《专业英语》课程教学资源（阅读资料）综述论文：A critical review of observation studies, modeling, and simulation of adaptive occupant behaviors in offices

Building and Environment 70(2013)31-47 Contents lists available at ScienceDirect Building and Environment ELSEVIER journal homepage:www.elsevier.com/locate/buildenv A critical review of observation studies,modeling,and simulation of CrossMark adaptive occupant behaviors in offices H.Burak Gunay,William O'Brien,Ian Beausoleil-Morrison b Carleton University,Department of Civil and Environmental Engineering.Canada Carleton Universiry.Department of Mechanical and Aerospace Engineering.Canada ARTICLE INFO ABSTRACT Article history: Occupants'behaviors account for significant uncertainty in building energy use.A better understanding Received 2 June 2013 of occupant behaviors is needed in order to manage this uncertainty;as such many studies have been Received in revised form dedicated to this topic.The current paper reviewed the research on adaptive occupant behaviors by 25July2013 sorting it into three categories.The first group encompasses all observational studies.The second group Accepted 31 July 2013 includes modeling studies.The third group incorporates the simulation studies.The current paper Keywords: presents the methodologies used in these studies,discusses the limitations associated with their application,and develops recommendations for future work.Generalized linear models-in particular Adaptive occupant behaviors Behavioral modeling logistic regression models-were found to be appropriate for modeling occupant behavior.Reversal of Occupant control of indoor environment adaptive behaviors (e.g.window closing)was modeled with deadband models or survival models Review Occupant models were typically simulated as discrete-time Markov processes.It was concluded that with appropriate selection of building geometry and materials and occupant-predicting control strategies. impact of occupant behaviors on the building performance can be reduced. 2013 Elsevier Ltd.All rights reserved. 1.Introduction affected if they have less control over their environment [8.12]. CIBSE [13]and ASHRAE [14]acknowledge this by including adap- Building Performance Simulation(BPS)based design,despite its tive comfort models for naturally ventilated buildings.Occupants potential for significant improvements in energy use and indoor can also adapt their personal characteristics such as adjusting their environment,has often been undermined by predictions that do typical beverage temperatures,location,posture,activity and not fully represent actual performance [1,2.Some of these dis- clothing levels.These personal adaptive behaviors can be restricted crepancies can be attributed to deviations from standard weather with social factors such as workplace dress codes however,even in data [3],modeling and simulation simplifications [4],occupancy the most sealed and fully conditioned buildings there are some profiles [5-7].unanticipated control behavior,and material/work- adaptive opportunities. manship related uncertainties.However,the uncertainty intro- Adaptive actions,aside from their impact on perceived comfort, duced by occupant behaviors are undeniable[8.9. often have significant impacts on energy use.Therefore,building Occupants adapt their environment and personal characteristics designers should foresee these occupant-use related impacts on to achieve their comfort in ways that are convenient to them rather energy consumption and incorporate them into design.However. than being necessarily energy-conserving [2,10.11].Environmental building designers tend to make static assumptions about occupant adjustments may involve decisions such as window/door opening, behavior,whereas field studies have indicated that occupants may blind/shade positioning,light switch on/off,carpet/hardwood floor act in unexpected ways and respond to crises of discomfort [2.15] covering,fan on/off,and thermostat up/down.In a given building. For example,an occupant may add carpet or hardwood flooring on occupants may or may not be given control over these actions,but top of concrete in a passive solar house;failure to consider this it was reported that occupants'comfort perception is negatively action will lead to inaccurate BPS predictions16.A better un- derstanding of occupant behaviors(aside from being a promising way to test buildings with expected occupant actions during the Corresponding author.Carleton University,Department of Civil and Environ- mental Engineering.1125 Colonel by Drive.Ottawa,Ontario K1S 5B6.Canada. design stage)has been recently acknowledged as a promising way Tel:+16135202600x8037:fax:+16135203951. to operate buildings [16].Clarke et al.[16].Thrun [17],Claridge and E-mail address:Liam_OBrien@carleton.ca (W.O'Brien). Abushakra [18],Guillemin and Molteni [19,20]and Dong et al.[21] 0360-1323/$-see front matter 2013 Elsevier Ltd.All rights reserved. http://dx.doiorg/10.1016/j.buildenv.2013.07.020

A critical review of observation studies, modeling, and simulation of adaptive occupant behaviors in offices H. Burak Gunay a , William O’Brien a,*, Ian Beausoleil-Morrison b a Carleton University, Department of Civil and Environmental Engineering, Canada b Carleton University, Department of Mechanical and Aerospace Engineering, Canada article info Article history: Received 2 June 2013 Received in revised form 25 July 2013 Accepted 31 July 2013 Keywords: Adaptive occupant behaviors Behavioral modeling Occupant control of indoor environment Review abstract Occupants’ behaviors account for significant uncertainty in building energy use. A better understanding of occupant behaviors is needed in order to manage this uncertainty; as such many studies have been dedicated to this topic. The current paper reviewed the research on adaptive occupant behaviors by sorting it into three categories. The first group encompasses all observational studies. The second group includes modeling studies. The third group incorporates the simulation studies. The current paper presents the methodologies used in these studies, discusses the limitations associated with their application, and develops recommendations for future work. Generalized linear models e in particular, logistic regression models e were found to be appropriate for modeling occupant behavior. Reversal of adaptive behaviors (e.g. window closing) was modeled with deadband models or survival models. Occupant models were typically simulated as discrete-time Markov processes. It was concluded that with appropriate selection of building geometry and materials and occupant-predicting control strategies, impact of occupant behaviors on the building performance can be reduced.
2013 Elsevier Ltd. All rights reserved. 1. Introduction Building Performance Simulation (BPS) based design, despite its potential for significant improvements in energy use and indoor environment, has often been undermined by predictions that do not fully represent actual performance [1,2]. Some of these dis￾crepancies can be attributed to deviations from standard weather data [3], modeling and simulation simplifications [4], occupancy profiles [5e7], unanticipated control behavior, and material/work￾manship related uncertainties. However, the uncertainty intro￾duced by occupant behaviors are undeniable [8,9]. Occupants adapt their environment and personal characteristics to achieve their comfort in ways that are convenient to them rather than being necessarily energy-conserving [2,10,11]. Environmental adjustments may involve decisions such as window/door opening, blind/shade positioning, light switch on/off, carpet/hardwood floor covering, fan on/off, and thermostat up/down. In a given building, occupants may or may not be given control over these actions, but it was reported that occupants’ comfort perception is negatively affected if they have less control over their environment [8,12]. CIBSE [13] and ASHRAE [14] acknowledge this by including adap￾tive comfort models for naturally ventilated buildings. Occupants can also adapt their personal characteristics such as adjusting their typical beverage temperatures, location, posture, activity and clothing levels. These personal adaptive behaviors can be restricted with social factors such as workplace dress codes however, even in the most sealed and fully conditioned buildings there are some adaptive opportunities. Adaptive actions, aside from their impact on perceived comfort, often have significant impacts on energy use. Therefore, building designers should foresee these occupant-use related impacts on energy consumption and incorporate them into design. However, building designers tend to make static assumptions about occupant behavior, whereas field studies have indicated that occupants may act in unexpected ways and respond to crises of discomfort [2,15]. For example, an occupant may add carpet or hardwood flooring on top of concrete in a passive solar house; failure to consider this action will lead to inaccurate BPS predictions [16]. A better un￾derstanding of occupant behaviors (aside from being a promising way to test buildings with expected occupant actions during the design stage) has been recently acknowledged as a promising way to operate buildings [16]. Clarke et al. [16], Thrun [17], Claridge and Abushakra [18], Guillemin and Molteni [19,20] and Dong et al. [21] * Corresponding author. Carleton University, Department of Civil and Environ￾mental Engineering, 1125 Colonel by Drive, Ottawa, Ontario K1S 5B6, Canada. Tel.: þ1 613 520 2600x8037; fax: þ1 613 520 3951. E-mail address: Liam_OBrien@carleton.ca (W. O’Brien). Contents lists available at ScienceDirect Building and Environment journal homepage: www.elsevier.com/locate/buildenv 0360-1323/$ e see front matter
2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.buildenv.2013.07.020 Building and Environment 70 (2013) 31e47

