildings: 2000 and beyond, BRE Electronic Publications, 2002.
[2] WBCSD, Energy Efficiency in Buildings PPP: Multi-annual Roadmap and Longer-
Term Strategy, Technical Report, (2010), http://dx.doi.org/10.2777/10074.
[3] M.H. Chung, E.K. Rhee, Potential opportunities for energy conservation in existing
buildings on university campus: a field survey in Korea, Energy Build. 78 (2014)
176–182, http://dx.doi.org/10.1016/j.enbuild.2014.04.018.
[4] A. Averchenkova, S. Matikainen, Assessing the Consistency of National Mitigation
Actions in the G20 with the Paris Agreement, (2016) http://www.lse.ac.uk/
GranthamInstitute/wp-content/uploads/2016/11/Averchenkova-and-Matikainen-
2016_web.pdf.
[5] T.A. Nguyen, M. Aiello, Energy intelligent buildings based on user activity: a
survey, Energy Build. 56 (2013) 244–257, http://dx.doi.org/10.1016/j.enbuild.
2012.09.005.
[6] D.J. Clements-Croome, Sustainable Intelligent Buildings for Better Health, Comfort
and Well-Being, Report for Denzero Project (2014), pp. 1–54.
[7] H. Poirazis, Å. Blomsterberg, M. Wall, Energy simulations for glazed office buildings
in Sweden, Energy Build. 40 (2008) 1161–1170, http://dx.doi.org/10.1016/j.
enbuild.2007.10.011.
[8] R.V. Andersen, B. Olesen, J. Toftum, Simulation of the effects of occupant behaviour
on indoor climate and energy consumption, Clima 2007 WellBeing Indoors
(2007) 7.
[9] D. Yan, W. O’brien, T. Hong, X. Feng, H.B. Gunay, F. Tahmasebi, A. Mahdavi,
Occupant behavior modeling for building performance simulation: current state and
future challenges, Energy Build. 107 (2015) 264–278, http://dx.doi.org/10.1016/j.
enbuild.2015.08.032.
[10] K. Sun, T. Hong, A framework for quantifying the impact of occupant behavior on
energy savings of energy conservation measures, Energy Build. 146 (2017)
383–396, http://dx.doi.org/10.1016/j.enbuild.2017.04.065.
[11] T. Hong, H.W. Lin, Occupant Behavior: Impact on Energy Use of Private Offices,
Lawrence Berkeley Natl. Lab, Berkeley, CA, 2013https://escholarship.org/uc/item/
6jp5w8kn#page-11.
[12] S. Karjalainen, Should we design buildings that are less sensitive to occupant be-
haviour? a simulation study of effects of behaviour and design on office energy
consumption, Energy Effic. 9 (2016) 1257–1270, http://dx.doi.org/10.1007/
s12053-015-9422-7.
[13] F. Haldi, D. Robinson, Interactions with window openings by office occupants,
Build. Environ. 44 (2009) 2378–2395, http://dx.doi.org/10.1016/j.buildenv.2009.
03.025.
[14] G.Y. Yun, P. Tuohy, K. Steemers, Thermal performance of a naturally ventilated
building using a combined algorithm of probabilistic occupant behaviour and de-
terministic heat and mass balance models, Energy Build. 41 (2009) 489–499,
http://dx.doi.org/10.1016/j.enbuild.2008.11.013.
[15] C.F. Reinhart, Lightswitch-2002: a model for manual and automated control of
electric lighting and blinds, Sol. Energy 77 (2004) 15–28, http://dx.doi.org/10.
1016/j.solener.2004.04.003.
[16] H.B. Rijal, P.G. Tuohy, M. Humphreys, J.F. Nicol, A.A. Samuel, J. Clarke, Using
results from field surveys to predict the effect of open windows on thermal comfort
and energy use in buildings, Energy Build. 39 (2007) 823–836, http://dx.doi.org/
10.1016/j.enbuild.2007.02.003.