2 H.B.Gunay et aL Building and Environment 70(2013)31-47 have been pioneering this approach to retrieve occupancy-related simulation studies.In these studies,occupant behavior models information using inverse models which later can be utilized to were simulated (e.g.discrete-time Markov Chains)with the create intelligent(i.e.learning.predicting,and adapting)control building energy models to predict the energy impacts of occupants strategies. behaviors for adapting building design and control.The current The existing scientific literature on which these pioneering paper presents the methodologies used in these studies,discusses research efforts were based covers a broad range of methodologies the limitations associated with their application,and develops to study adaptive occupant behaviors.However,existing review recommendations for future work.Due to substantial contextual papers on occupant behaviors give high resolution insights into differences,occupant behaviors in residential buildings,although particular adaptive behaviors such as only manual control of win- they account for about the same amount of energy use [27].were dows [22,23].window shading devices [24,25]or lighting [20.26] not included in this paper.These contextual differences can be with an emphasis on the observational methodologies and their explained with the responsibility of energy bills,need for privacy. limitations.In this paper,a comprehensive,yet broad,approach social factors,type of activities/task,et cetera.The long-term was taken to cover common findings and limitations of the occu- objective of this research project is to develop building design pant behavior research in general with an equal emphasis on the and operation strategies which better account for occupants'be- observational,modeling,and simulation methodologies haviors,habits,and preferences The current paper reviewed the research on adaptive occupant behaviors in offices by sorting it into three categories as shown in 2.System observation Fig.1.These categories were formed to represent the logical flow of research approach for any phenomena:observe-model To assess the adaptive actions of occupants,researchers have simulate.This will help revealing the research needed from each observed a system to be able to correlate a state(e.g.window po- category.The first group encompasses all observational studies.In sition)with a set of variables (e.g.indoor air temperature).The these studies,researchers observed a system (e.g.naturally venti- validity of extending the conclusions of these observations to lated office building)for a period of time(e.g.heating season)in another context may be restricted to the characteristics of the order to develop a correlation between the observed state (e.g. observed building envelope and operation [28.Moreover,tech- operable window or window shades)and the monitored variables niques employed to collect information about the adaptive be- (e.g.indoor temperature).The second group includes modeling haviors (e.g.time-lapse photography,sensors)and the monitored studies.In these studies,occupant behavior models were predicted physical (e.g.indoor/outdoor thermal and non-thermal)and non- by assuming an idealized probability distribution(e.g.binomial)via physical (e.g.privacy,view to outside)variables constitute limita- a regression analysis(e.g.logistic)to reveal the predictor variables tions for the future models proposed based on these observations. that drive an adaptive behavior.The third group incorporates the This section identifies the factors that may affect the generality of Adaptive Occupant Behavior State Monitoring State discretization method (e.g.open/closed or open/half-open/closed) System Observation State monitoring method Overview of system (e.g.photography or sensory) (e.g.south-facing office building) State monitoring frequency Size of system (e.g.two per day) (e.g.300 offices) Observation period Variable Monitoring (e.g.heating season) Monitored variables (e.g.workplane illuminance,temperature) Model Prediction Adaptive behavior model type (e.g.logistic regression) Model validation method (e.g.cross-validation) Reversal of adaptive behavior model (e.g.survival models) Simulation Model simulation method (e.g.discrete time Markov Chains) Simulation verification method (e.g.isolate tests) Fig.1.Research and modeling approach on adaptive occupant behavior

have been pioneering this approach to retrieve occupancy-related information using inverse models which later can be utilized to create intelligent (i.e. learning, predicting, and adapting) control strategies. The existing scientific literature on which these pioneering research efforts were based covers a broad range of methodologies to study adaptive occupant behaviors. However, existing review papers on occupant behaviors give high resolution insights into particular adaptive behaviors such as only manual control of win￾dows [22,23], window shading devices [24,25] or lighting [20,26] with an emphasis on the observational methodologies and their limitations. In this paper, a comprehensive, yet broad, approach was taken to cover common findings and limitations of the occu￾pant behavior research in general with an equal emphasis on the observational, modeling, and simulation methodologies. The current paper reviewed the research on adaptive occupant behaviors in offices by sorting it into three categories as shown in Fig. 1. These categories were formed to represent the logical flow of research approach for any phenomena: observe / model / simulate. This will help revealing the research needed from each category. The first group encompasses all observational studies. In these studies, researchers observed a system (e.g. naturally venti￾lated office building) for a period of time (e.g. heating season) in order to develop a correlation between the observed state (e.g. operable window or window shades) and the monitored variables (e.g. indoor temperature). The second group includes modeling studies. In these studies, occupant behavior models were predicted by assuming an idealized probability distribution (e.g. binomial) via a regression analysis (e.g. logistic) to reveal the predictor variables that drive an adaptive behavior. The third group incorporates the simulation studies. In these studies, occupant behavior models were simulated (e.g. discrete-time Markov Chains) with the building energy models to predict the energy impacts of occupants’ behaviors for adapting building design and control. The current paper presents the methodologies used in these studies, discusses the limitations associated with their application, and develops recommendations for future work. Due to substantial contextual differences, occupant behaviors in residential buildings, although they account for about the same amount of energy use [27], were not included in this paper. These contextual differences can be explained with the responsibility of energy bills, need for privacy, social factors, type of activities/task, et cetera. The long-term objective of this research project is to develop building design and operation strategies which better account for occupants’ be￾haviors, habits, and preferences. 2. System observation To assess the adaptive actions of occupants, researchers have observed a system to be able to correlate a state (e.g. window po￾sition) with a set of variables (e.g. indoor air temperature). The validity of extending the conclusions of these observations to another context may be restricted to the characteristics of the observed building envelope and operation [28]. Moreover, tech￾niques employed to collect information about the adaptive be￾haviors (e.g. time-lapse photography, sensors) and the monitored physical (e.g. indoor/outdoor thermal and non-thermal) and non￾physical (e.g. privacy, view to outside) variables constitute limita￾tions for the future models proposed based on these observations. This section identifies the factors that may affect the generality of Fig. 1. Research and modeling approach on adaptive occupant behavior. 32 H.B. Gunay et al. / Building and Environment 70 (2013) 31e47

H.B.Gunay et al.Building and Environment 70 (2013)31-47 33 the observations and develops recommendations that will incor- parameters,despite their importance being qualitatively acknowl- porate the bias introduced by these factors. edged,have not been included in the reviewed literature.Haldi and Robinson [46]suggested building statistical models that incorpo- 2.1.Behaviors that adapt the indoor environment rate non-temperature physical variables as predictors as a future research effort.However,as window openings connect the ambient Occupants adapt their indoor environment with their alter- conditions to the indoor environment,the type of the window ations to operable windows and window shading devices,lights. opening can also play a significant role on the occupant's prefer- fans,carpets,and thermostats.Various research projects have ences.For example,bottom hung inside opening windows provide conducted observational studies where they investigated these weather protection[22.Thus,in this case,wind and rain may be adaptive behaviors using a variety of methods.This section pre- discarded from the monitored variables.But for the side hung or sents these methodologies and discusses the limitations and sliding windows,it may be more appropriate to include the associated challenges. weather variables to the monitored variables. Similarly,to predict a window shade deployment model,indoor 2.1.1.Physical variables variables,such as indoor temperature 29,32,47,49,indoor daylight Researchers have either used their prior knowledge on the 29,32,47,49,50],transmitted solar radiation [29,32,50,51]:and observed system or they carried out questionnaire surveys to nar- outdoor variables,such as outdoor temperature [39,47]and row down the variables that should be measured,such as tem- external solar radiation [49.50,52-54].have been monitored. perature,relative humidity(RH),noise,and workplane illuminance. Zhang and Barrett [47]observed that window shade deployment so that the adaptive occupant behaviors can be predicted with did not follow the outdoor temperature.Arguably.Lindsay and these monitored variables.For example,Inkarojrit [29]carried out a Littlefair [52]and Foster and Oreszczyn55]claimed that indoor survey to identify the main motivations for closing window blinds temperature and external solar radiation cannot be a predictor in private offices.In this study,it was reported that the majority of variable for the window shade deployment.Also,Reinhart [56 occupants who closed their blinds do so to protect their worksta- used only visual/optical variables that may lead to blind lowering. tions and screens from direct or reflected glare from sunlight,while A reasonable explanation for this controversy is that occupants use 27.4%of the participants claimed that they use their blinds to window shades to mitigate both visual and thermal discomfort reduce the heat from the sun and only 12.3%stated privacy and (excluding the non-physical factors such as view or privacy).which security as a reason for blind closure.Eilers et al.[30]surveyed can be caused by temperature,solar radiation,glare,et cetera. office occupants and confirmed that the majority of the subjects Clearly,it is crucial to monitor independent variables that can who closed their blinds do so to reduce the glare on their computer represent the window shade deployment.Reviewed literature screen.Similarly.Warren and Parkins [31]carried out a survey in suggests that these variables are:the depth of penetration of the which occupants stated that lAQ was the main reason for opening direct sunlight as a function of the solar altitude [47,51,57]and glare windows during the heating season and noise was the main reason 52,58,59].Inoue et al.[51]suggested that the depth of penetration for closing windows during the summer season.These surveys, of direct sunlight changes with the mean blind occlusion.This when used prior to the physical measurements,can give pre- observation was later confirmed by Reinhart and Voss [57]with a liminary insight into determining the variables that should be theoretical solar penetration depth measured from the top of the measured and the spatial distribution of sensors that will be placed window.Subsequently,Reinhart and Wienold 60]performed in the office during a study.Furthermore,it can be used to identify representative daylight design strategies that incorporate occu- the subtleties that are difficult to measure (e.g.rattling blinds pants'reaction against direct sunlight and glare. caused by wind passing over deployed venetian blinds)before starting the data collection 32]. 2.1.2.Non-physical parameters Early studies on the window opening behavior started by Physical variables (i.e.thermal,visual,acoustic,indoor air monitoring the outdoor variables [31.33-36]with the following environment)influence the chance that an office occupant will reasoning:(1)once these observations are integrated to BPS as experience discomfort.However,it is his/her social,economic,and occupant models the indoor variables become outputs of the BPS, psychological influences,which are driven by non-physical (i.e. therefore,the indoor variables cannot be more reliable than out- latent)variables that lead to these adaptive actions [61.These non- door variables [37.38];and (2)indoor temperature can be defined physical (latent)variables are parameters that are not measurable as a spatial distribution rather than a single scalar [38]which re- with typical sensors such as view and connection to the outside. quires time consuming and expensive instrumentation of the privacy or daylight-health perception. sensors and data loggers.Subsequently,it was suggested that oc- It is evident that one of the main design purposes of windows is cupants only have an indirect perception of outdoor physical vari- to provide a clear view and physical connection to the outside 60. ables [38].consequently indoor variables (at least the indoor Green building rating systems (e.g.LEED)define a view as "a thermal variables)have been incorporated amongst monitored straight visual connection from an interior point to a point outside variables in many of the recent studies [10,38-41.In particular, through a facade opening located within a certain height range within using a reasonable balance between the indoor and outdoor vari- a facade"[62].Inoue et al.[51]reported that most occupants ables to describe the window opening behavior can be suggested as preferred to have seats close to the windows,however these seats follows:(1)window opening behavior can be described with the were known to be the most susceptible locations to glare and solar indoor variables (e.g.indoor temperature);and (2)window closing radiation.Based on the surveys to study the stimulating factors for behavior can be explained with both indoor and outdoor variables window opening,it was reported that some of the window opening (e.g.indoor and outdoor temperature)[39,42].This would take into behaviors,despite allowing some ambient noise,may be explained account the transmitted effects of the outdoors once the window is to maintain a direct connection to outdoors [63,64].These findings open,while ignoring them once it is closed.Non-temperature can be interpreted as occupants prefer to tolerate some discomfort physical variables that can affect the window opening/closing in order to have a better quality of view and connection to the behavior have been listed as the indoor air quality [31,43-45.the outdoors. outdoor noise level 22.31,46,47],RH39].wind speed and direction Window shading devices may obstruct the view to the outside [22.34,47].and rain [41,47,48].These non-temperature physical Haldi and Robinson[65]carried out a study on window control

the observations and develops recommendations that will incor￾porate the bias introduced by these factors. 2.1. Behaviors that adapt the indoor environment Occupants adapt their indoor environment with their alter￾ations to operable windows and window shading devices, lights, fans, carpets, and thermostats. Various research projects have conducted observational studies where they investigated these adaptive behaviors using a variety of methods. This section pre￾sents these methodologies and discusses the limitations and associated challenges. 2.1.1. Physical variables Researchers have either used their prior knowledge on the observed system or they carried out questionnaire surveys to nar￾row down the variables that should be measured, such as tem￾perature, relative humidity (RH), noise, and workplane illuminance, so that the adaptive occupant behaviors can be predicted with these monitored variables. For example, Inkarojrit [29] carried out a survey to identify the main motivations for closing window blinds in private offices. In this study, it was reported that the majority of occupants who closed their blinds do so to protect their worksta￾tions and screens from direct or reflected glare from sunlight, while 27.4% of the participants claimed that they use their blinds to reduce the heat from the sun and only 12.3% stated privacy and security as a reason for blind closure. Eilers et al. [30] surveyed office occupants and confirmed that the majority of the subjects who closed their blinds do so to reduce the glare on their computer screen. Similarly, Warren and Parkins [31] carried out a survey in which occupants stated that IAQ was the main reason for opening windows during the heating season and noise was the main reason for closing windows during the summer season. These surveys, when used prior to the physical measurements, can give pre￾liminary insight into determining the variables that should be measured and the spatial distribution of sensors that will be placed in the office during a study. Furthermore, it can be used to identify the subtleties that are difficult to measure (e.g. rattling blinds caused by wind passing over deployed venetian blinds) before starting the data collection [32]. Early studies on the window opening behavior started by monitoring the outdoor variables [31,33e36] with the following reasoning: (1) once these observations are integrated to BPS as occupant models the indoor variables become outputs of the BPS, therefore, the indoor variables cannot be more reliable than out￾door variables [37,38]; and (2) indoor temperature can be defined as a spatial distribution rather than a single scalar [38] which re￾quires time consuming and expensive instrumentation of the sensors and data loggers. Subsequently, it was suggested that oc￾cupants only have an indirect perception of outdoor physical vari￾ables [38], consequently indoor variables (at least the indoor thermal variables) have been incorporated amongst monitored variables in many of the recent studies [10,38e41]. In particular, using a reasonable balance between the indoor and outdoor vari￾ables to describe the window opening behavior can be suggested as follows: (1) window opening behavior can be described with the indoor variables (e.g. indoor temperature); and (2) window closing behavior can be explained with both indoor and outdoor variables (e.g. indoor and outdoor temperature) [39,42]. This would take into account the transmitted effects of the outdoors once the window is open, while ignoring them once it is closed. Non-temperature physical variables that can affect the window opening/closing behavior have been listed as the indoor air quality [31,43e45], the outdoor noise level [22,31,46,47], RH [39], wind speed and direction [22,34,47], and rain [41,47,48]. These non-temperature physical parameters, despite their importance being qualitatively acknowl￾edged, have not been included in the reviewed literature. Haldi and Robinson [46] suggested building statistical models that incorpo￾rate non-temperature physical variables as predictors as a future research effort. However, as window openings connect the ambient conditions to the indoor environment, the type of the window opening can also play a significant role on the occupant’s prefer￾ences. For example, bottom hung inside opening windows provide weather protection [22]. Thus, in this case, wind and rain may be discarded from the monitored variables. But for the side hung or sliding windows, it may be more appropriate to include the weather variables to the monitored variables. Similarly, to predict a window shade deployment model, indoor variables, such as indoor temperature [29,32,47,49], indoor daylight [29,32,47,49,50], transmitted solar radiation [29,32,50,51]; and outdoor variables, such as outdoor temperature [39,47] and external solar radiation [49,50,52e54], have been monitored. Zhang and Barrett [47] observed that window shade deployment did not follow the outdoor temperature. Arguably, Lindsay and Littlefair [52] and Foster and Oreszczyn [55] claimed that indoor temperature and external solar radiation cannot be a predictor variable for the window shade deployment. Also, Reinhart [56] used only visual/optical variables that may lead to blind lowering. A reasonable explanation for this controversy is that occupants use window shades to mitigate both visual and thermal discomfort (excluding the non-physical factors such as view or privacy), which can be caused by temperature, solar radiation, glare, et cetera. Clearly, it is crucial to monitor independent variables that can represent the window shade deployment. Reviewed literature suggests that these variables are: the depth of penetration of the direct sunlight as a function of the solar altitude [47,51,57] and glare [52,58,59]. Inoue et al. [51] suggested that the depth of penetration of direct sunlight changes with the mean blind occlusion. This observation was later confirmed by Reinhart and Voss [57] with a theoretical solar penetration depth measured from the top of the window. Subsequently, Reinhart and Wienold [60] performed representative daylight design strategies that incorporate occu￾pants’ reaction against direct sunlight and glare. 2.1.2. Non-physical parameters Physical variables (i.e. thermal, visual, acoustic, indoor air environment) influence the chance that an office occupant will experience discomfort. However, it is his/her social, economic, and psychological influences, which are driven by non-physical (i.e. latent) variables that lead to these adaptive actions [61]. These non￾physical (latent) variables are parameters that are not measurable with typical sensors such as view and connection to the outside, privacy or daylight-health perception. It is evident that one of the main design purposes of windows is to provide a clear view and physical connection to the outside [60]. Green building rating systems (e.g. LEED) define a view as “a straight visual connection from an interior point to a point outside through a facade opening located within a certain height range within a facade” [62]. Inoue et al. [51] reported that most occupants preferred to have seats close to the windows, however these seats were known to be the most susceptible locations to glare and solar radiation. Based on the surveys to study the stimulating factors for window opening, it was reported that some of the window opening behaviors, despite allowing some ambient noise, may be explained to maintain a direct connection to outdoors [63,64]. These findings can be interpreted as occupants prefer to tolerate some discomfort in order to have a better quality of view and connection to the outdoors. Window shading devices may obstruct the view to the outside. Haldi and Robinson [65] carried out a study on window control H.B. Gunay et al. / Building and Environment 70 (2013) 31e47 33