[17] R.K. Andersen, V. Fabi, S.P. Corgnati, Predicted and actual indoor environmental
quality: verification of occupants' behaviour models in residential buildings, Energy
Build. 127 (2016) 105–115, http://dx.doi.org/10.1016/j.enbuild.2016.05.074.
[18] V. Fabi, R.V. Andersen, S.P. Corgnati, Influence of occupant's heating set-point
preferences on indoor environmental quality and heating demand in residential
buildings, HVAC&R Res. 19 (2013) 37–41, http://dx.doi.org/10.1080/10789669.
2013.789372.
[19] S. D'Oca, V. Fabi, S.P. Corgnati, R.K. Andersen, Effect of thermostat and window
opening occupant behavior models on energy use in homes, Build. Simul. 7 (2014)
683–694, http://dx.doi.org/10.1007/s12273-014-0191-6.
[20] S. Wei, R. Jones, S. Goodhew, P. De Wilde, Occupants' space heating behaviour in a
simulation intervention loop, 13th Conf. Int. Build. Perform. Simul. Assoc.
Chambéry, Fr. August 26-28, 2013, pp. 1991–1998 http://www.ibpsa.org/
proceedings/BS2013/p_1152.pdf.
[21] D. Daum, F. Haldi, N. Morel, A personalized measure of thermal comfort for
building controls, Build. Environ. 46 (2011) 3–11, http://dx.doi.org/10.1016/j.
buildenv.2010.06.011.
[22] G.M. Revel, M. Arnesano, Perception of the thermal environment in sports facilities
through subjective approach, Build. Environ. 77 (2014) 12–19, http://dx.doi.org/
10.1016/j.buildenv.2014.03.017.
[23] G.M. Revel, M. Arnesano, Measuring overall thermal comfort to balance energy use
in sports facilities, Measurement 55 (2014) 382–393, http://dx.doi.org/10.1016/j.
measurement.2014.05.027.
[24] G.M. Revel, M. Arnesano, F. Pietroni, Development and validation of a low-cost
infrared measurement system for real-time monitoring of indoor thermal comfort,
Meas. Sci. Technol. 25 (2014), http://dx.doi.org/10.1088/0957–0233/25/8/
085101 10pp.
[25] M. Arnesano, G.M. Revel, F. Seri, A tool for the optimal sensor placement to opti-
mize temperature monitoring in large sports spaces, Autom. Constr. 68 (2015)
223–234, http://dx.doi.org/10.1016/j.autcon.2016.05.012.
[26] O. Vermesan, T. Crosbie, R. Bahr, N. Dawood, G.M. Revel, A wireless sensor net-
work for intelligent building energy management based on multi communication
standards – a case study, J. Inf. Technol. Constr. 17 (2012) 43–62 http://www.
237
Building and Environment 127 (2018) 221–238
itcon.org/2011/3.
[27] F. Stazi, F. Naspi, M. D'Orazio, A literature review on driving factors and contextual
events influencing occupants' behaviours in buildings, Build. Environ. 118 (2017)
40–66, http://dx.doi.org/10.1016/j.buildenv.2017.03.021.
[28] S. Barlow, D. Fiala, Occupant comfort in UK offices-How adaptive comfort theories
might influence future low energy office refurbishment strategies, Energy Build. 39
(2007) 837–846, http://dx.doi.org/10.1016/j.enbuild.2007.02.002.
[29] V. Fabi, R.K. Andersen, S. Corgnati, Accounting for the uncertainty related to
building occupants with regards to visual comfort: a literature survey on drivers and
models, Buildings 6 (2016), http://dx.doi.org/10.3390/buildings6010005.
[30] P. Warren, L. Parkins, Window-opening behaviour in office buildings, Build. Serv.
Eng. Res. Technol. 5 (1984) 89–101.
[31] N. Li, J. Li, R. Fan, H. Jia, Probability of occupant operation of windows during
transition seasons in office buildings, Renew. Energy 73 (2015) 84–91, http://dx.
doi.org/10.1016/j.renene.2014.05.065.
[32] K. Schakib-Ekbatan, F.Z. Çakıcı, M. Schweiker, A. Wagner, Does the occupant be-
havior match the energy concept of the building? – Analysis of a German naturally
ventilated office building, Build. Environ. 84 (2015) 142–150, http://dx.doi.org/
10.1016/j.buildenv.2014.10.018.
[33] H.B. Rijal, P.G. Tuohy, J.F. Nicol, M.a. Humphreys, A.a.a. Samuel, J.a. Clarke,
Development of adaptive algorithms for the operation of windows, fans and doors
to predict thermal comfort and energy use in Pakistani buildings, ASHRAE Trans.
114 (2008) 555–573 http://strathprints.strath.ac.uk/9382/.
[34] 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, http://dx.doi.
org/10.1016/j.buildenv.2014.10.021.
[35] Y. Zhang, P. Barrett, Factors influencing the occupants' window opening behaviour
in a naturally ventilated office building, Build. Environ. 50 (2012) 125–134, http://
dx.doi.org/10.1016/j.buildenv.2011.10.018.
[36] S. Herkel, U. Knapp, J. Pfafferott, Towards a model of user behaviour regarding the
manual control of windows in office buildings, Build. Environ. 43 (2008) 588–600,
http://dx.doi.org/10.1016/j.buildenv.2006.06.031.
[37] G.Y. Yun, K. Steemers, Time-dependent occupant behaviour models of window
control in summer, Build. Environ. 43 (2008) 1471–1482, http://dx.doi.org/10.
1016/j.buildenv.2007.08.001.
[38] M. Santamouris, A. Synnefa, M. Asssimakopoulos, I. Livada, K. Pavlou,
M. Papaglastra, N. Gaitani, D. Kolokotsa, V. Assimakopoulos, Experimental in-
vestigation of the air flow and indoor carbon dioxide concentration in classrooms
with intermittent natural ventilation, Energy Build. 40 (2008) 1833–1843, http://
dx.doi.org/10.1016/j.enbuild.2008.04.002.
[39] R. Fritsch, a. Kohler, M. Nygård-Ferguson, J.-L. Scartezzini, A stochastic model of
user behaviour regarding ventilation, Build. Environ. 25 (1990) 173–181, http://
dx.doi.org/10.1016/0360-1323(90)90030-U.
[40] C.F. Reinhart, K. Voss, Monitoring manual control of electric lighting and blinds,
Light. Res. Technol. 35 (2003) 243–258, http://dx.doi.org/10.1191/
1365782803li064oa.
[41] D.R.G. Hunt, The use of artificial lighting in relation to daylight levels and occu-
pancy, Build. Environ. 14 (1979) 21–33, http://dx.doi.org/10.1016/0360-1323(79)
90025-8.
[42] C. Wang, D. Yan, H. Sun, Y. Jiang, A generalized probabilistic formula relating
occupant behavior to environmental conditions, Build. Environ. 95 (2016) 53–62,
http://dx.doi.org/10.1016/j.buildenv.2015.09.004.
[43] S. Pigg, M. Eilers, I. Consultant, J. Reed, T. Works, Behavioral aspects of lighting
and occupancy sensors in privates offices: a case study of a university office
building, ACEEE 1996 Summer Study Energy Effic. Build. (1996) 161–170.
[44] P. Correia da Silva, V. Leal, M. Andersen, Occupants interaction with electric
lighting and shading systems in real single-occupied offices: results from a mon-
itoring campaign, Build. Environ. 64 (2013) 152–168, http://dx.doi.org/10.1016/j.
buildenv.2013.03.015.
[45] P.R. Boyce, J.a Veitch, G.R. Newsham, C.C. Jones, J. Heerwagen, M. Myer,
C.M. Hunter, Occupant use of switching and dimming controls in offices, Light. Res.
Technol. 38 (2006) 358–376, http://dx.doi.org/10.1177/1477153506070994.