H.B.Gunay et aL Building and Environment 70 (2013)31-47 with separate upper and lower blinds and reported that upper openings were suppressed with the larger number of window blinds were slightly more frequently used.The upper blinds were openings during the cooling season.However,it failed to explain found to be fully drawn four times more than the lower blinds. the fact that up to 20%of the windows were left open during the However,the relationship between the view and window shade heating season.This shows that the validity of observations may be use was inconclusive due to variability introduced by the presence limited to a particular season.For example,windows may be of anidolic reflectors.Other researchers 32,47,51,65]have also opened for promoting ventilation during the heating season,while acknowledged the view to the outside as a possible predictor var- during the cooling season it may occur in order to achieve both iable,yet a conclusive finding has not been suggested mainly cooling and ventilation [31.33.34.54.This suggests that proposing because of the interferences from other variables.For example, a general window opening model that is valid for both the heating Rubin,et al.[54]stated that the view to the other office buildings and the cooling season may not be possible. can conflict with the preference to maintain a private indoor space. Similar observations were reported in the studies on window Inkarojrit[32]reported that occupants'desire to maintain privacy shades.Mahdavi et al.[49]carried out a survey on three office as a secondary reason for choosing the blind positions.About 12%of buildings,which revealed that the proportion of the mean shade participants stated that privacy and security concerns represent deployment is up to 30%higher during the cooling season than the one of the reasons why they deploy their window shades.More- heating season.This was explained with the relatively higher solar over,Foster and Oreszczyn [55]unexpectedly observed higher radiation on the facade during cooling season.Even after sub mean blind occlusion rates in the north facade than the west stantial changes took place in the solar radiation and illuminance, facade.This was attributed to the fact that north facade of the occupants usually did not react to change the shade position building was facing another office building,which in turn may be [47.72].Window shades were rarely observed to be operated more explained with the efforts of occupants to preserve their privacy. than once a day [53,54]and even then,Bordass,et al.[15]reported Similarly,Reinhart and Voss57]aimed to correct the bias in their that window shades were typically set to mitigate the worst-case observations due to the privacy concerns and suggested that if condition.Thus,Zhang and Barrett [47]stated that window shade blinds were lowered at ambient horizontal illuminance less than position was based on occupants'long term perception and expe- 1000 lux,it would have occurred due to occupants'desire to rience rather than an instantaneous reaction against a particular maintain privacy.Therefore,a major task for BPS users that incor- stimulus.On the contrary,Haldi and Robinson 65 reported that porate occupant behavior into their studies is to predict the bias seasonal effects depend on other independent variables such as introduced by these non-physical variables and adapt the model indoor temperature or daylight level,thus were found statistically accordingly. insignificant.These results suggest that window shade deploy- Heerwagen and Heerwagen [63]carried out a survey on office ment,unlike window opening,may be used to develop a single occupants in a heating and cooling season and revealed that oc- model that is valid for both the cooling and the heating seasons. cupants widely believe daylight is crucial for their general health Begemann et al.[73]suggested that occupant light switch-on and essential for their work environment.Veitch et al.66 preferences were based on the desire to balance the variation be- confirmed that people believe daylight is superior to artificial tween the window brightness and the interior surfaces.Therefore, lighting for health.Participants reported that the quality of light it is expected that on a sunny summer day occupants tend to switch sources is crucial for their well-being and the florescent lighting on their lights to mitigate the large daylight gradients.This expla- can cause headaches and eyestrain [67.Therefore,the occupants nation can justify the lack of seasonal light energy usage variation preference to sit close to the windows can be explained with their in the UK,by inferring that the occupants can tolerate lower health concerns related to the artificial lighting along with benefit workplane illuminances during winter when the daylight level is of view and connection to the outside. lower and expect higher workplane illuminances during summer Visibility of energy use,which can be influenced with the when the daylight level is higher [741.Therefore,the workplane availability of various feedback sources,affects the behavioral illuminance should not be used alone as a predictor variable to adaptation of occupants[68.These direct and indirect feedbacks model the light switch-on behavior and it may be appropriate to may emerge from simple and more intuitive energy use dashboards incorporate window illuminance as a secondary predictor variable. [69].utility bills [70].competitions or awards [70].Darby [71] estimated that savings of up to ten percent can be achieved 2.1.4.Facade orientation through various feedback strategies,which suggests that occupants Facade orientation affects the magnitude and temporal distri- adapt their behaviors to save energy.In other words,the likelihood bution of the solar gains.For example,for the Northern Hemi- of undertaking a manual control action (e.g.turning off the lights sphere,the north facades receive the least solar gains,while south before departure)can be influenced with the visibility of energy facades receive the most useful solar radiation during the winter. use. Also,the solar penetration varies daily in zones adjacent to the east and west facades,but it varies more seasonally in the south zones 2.1.3.Seasonal effects As a result,naturally ventilated south facing offices tend to have Long-term observational studies revealed noticeable variations higher indoor temperatures than the east,west,and north facing in occupant adaptive behaviors between cooling and heating sea- offices [38].In line with this,the likelihood of opening windows in sons.For example,Fritsch,et al.[36]carried out an observational the south facade was observed to be 30%higher than the north study on the occupant control of windows in four offices in a facade [38].Zhang and Barrett [75]reported that the mean pro- heating season and a cooling season.In the heating season,the portion of windows open was 7.3%in the south facade:6.3%,5.6% window opening behavior was found to not follow the outdoor and 3.6%in the east,west,and north facades,respectively.This temperature,while during the cooling season the outdoor tem- implies that window opening behavior in east and west facades perature was a major factor leading to window opening.Similarly, follows a trend more similar to the south facade than the north Rijal,et al.[10]carried out a long term observational study to facade.Moreover,the peak percentage of window opening predict a window opening model,which was based on the aggre- times shifted following the peak solar radiation rather than the gated observations for cooling and heating seasons.The model was indoor temperature [38.This may be explained with the discom- in agreement with the cooling season observations of Fritsch et al. fort due to the transmitted solar radiation incident on the work- 36;perhaps the lower number of heating season window station and occupant.Similarly,mean window shade occlusion was

with separate upper and lower blinds and reported that upper blinds were slightly more frequently used. The upper blinds were found to be fully drawn four times more than the lower blinds. However, the relationship between the view and window shade use was inconclusive due to variability introduced by the presence of anidolic reflectors. Other researchers [32,47,51,65] have also acknowledged the view to the outside as a possible predictor var￾iable, yet a conclusive finding has not been suggested mainly because of the interferences from other variables. For example, Rubin, et al. [54] stated that the view to the other office buildings can conflict with the preference to maintain a private indoor space. Inkarojrit [32] reported that occupants’ desire to maintain privacy as a secondary reason for choosing the blind positions. About 12% of participants stated that privacy and security concerns represent one of the reasons why they deploy their window shades. More￾over, Foster and Oreszczyn [55] unexpectedly observed higher mean blind occlusion rates in the north facade than the west facade. This was attributed to the fact that north facade of the building was facing another office building, which in turn may be explained with the efforts of occupants to preserve their privacy. Similarly, Reinhart and Voss [57] aimed to correct the bias in their observations due to the privacy concerns and suggested that if blinds were lowered at ambient horizontal illuminance less than 1000 lux, it would have occurred due to occupants’ desire to maintain privacy. Therefore, a major task for BPS users that incor￾porate occupant behavior into their studies is to predict the bias introduced by these non-physical variables and adapt the model accordingly. Heerwagen and Heerwagen [63] carried out a survey on office occupants in a heating and cooling season and revealed that oc￾cupants widely believe daylight is crucial for their general health and essential for their work environment. Veitch et al. [66] confirmed that people believe daylight is superior to artificial lighting for health. Participants reported that the quality of light sources is crucial for their well-being and the florescent lighting can cause headaches and eyestrain [67]. Therefore, the occupants’ preference to sit close to the windows can be explained with their health concerns related to the artificial lighting along with benefit of view and connection to the outside. Visibility of energy use, which can be influenced with the availability of various feedback sources, affects the behavioral adaptation of occupants [68]. These direct and indirect feedbacks may emerge from simple and more intuitive energy use dashboards [69], utility bills [70], competitions or awards [70]. Darby [71] estimated that savings of up to ten percent can be achieved through various feedback strategies, which suggests that occupants adapt their behaviors to save energy. In other words, the likelihood of undertaking a manual control action (e.g. turning off the lights before departure) can be influenced with the visibility of energy use. 2.1.3. Seasonal effects Long-term observational studies revealed noticeable variations in occupant adaptive behaviors between cooling and heating sea￾sons. For example, Fritsch, et al. [36] carried out an observational study on the occupant control of windows in four offices in a heating season and a cooling season. In the heating season, the window opening behavior was found to not follow the outdoor temperature, while during the cooling season the outdoor tem￾perature was a major factor leading to window opening. Similarly, Rijal, et al. [10] carried out a long term observational study to predict a window opening model, which was based on the aggre￾gated observations for cooling and heating seasons. The model was in agreement with the cooling season observations of Fritsch et al. [36]; perhaps the lower number of heating season window openings were suppressed with the larger number of window openings during the cooling season. However, it failed to explain the fact that up to 20% of