[46] T. Moore, D.J. Carter, A. Slater, Long-term patterns of use of occupant controlled
office lighting, Light. Res. Technol. 35 (2003) 43–59, http://dx.doi.org/10.1191/
1477153503li061oa.
[47] G.Y. Yun, H.J. Kong, H. Kim, J.T. Kim, A field survey of visual comfort and lighting
energy consumption in open plan offices, Energy Build. 46 (2012) 146–151, http://
dx.doi.org/10.1016/j.enbuild.2011.10.035.
[48] A. Michael, D. Demosthenous, M. Philokyprou, Natural ventilation for cooling in
mediterranean climate: a case study in vernacular architecture of Cyprus, Energy
Build. 144 (2017) 333–345, http://dx.doi.org/10.1016/j.enbuild.2017.03.040.
[49] V. Geros, M. Santamouris, A. Tsangrasoulis, G. Guarracino, Experimental evaluation
of night ventilation phenomena, Energy Build. 29 (1999) 141–154, http://dx.doi.
org/10.1016/S0378-7788(98)00056-5.
[50] M. Santamouris, Building Ventilation the State of the Art, Routledge, 2006, http://
dx.doi.org/10.4324/9781849770620.
[51] IEA EBC, Annex 66: Definition and Simulation of Occupant Behavior in Buildings,
(2013) http://www.annex66.org/.
[52] Z. Belafi, T. Hong, A library of building occupant behavior models represented in a
standardized schema, BEHAVE 2016 4th Eur. Conf. Behav. Energy Effic.
(2016) 8–9.
[53] F. Haldi, D. Robinson, On the behaviour and adaptation of office occupants, Build.
Environ. 43 (2008) 2163–2177, http://dx.doi.org/10.1016/j.buildenv.2008.01.
003.
[54] T. Hong, D. Yan, S. D'Oca, C. Chen, Ten questions concerning occupant behavior in
buildings: the big picture, Build. Environ. 114 (2016) 518–530, http://dx.doi.org/
F. Naspi et al.
10.1016/j.buildenv.2016.12.006.
[55] T. Hong, S. D'Oca, W.J.N. Turner, S.C. Taylor-Lange, An ontology to represent en-
ergy-related occupant behavior in buildings. part I: Introduction to the DNAs fra-
mework, Build. Environ. 92 (2015) 764–777, http://dx.doi.org/10.1016/j.
buildenv.2015.02.019.
[56] T. Hong, S. D'Oca, S.C. Taylor-Lange, W.J.N. Turner, Y. Chen, S.P. Corgnati, An
ontology to represent energy-related occupant behavior in buildings. part II: im-
plementation of the DNAS framework using an XML schema, Build. Environ. 94
(2015) 196–205, http://dx.doi.org/10.1016/j.buildenv.2015.08.006.
[57] D. Calì, R.K. Andersen, D. Müller, B. Olesen, Analysis of occupants' behavior related
to the use of windows in German households, Build. Environ. 103 (2016) 54–69,
http://dx.doi.org/10.1016/j.buildenv.2016.03.024.
[58] R.V. Jones, A. Fuertes, E. Gregori, A. Giretti, Stochastic behavioural models of oc-
cupants' main bedroom window operation for UK residential buildings, Build.
Environ. 118 (2017) 144–158, http://dx.doi.org/10.1016/j.buildenv.2017.03.033.
[59] S. Wei, R. Buswell, D. Loveday, Factors affecting “end-of-day” window position in a
non-air-conditioned office building, Energy Build. 62 (2013) 87–96, http://dx.doi.
org/10.1016/j.enbuild.2013.02.060.
[60] Y. Zhang, P. Barrett, Factors influencing occupants' blind-control behaviour in a
naturally ventilated office building, Build. Environ. 54 (2012) 137–147, http://dx.
doi.org/10.1016/j.buildenv.2012.02.016.
[61] ISO, 7726:2001 Ergonomics of the Thermal Environment - Instruments for
Measuring Physical Quantities, (2001).