the windows were left open during the heating season. This shows that the validity of observations may be limited to a particular season. For example, windows may be opened for promoting ventilation during the heating season, while during the cooling season it may occur in order to achieve both cooling and ventilation [31,33,34,54]. This suggests that proposing a general window opening model that is valid for both the heating and the cooling season may not be possible. Similar observations were reported in the studies on window shades. Mahdavi et al. [49] carried out a survey on three office buildings, which revealed that the proportion of the mean shade deployment is up to 30% higher during the cooling season than the heating season. This was explained with the relatively higher solar radiation on the facade during cooling season. Even after sub￾stantial changes took place in the solar radiation and illuminance, occupants usually did not react to change the shade position [47,72]. Window shades were rarely observed to be operated more than once a day [53,54] and even then, Bordass, et al. [15] reported that window shades were typically set to mitigate the worst-case condition. Thus, Zhang and Barrett [47] stated that window shade position was based on occupants’ long term perception and expe￾rience rather than an instantaneous reaction against a particular stimulus. On the contrary, Haldi and Robinson [65] reported that seasonal effects depend on other independent variables such as indoor temperature or daylight level, thus were found statistically insignificant. These results suggest that window shade deploy￾ment, unlike window opening, may be used to develop a single model that is valid for both the cooling and the heating seasons. Begemann et al. [73] suggested that occupant light switch-on preferences were based on the desire to balance the variation be￾tween the window brightness and the interior surfaces. Therefore, it is expected that on a sunny summer day occupants tend to switch on their lights to mitigate the large daylight gradients. This expla￾nation can justify the lack of seasonal light energy usage variation in the UK, by inferring that the occupants can tolerate lower workplane illuminances during winter when the daylight level is lower and expect higher workplane illuminances during summer when the daylight level is higher [74]. Therefore, the workplane illuminance should not be used alone as a predictor variable to model the light switch-on behavior and it may be appropriate to incorporate window illuminance as a secondary predictor variable. 2.1.4. Facade orientation Facade orientation affects the magnitude and temporal distri￾bution of the solar gains. For example, for the Northern Hemi￾sphere, the north facades receive the least solar gains, while south facades receive the most useful solar radiation during the winter. Also, the solar penetration varies daily in zones adjacent to the east and west facades, but it varies more seasonally in the south zones. As a result, naturally ventilated south facing offices tend to have higher indoor temperatures than the east, west, and north facing offices [38]. In line with this, the likelihood of opening windows in the south facade was observed to be 30% higher than the north facade [38]. Zhang and Barrett [75] reported that the mean pro￾portion of windows open was 7.3% in the south facade; 6.3%, 5.6%, and 3.6% in the east, west, and north facades, respectively. This implies that window opening behavior in east and west facades follows a trend more similar to the south facade than the north facade. Moreover, the peak percentage of window opening times shifted following the peak solar radiation rather than the indoor temperature [38]. This may be explained with the discom￾fort due to the transmitted solar radiation incident on the work￾station and occupant. Similarly, mean window shade occlusion was 34 H.B. Gunay et al. / Building and Environment 70 (2013) 31e47

H.B.Gunay et al.Building and Environment 70(2013)31-47 子 reported lowest on north facades and highest on south facades that the number of monitored blind deployments during arrival [29,30,49,54.55].Shades on the north facade were rarely observed was 5.5 times more than that was during presence.Another as fully closed [30.76].Rea [53]and Zhang and Barrett [47]reported explanation suggests that on a sunny day the likelihood of turning that the mean shade occlusion in the east and west facades were the lights on and/or lowering the blinds increase,because the between that of the north facade and the south facade,however occupant wants to balance the brightness of the window areas with they were closer to that of the south facade.Given that the east and those of the interior 741.The increased probability of adaptive west facades are known to have greatest solar penetration depth measures upon arrival could also indicate that the adaptive mea- during the occupied hours and the south perimeter zones often sures taken for the previous occupancy period are no longer have the highest temperatures,the relative importance of the appropriate for current conditions.It is also worth noting that the temperature and the beam solar radiation may be discernible at convenience to undertake a certain adaptive behavior,once an different facade orientations.For example,to avoid frequent blind occupant is already standing upon arrival or prior to departure may use,occupants in the east and west facades can be more likely to increase the ease with which the adaptive action can be made [79]. leave their blinds fully closed.However,it has been suggested that However,no increase in the monitored blind deployment actions the effect of facade orientation itself can be treated as a dependant was observed during the occupant departure.It was concluded that variable of others such as temperature or beam solar radiation occupants do not adjust their blinds for predictive purposes during 24,46,65]:comparing adaptive behaviors in different facade ori- their absence [74].It should be noted that window opening/closing entations can give better insight into the relative significance of actions,unlike the shade deployment actions,were reported with a variables for future models. discernible frequency during departures [28].Likewise,the light switching was observed to take place during arrival and departure 2.1.5.HVAC system and operation 30,80,81].Switch-on actions during arrivals were frequently The type of HVAC system and operation may affect the adaptive explained by the daylight illuminances in the workplane [80.81]. occupant behaviors because occupants do not need to take as many. while the switch-off actions upon departure were explained with if any,adaptive measures if comfort conditions are automatically length of absence [561.Eilers et al.[30]showed that only about half provided.For example,occupants in a naturally ventilated building of the occupants switched off their lights if the departure was fol- may use their windows for different reasons than the occupants in lowed by an absence of two to four hours.Also,this ratio further a mechanically ventilated/cooled building.It was reported that if decreased once there were occupancy sensors or dimmed,indirect the indoor conditions were tightly controlled,occupants were lighting systems [30.57.It is also worth noting that not all these found less likely to undertake adaptive behaviors[24].For example, occupant behaviors aim at adapting their environment or adapting Rijal,et al.10]investigated the effect of open windows on thermal to their environment;instead,they can be habitual actions.For comfort and energy use in two mixed-mode buildings(i.e.build- example,occupants'action to turn on lights upon arrival,regardless ings with mechanical cooling with operable windows [77])and of brightness,can attest their arrival in a habitual manner [26]. seven naturally ventilated buildings.Occupants in the air- Therefore,not only the mere presence of the occupant,but also the conditioned buildings used their windows significantly less than state of presence (e.g.just arrived on a sunny day)should be occupants in the naturally ventilated buildings.Similarly,a study by incorporated in observational studies to be able to properly model Inkarojrit [32]revealed that the mean shade occlusion rate for the the adaptive occupant behavior. offices with air conditioning was 30%in comparison to the 49%for The number of occupants responsible for opening a particular those without.This can be interpreted that the validity of these window can also impact the overall window opening behavior observations should be restrained with the context of the moni- 46,781.Haldi and Robinson [46]observed a slight variation in the tored building or similar buildings window opening behavior in offices with one or two occupants. This was confirmed by similar observations by Herkel et al.[78]in 2.1.6.Occupancy pattern two or three person offices while studying manual blinds control Occupants'distance to the controlled device,time after their and by Moore et al.[74]in one to nine person offices while studying arrival or to their departure,or the number of occupants sharing light switching.However,in this case the aforementioned arrival the same controlled device may affect the likelihood of an adaptive and departure time intervals require further explanation.For behavior.Number of adaptive occupant behaviors per unit time was example,an occupant may walk into an already occupied office and found notably higher just after arrival and just before departure.For open the window.Considering this action as an intermediate example,Pfafferott and Herkel [40]suggested that window open- window opening behavior may mislead the model prediction ing is usually employed during arrival and departure.In between process.Haldi and Robinson [46]suggested a simplifying assump- the arrival and the departure,occupants were found less likely to tion that all occupants act independently and adaptive behavior is open or close a window.The arrival interval was defined as a time controlled by the most active (ie.occupant who uses the adaptive interval that was followed by the arrival and the departure interval control actions more frequently)occupant.Future studies may seek was defined as a time interval that was preceded by the departure for a justification for this assumption.On the contrary,occupants Haldi and Robinson [46]confirmed this observation and claimed tend to be more reluctant to use their blinds if others are present that the first five minutes after arrival and the last five minutes because of social constraints [24].It was reported that such control before the departure define a threshold limit for these arrival and actions in large offices were more frequently performed once departure intervals.Herkel et al.[78]suggested that arrival and most people had left because the action was judged to not impact departure time interval to be 15 min.Similar observations were anyone [82]. reported in the reviewed studies on the window shades and light The indoor conditions are a temporospatial distribution of a switching.This distinct behavior during arrival and departure was physical variable.For example,the illuminance even in a small of- explained with the occupant's preference to minimize the variation fice can vary significantly.In fact,Reinhart and Wienold [60 between the indoors and outdoors [73.This explanation suggests considered repositioning in the office as an independent adaptive that if an occupant walks in a dark office on a sunny day,he/she measure.Subsequently,Jakubiec and Reinhart [83]introduced likely switches the lights on upon arrival.This explanation can be adaptive zone concept in which occupants change position and refuted as the likelihood of blind deployment increases upon arrival view directions,rather than readily accepting the discomfort from in a sunny day as well.For example,Haldi and Robinson [65]stated glare or closing the blinds.It was reported that in a side-lit office

reported lowest on north facades and highest on south facades [29,30,49,54,55]. Shades on the north facade were rarely observed as fully closed [30,76]. Rea [53] and Zhang and Barrett [47] reported that the mean shade occlusion in the east and west facades were between that of the north facade and the south facade, however they were closer to that of the south facade. Given that the east and west facades are known to have greatest solar penetration depth during the occupied hours and the south perimeter zones often have the highest temperatures, the relative importance of the temperature and the beam solar radiation may be discernible at different facade orientations. For example, to avoid frequent blind use, occupants in the east and west facades can be more likely to leave their blinds fully closed. However, it has been suggested that the effect of facade orientation itself can be treated as a dependant variable of others such as temperature or beam solar radiation [24,46,65]; comparing adaptive behaviors in different facade ori￾entations can give better insight into the relative significance of variables for future models. 2.1.5. HVAC system and operation The type of HVAC system and operation may affect the adaptive occupant behaviors because occupants do not need to take as many, if any, adaptive measures if comfort conditions are automatically provided. For example, occupants in a naturally ventilated building may use their windows for different reasons than the occupants in a mechanically ventilated/cooled building. It was reported that if the indoor conditions were tightly controlled, occupants were found less likely to undertake adaptive behaviors [24]. For example, Rijal, et al. [10] investigated the effect of open windows on thermal comfort and energy use in two mixed-mode buildings (i.e. build￾ings with mechanical cooling with operable windows [77]) and seven naturally ventilated buildings. Occupants in the air￾conditioned buildings used their windows significantly less than occupants in the naturally ventilated buildings. Similarly, a study by Inkarojrit [32] revealed that the mean shade occlusion rate for the offices with air conditioning was 30% in comparison to the 49% for those without. This can be interpreted that the validity of these observations should be restrained with the context of the moni￾tored building or similar buildings. 2.1.6. Occupancy pattern Occupants’ distance to the controlled device, time after their arrival or to their departure, or the number of occupants sharing the same controlled device may affect the likelihood of an adaptive behavior. Number of adaptive occupant behaviors per unit time was found notably higher just after arrival and just before departure. For example, Pfafferott and Herkel [40] suggested that window open￾ing is usually employed during arrival and departure. In between the arrival and the departure, occupants were found less likely to open or close a window. The arrival interval was defined as a time interval that was followed by the arrival and the departure interval was defined as a time interval that was preceded by the departure. Haldi and Robinson [46] confirmed this observation and claimed that the first five minutes after arrival and the last five minutes before the departure define a threshold limit for these arrival and departure intervals. Herkel et al. [78] suggested that arrival and departure time interval to be 15 min. Similar observations were reported in the reviewed studies on the window shades and light switching. This distinct behavior during arrival and departure was explained with the occupant’s preference to minimize the variation between the indoors and outdoors [73]. This explanation suggests that if an occupant walks in a dark office on a sunny day, he/she likely switches the lights on upon arrival. This explanation can be refuted as the likelihood of blind deployment increases upon arrival in a sunny day as well. For example, Haldi and Robinson [65] stated that the number of monitored blind deployments during arrival was 5.5 times more than that was during presence. Another explanation suggests that on a sunny day the likelihood of turning the lights on and/or lowering the blinds increase, because the occupant wants to balance the brightness of the window areas with those of the interior [74]. The increased probability of adaptive measures upon arrival could also indicate that the adaptive mea￾sures taken for the previous occupancy period are no longer appropriate for current conditions. It is also worth noting that the convenience to undertake a certain adaptive behavior, once an occupant is already standing upon arrival or prior to departure may increase the ease with which the adaptive action can be made [79]. However, no increase in the monitored blind deployment actions was observed during the occupant departure. It was concluded that occupants do not adjust their blinds for predictive purposes during their absence [74]. It should be noted that window opening/closing actions, unlike the shade deployment actions, were reported with a discernible frequency during departures [28]. Likewise, the light switching was observed to take place during arrival and departure [30,80,81]. Switch-on actions during arrivals were frequently explained by the daylight illuminances in the workplane [80,81], while the switch-off actions upon departure were explained with length of absence [56]. Eilers et al. [30] showed that only about half of the occupants switched off their lights if the departure was fol￾lowed by an absence of two to four hours. Also, this ratio further decreased once there were occupancy sensors or dimmed, indirect lighting systems [30,57]. It is also worth noting that not all these occupant behaviors aim at adapting their environment or adapting to their environment; instead, they can be habitual actions. For example, occupants’ action to turn on lights upon arrival, regardless of brightness, can attest their arrival in a habitual manner [26]. Therefore, not only the mere presence of the occupant, but also the state of presence (e.g. just arrived on a sunny day) should be incorporated in observational studies to be able to properly model the adaptive occupant behavior. The number of occupants responsible for opening a particular window can also impact the overall window opening behavior [46,78]. Haldi and Robinson [46] observed a slight variation in the window opening behavior in offices with one or two occupants. This was confirmed by similar observations by Herkel et al. [78] in two or three person offices while studying manual blinds control and by Moore et al. [74] in one to nine person offices while studying light switching. However, in this case the aforementioned arrival and departure time intervals require further explanation. For example, an occupant may walk into an already occupied office and open the window. Considering this action as an intermediate window opening behavior may mislead the model prediction process. Haldi and Robinson [46] suggested a simplifying assump￾tion that all occupants act independently and adaptive behavior is controlled by the most active (i.e. occupant who uses the adaptive control actions more frequently) occupant. Future studies may seek for a justification for this assumption. On the contrary, occupants tend to be more reluctant to use their blinds if others are present because of social constraints [24]. It was reported that such control actions in large offices were more frequently performed once most people had left because the action was judged to not impact anyone [82]. The indoor conditions are a temporospatial distribution of a physical variable. For example, the illuminance even in a small of- fice can vary significantly. In fact, Reinhart and Wienold [60] considered repositioning in the office as an independent adaptive measure. Subsequently, Jakubiec and Reinhart [83] introduced adaptive zone concept in which occupants change position and view directions, rather than readily accepting the discomfort from glare or closing the blinds. It was reported that in a side-lit office H.B. Gunay et al. / Building and Environment 70 (2013) 31e47 35

36 H.B.Gunay et aL Building and Environment 70(2013)31-47 with venetian blinds,small movements account for a substantial many of the aforementioned limitations.It should also be noted reduction in number of hours occupants experience discomfort that if the discretization of the window state and the sampling glare (from 735 to 18 occupied hours)[83].This underlines the intervals are not fine enough,the data may be misleading,since the importance of interior design of office spaces,which allows occu- small window openings and short opening/closing intervals may be pants to reposition or change view direction,to reduce occupant lost.Warren and Parkins [31]and Brager et al.84]photographed discomfort and in turn,occupants'need for environmental behav- the facade of the building twice a day to quantify the window state ioral adaptations.The location (e.g.distance to the window)where which was discretized as open,slightly-open,or closed.Inkarojrit the occupant is seated in an office and the freedom to reposition or and Paliaga [38]automated the time lapse-photography technique change view direction should be noted as this accounts for some so that three major facades of a naturally ventilated office building variance in the reviewed studies [51,571.The physical variables were photographed at fixed times(four times per day for nine days) should be collected as close as possible to the workstations or during a shoulder season,while windows were recorded as a binary occupants. state(i.e.open/closed).Reinhart [56]suggested that observational round-offs due to crude state-discretizations (e.g.fully-closed or 2.1.7.Observation techniques,time and state discretization open-blinds)can cause critical information loss about smaller The technique employed to monitor the change in the state(e.g. behavioral adaptations.For example,some occupants may opt to window opening),the sampling interval (e.g.twice per day).and close their blinds up to a level to block direct sunlight by leaving the state discretization (e.g.open/closed or open/half-open/closed) some opening to maintain view to outdoors;but this may not be may represent limitations for the model prediction.Warren and noted as a behavioral adaptation if the state was discretized as on Parkins [31]carried out a survey on five buildings with a total of 196 off.Thus,a future research effort to understand the sensitivity of offices by employing a time-lapse photography technique.This such algorithms on the state and time discretization can be technique,despite being non-invasive and relatively inexpensive, beneficial. does not give