[62] ISO, 7730:2006 Ergonomics of the Thermal Environment - Analytical determination
and Interpretation of the Thermal Comfort Using Calculation of the PMV and PPD
Indices and Local Thermal Comfort Criteria, (2006).
[63] ISO, 9920:2007, Ergonomics of the Thermal Environment - Estimation of Thermal
Insulation and Water Vapour Resistance of a Clothing Ensemble, (2007).
[64] D.W. Scott, Sturges' rule, Wiley Interdiscip. Rev. Comput. Stat. 1 (2009) 303–306,
http://dx.doi.org/10.1002/wics.35.
[65] F. Stazi, F. Naspi, M. D'Orazio, Modelling window status in school classrooms. re-
sults from a case study in Italy, Build. Environ. 111 (2017) 24–32, http://dx.doi.
org/10.1016/j.buildenv.2016.10.013.
[66] B. Hu, J. Shao, M. Palta, Pseudo-R2 in logistic regression model, Stat. Sin. 16 (2006)
847–860.
238
Building and Environment 127 (2018) 221–238
[67] T. Domencich, D. McFadden, Urban Travel Demand: a Behavioral Analysis, North-
Holland Publishing, 1975, p. 1975.
[68] F. Haldi, D. Robinson, Modelling occupants' personal characteristics for thermal
comfort prediction, Int. J. Biometeorol. 55 (2011) 681–694, http://dx.doi.org/10.
1007/s00484-010-0383-4.
[69] F. Haldi, D. Robinson, A comprehensive stochastic model of blind usage: theory and
validation, Build. Simul. Conf. 2009, pp. 545–552 Glasgow, Scotland.
[70] European commetee for Standardization, EN 12464-1:2011 Light and Lighting -
Lighting of Work Places - Part 1: Indoor Work Places, (2011).
[71] G.Y. Yun, H. Kim, J.T. Kim, Effects of occupancy and lighting use patterns on
lighting energy consumption, Energy Build. 46 (2012) 152–158, http://dx.doi.org/
10.1016/j.enbuild.2011.10.034.
[72] A. Roetzel, A. Tsangrassoulis, U. Dietrich, S. Busching, A review of occupant control
on natural ventilation, Renew. Sustain. Energy Rev. 14 (2010) 1001–1013, http://
dx.doi.org/10.1016/j.rser.2009.11.005.
[73] D. Lindelöf, N. Morel, A field investigation of the intermediate light switching by
users, Energy Build. 38 (2006) 790–801, http://dx.doi.org/10.1016/j.enbuild.
2006.03.003.
[74] V. Fabi, R.V. Andersen, S. Corgnati, B.W. Olesen, Occupants' window opening be-
haviour: a literature review of factors influencing occupant behaviour and models,
Build. Environ. 58 (2012) 188–198, http://dx.doi.org/10.1016/j.buildenv.2012.07.
009.
[75] K. Van Den Wymelenberg, Patterns of occupant interaction with window blinds: a
literature review, Energy Build. 51 (2012) 165–176, http://dx.doi.org/10.1016/j.
enbuild.2012.05.008.
[76] M. Schweiker, F. Haldi, M. Shukuya, D. Robinson, Verification of stochastic models
of window opening behaviour for residential buildings, J. Build. Perform. Simul. 5
(2012) 55–74, http://dx.doi.org/10.1080/19401493.2011.567422.
[77] F. Stazi, F. Naspi, G. Ulpiani, C. Di Perna, Indoor air quality and thermal comfort
optimization in classrooms developing an automatic system for windows opening
and closing, Energy Build. 139 (2017) 732–746, http://dx.doi.org/10.1016/j.
enbuild.2017.01.017.
[78] J.F. Nicol, Characterising occupant behavior in buildings: towards a stochastic
model of occupant use of windows, lights, blinds heaters and fans, Seventh Int.
IBPSA Conf. (2001) 1073–1078.