any insight about the indoor environment [24.This is The reviewed literature on the window shades using the time- perhaps why Warren and Parkins [31]used weather variables lapse photography technique involves similar discretization in rather than the indoor variables to suggest a correlation.Brager time and state.The frequency of photographs ranges from once a et al.84]tackled this limitation with continuous desktop indoor week [54]to once each hour [85]during the official work hours. monitoring and questionnaire surveys along with the time-lapse Despite the predominant seasonal effects in the window shade photography in a naturally ventilated office building with 230 deployment,the diurnal variations were reported as noticeable people.Likewise,the positions of the window shading system were particularly for south,east,and west facades.Thus,it has been monitored using the time-lapse photography technique [32.52- suggested that the photography technique should be employed at 551.The limitation of using this technique for studying windows least twice a day [24].The discretization of the window shades was and window shades is that the slat angles of Venetian blinds and constrained with the image resolution and the post-processing the hinged-window openings may not be distinguishable[24].The technique (i.e.digital image processing or manual analysis).The insufficient image resolution [52.55.visual obstructions [29,54]. window shade states were discretized between 2 [49]and 10[85 and uncertainty in weather(e.g.fog)[54]represent other limita- discrete shade positions.Unlike roller blinds,the slat angle in tions for the technique.However,with the development of image Venetian blinds should be incorporated in the state discretization, recognition algorithms this technique can be more widely used for since the slat angle (i.e.tilted upwards or downwards)affects the retrieving information about the windows and window shades. amount of transmitted daylight by a factor of 10 [521.However,in O'Brien et al.85]demonstrated that this when coupled with image the reviewed literature,the Venetian blinds were discretized as recognition algorithms,as shown in Fig.2.can be used to improve open or closed [47]and open/tilted downwards/closed [55. ■里 ■ 12345678910111213141516 (a) (b) (c) Fig.2.(a)Original photograph.(b)Preprocessing image after cropping and transformation.(c)Post-processing image with window shades deployments discretized with 10 in- termediate states (taken from O'Brien,Kapsis et al.241)

with venetian blinds, small movements account for a substantial reduction in number of hours occupants experience discomfort glare (from 735 to 18 occupied hours) [83]. This underlines the importance of interior design of office spaces, which allows occu￾pants to reposition or change view direction, to reduce occupant discomfort and in turn, occupants’ need for environmental behav￾ioral adaptations. The location (e.g. distance to the window) where the occupant is seated in an office and the freedom to reposition or change view direction should be noted as this accounts for some variance in the reviewed studies [51,57]. The physical variables should be collected as close as possible to the workstations or occupants. 2.1.7. Observation techniques, time and state discretization The technique employed to monitor the change in the state (e.g. window opening), the sampling interval (e.g. twice per day), and the state discretization (e.g. open/closed or open/half-open/closed) may represent limitations for the model prediction. Warren and Parkins [31] carried out a survey on five buildings with a total of 196 offices by employing a time-lapse photography technique. This technique, despite being non-invasive and relatively inexpensive, does not give any insight about the indoor environment [24]. This is perhaps why Warren and Parkins [31] used weather variables rather than the indoor variables to suggest a correlation. Brager et al. [84] tackled this limitation with continuous desktop indoor monitoring and questionnaire surveys along with the time-lapse photography in a naturally ventilated office building with 230 people. Likewise, the positions of the window shading system were monitored using the time-lapse photography technique [32,52e 55]. The limitation of using this technique for studying windows and window shades is that the slat angles of Venetian blinds and the hinged-window openings may not be distinguishable [24]. The insufficient image resolution [52,55], visual obstructions [29,54], and uncertainty in weather (e.g. fog) [54] represent other limita￾tions for the technique. However, with the development of image recognition algorithms this technique can be more widely used for retrieving information about the windows and window shades. O’Brien et al. [85] demonstrated that this when coupled with image recognition algorithms, as shown in Fig. 2, can be used to improve many of the aforementioned limitations. It should also be noted that if the discretization of the window state and the sampling intervals are not fine enough, the data may be misleading, since the small window openings and short opening/closing intervals may be lost. Warren and Parkins [31] and Brager et al. [84] photographed the facade of the building twice a day to quantify the window state which was discretized as open, slightly-open, or closed. Inkarojrit and Paliaga [38] automated the time lapse-photography technique so that three major facades of a naturally ventilated office building were photographed at fixed times (four times per day for nine days) during a shoulder season, while windows were recorded as a binary state (i.e. open/closed). Reinhart [56] suggested that observational round-offs due to crude state-discretizations (e.g. fully-closed or open-blinds) can cause critical information loss about smaller behavioral adaptations. For example, some occupants may opt to close their blinds up to a level to block direct sunlight by leaving some opening to maintain view to outdoors; but this may not be noted as a behavioral adaptation if the state was discretized as on/ off. Thus, a future research effort to understand the sensitivity of such algorithms on the state and time discretization can be beneficial. The reviewed literature on the window shades using the time￾lapse photography technique involves similar discretization in time and state. The frequency of photographs ranges from once a week [54] to once each hour [85] during the official work hours. Despite the predominant seasonal effects in the window shade deployment, the diurnal variations were reported as noticeable particularly for south, east, and west facades. Thus, it has been suggested that the photography technique should be employed at least twice a day [24]. The discretization of the window shades was constrained with the image resolution and the post-processing technique (i.e. digital image processing or manual analysis). The window shade states were discretized between 2 [49] and 10 [85] discrete shade positions. Unlike roller blinds, the slat angle in Venetian blinds should be incorporated in the state discretization, since the slat angle (i.e. tilted upwards or downwards) affects the amount of transmitted daylight by a factor of 10 [52]. However, in the reviewed literature, the Venetian blinds were discretized as open or closed [47] and open/tilted downwards/closed [55]. Fig. 2. (a) Original photograph, (b) Preprocessing image after cropping and transformation, (c) Post-processing image with window shades deployments discretized with 10 in￾termediate states (taken from O’Brien, Kapsis et al. [24]). 36 H.B. Gunay et al. / Building and Environment 70 (2013) 31e47

H.B.Gunay et al.Building and Environment 70(2013)31-47 37 Instead of time-lapse photography.sensors (e.g.contact and setpoint indoor illuminance is reached;or ventilation systems that proximity sensors in windows or roller blinds)have also been provide fresh air for a set number of occupants and for a set occu- employed to acquire window opening behavior at a finer resolution pancy schedule.These set values are static assumptions published in (i.e.one to fifteen measurements per minute).For example,Yun various comfort standards [13.14.94]to address the conservative and Steemers (41]investigated the window opening behavior using design needs;however they neglect individual variations amongst the sensors in a naturally ventilated office building in parallel with occupants'preferences,behaviors,and habits.In fact,in many cases a questionnaire survey to confirm the occupancy and to estimate findings of the research on occupant control of blinds and lighting the ASHRAE [14]predicted mean vote (PMV)during a cooling [57,65 do not match with the design practices to automate the season.Use of sensors is less prone to human error and more blinds and lighting [20.64.68,90].For example,to avoid occupant efficient in data acquisition and post-processing [24].Also,longer complaints due to glare,blinds automation systems typically use studies can be performed with this technique such as Haldi and conservative setpoint values less than 2 klux [20].However,it was Robinson [39].who used a data record for six years for a naturally reported that occupants rarely close their blinds at workplane illu- ventilated office building.However,the size of sample group in minances less than 2 klux [57.651.This conservativeness,to avoid these studies was usually smaller than 50 [28,39,41,46,50.57,65.86] occupant complaints due to glare,resulted in occupant overrides to which was notably less than the large-scale surveys carried out on preserve their view and connection to the outside.This suggests that up to 1200 windows[85]using the time-lapse photography tech- occupants'preferences can be wildly different and questions the nique [32,36,38,51,53,55,82].In some of these studies [32.50]. existence of a standard indoor climate which can make every occu- hand-held sensors,instead of permanent sensors,were used in pant happy.This is also implicitly acknowledged in the current parallel with questionnaires and visual investigations.This ASHRAE comfort standard 14.As reported in Olesen and Brager approach gives immeasurable insights(e.g.rattling blinds)24]into 95].ASHRAE [14]defines conditions acceptable to a majority of a a smaller scale research group for a shorter and an irregular dura- group of occupants,whereby the majority is defined as 80%Instead tion and may become useful if coupled with other techniques such of trying to sustain a standard indoor environment,controllers in the as time-lapse photography. building automation systems can be used to run a self-adaptive al- gorithm to learn occupants'customized indoor climate preferences 2.1.8.Automated/manual controls from their behaviors.Guillemin96 showed that occupant overrides To reduce occupants'energy impact,building systems with can be reduced from 25%to 5%of the automated control actions by which occupants widely interact to adapt their indoor environments training a self-adaptive occupant-learning algorithm in the individ- (e.g.windows,window shading devices,lights,thermostats)have ual controller level. been automated in many applications.However,evidence from these applications suggest that occupants frequently override these 2.1.9.Reversal of an adaptive behavior automation systems indicating their dissatisfaction;and these It is known that occupants adapt their environment to preserve frequent overrides deemed many automation applications poorly their comfort.However the reversal of the same adaptive behaviors functioning.For example,Reinhart and Voss [57]reported that in are not fully understood [56.Only a small group of the reviewed 1263 out of 1432 attempts(88%)to close the blinds automatically. studies reported that the reversal of adaptive occupant behaviors, the control algorithm was overridden by the occupants.Reinhart such as closing a window.opening a shade,or switching-off the explained this as "You'd have a prized window seat,then Zoop!The lights,take place after the source of discomfort dissipates [97.This shades go down.There was a realization that occupants don't accept arises because the incentives of an adaptive behavior are different that".In line with this.Leaman and Bordass (87]stated that auto- than the incentives to reverse it,as shown in Fig.3.For example mation systems that exclude occupants from the control-loop(e.g. occupants open a window to improve the IAQ and thermal comfort closing blinds before glare conditions exist for occupants)can but acoustic discomfort (e.g.noise)or thermal discomfort motivate infuriate occupants.Other studies [37,88-90]confirmed these ob- them to close it.Similarly,occupants lower a window shading de- servations;many of such cases with automated blind and lighting vice to satisfy their visual and thermal comfort and privacy,but the controls were deactivated due to complaints or needed to be preference to get more daylight and to have better view to the improved/customized to meet individual preferences with a post- outside motivates them to raise it.This effect is expected to be even occupancy commissioning process.Even more surprisingly,Carter. more significant if electrical and mechanical systems provide et al.[91]reported that manually controllable lighting fixtures alternative means for comfort.Therefore,it is expected that the which do not even meet the lighting standards were perceived more predictive models for adaptive behaviors and their reversals are satisfactory than the daylight linked automated lighting controls. different.For example,Reinhart and Voss [57]reported that blinds Two different explanations have been suggested in the literature were manually closed at external facade illuminance of 50 klux and to explain occupants'discontent with the automation applications: opened at 25 klux.On the other extreme,it was observed that some (1)Desire for ability to control:Leaman and Bordass92]showed that occupants did not raise their blinds for multiple months once they there is a strong relationship between occupants'perception of were closed [54,551.Similarly,it was observed that occupants do control over their environment and productivity.In fact,a common not notice,or perhaps simply ignore,the availability of daylight and industry practice was reported as placing dummy (ie.placebo) failed to switch-off their lights at the indoor daylight level that is controllers (e.g.thermostats)so that occupants overestimate their equivalent to the artificial lighting [57.Sutter et al.[50]confirmed control on their offices,i.e.illusion of control [931.Galasiu and Veitch these observations that occupants raise their blinds at illuminance 26]interpreted this as the occupants'preference to have the capa- levels lower than at which they are lowered and defined this as a bility to choose their environment rather than being obligated to "hysteresis phenomenon".This concept was later adopted by others accept the environment chosen for them.(2)Desire for a customized [10,39,42,46,65.These studies observed the window,window indoor climate:Controllers in building automation systems are shading,or light switch actions,not as mere adjustments.They decision-makers such that they train logics to control a device.For rather classified these adjustments as opening/closing,lowering/ example,if the temperature is above the setpoint,the controller tells raising,and switching on/off.Therefore,observations that involve the air-conditioner to turn-on.Similar setpoint-based deterministic adaptive occupant behaviors should be distinguished as actions to building automation strategies are commonplace in practice.For mitigate discomfort and as the reversal of these actions after the example,automated blinds that close or lights that turn on,when the source of discomfort fades

Instead of time-lapse photography, sensors (e.g. contact and proximity sensors in windows or roller blinds) have also been employed to acquire window opening behavior at a finer resolution (i.e. one to fifteen measurements per minute). For example, Yun and Steemers [41] investigated the window opening behavior using the sensors in a naturally ventilated office building in parallel with a questionnaire survey to confirm the occupancy and to estimate the ASHRAE [14] predicted mean vote (PMV) during a cooling season. Use of sensors is less prone to human error and more efficient in data acquisition and post-processing [24]. Also, longer studies can be performed with this technique such as Haldi and Robinson [39], who used a data record for six years for a naturally ventilated office building. However, the size of sample group in these studies was usually smaller than 50 [28,39,41,46,50,57,65,86], which was notably less than the large-scale surveys carried out on up to 1200 windows [85] using the time-lapse photography tech￾nique [32,36,38,51,53,55,82]. In some of these studies [32,50], hand-held sensors, instead of permanent sensors, were used in parallel with questionnaires and visual investigations. This approach gives immeasurable insights (e.g. rattling blinds) [24] into a smaller scale research group for a shorter and an irregular dura￾tion and may become useful if coupled with other techniques such as time-lapse photography. 2.1.8. Automated/manual controls To reduce occupants’ energy impact, building systems with which occupants widely interact to adapt their indoor environments (e.g. windows, window shading devices, lights, thermostats) have been automated in many applications. However, evidence from these applications suggest that occupants frequently override these automation systems indicating their dissatisfaction; and these frequent overrides deemed many automation applications poorly functioning. For example, Reinhart and Voss [57] reported that in 1263 out of 1432 attempts (88%) to close the blinds automatically, the control algorithm was overridden by the occupants. Reinhart explained this as “You’d have a prized window seat, then Zoop! The shades go down. There was a realization that occupants don’t accept that”. In line with this, Leaman and Bordass [87] stated that auto￾mation systems that exclude occupants from the control-loop (e.g. closing blinds before glare conditions exist for occupants) can infuriate occupants. Other studies [37,88e90] confirmed these ob￾servations; many of such cases with automated blind and lighting controls were deactivated due to complaints or needed to be improved/customized to meet individual preferences with a post￾occupancy commissioning process. Even more surprisingly, Carter, et al. [91] reported that manually controllable lighting fixtures which do not even meet the lighting standards were perceived more satisfactory than the daylight linked automated lighting controls. Two different explanations have been suggested in the literature to explain occupants’ discontent with the automation applications: (1) Desire for ability to control: Leaman and Bordass [92] showed that there is a strong relationship between occupants’ perception of control over their environment and productivity. In fact, a common industry practice was reported as placing dummy (i.e. placebo) controllers (e.g. thermostats) so that occupants overestimate their control on their offices, i.e. illusion of control [93]. Galasiu and Veitch [26] interpreted this as the occupants’ preference to have the capa￾bility to choose their environment rather than being obligated to accept the environment chosen for them. (2) Desire for a customized indoor climate: Controllers in building automation systems are decision-makers such that they train logics to control a device. For example, if the temperature is above the setpoint, the controller tells the air-conditioner to turn-on. Similar setpoint-based deterministic building automation strategies are commonplace in practice. For example, automated blinds that close or lights that turn on, when the setpoint indoor illuminance is reached; or ventilation systems that provide fresh air for a set number of occupants and for a set occu￾pancy schedule. These set values are static assumptions published in various comfort standards [13,14,94] to address the conservative design needs; however they neglect individual variations amongst occupants’ preferences, behaviors, and habits. In fact, in many cases findings of the research on occupant control of blinds and lighting [57,65] do not match with the design practices to automate the blinds and lighting [20,64,68,90]. For example, to avoid occupant complaints due to glare, blinds automation systems typically use conservative setpoint values less than 2 klux [20]. However, it was reported that occupants rarely close their blinds at workplane illu￾minances less than 2 klux [57,65]. This conservativeness, to avoid occupant complaints due to glare, resulted in occupant overrides to preserve their view and connection to the outside. This suggests that occupants’ preferences can be wildly different and questions the existence of a standard indoor climate which can make every occu￾pant happy. This is also implicitly acknowledged in the current ASHRAE comfort standard [14]. As reported in Olesen and Brager [95], ASHRAE [14] defines conditions acceptable to a majority of a group of occupants, whereby the majority is defined as 80%. Instead of trying to sustain a standard indoor environment, controllers in the building automation systems can be used to run a self-adaptive al￾gorithm to learn occupants’ customized indoor climate preferences from their behaviors. Guillemin [96] showed that occupant overrides can be reduced from 25% to 5% of the automated control actions by training a self-adaptive occupant-learning algorithm in the individ￾ual controller level. 2.1.9. Reversal of an adaptive behavior It is known that occupants adapt their environment to preserve their comfort. However the reversal of the same adaptive behaviors are not fully understood [56]. Only a small group of the reviewed studies reported that the reversal of adaptive occupant behaviors, such as closing a window, opening a shade, or switching-off the lights, take place after the source of discomfort dissipates [97]. This arises because the incentives of an adaptive behavior are different than the incentives to reverse it, as shown in Fig. 3. For example, occupants open a window to improve the IAQ and thermal comfort, but acoustic discomfort (e.g. noise) or thermal discomfort motivate them to close it. Similarly, occupants lower a window shading de￾vice to satisfy their visual and thermal comfort and privacy, but the preference to get more daylight and to have better view to the outside motivates them to raise it. This effect is expected to be even more significant if electrical and mechanical systems provide alternative means for comfort. Therefore, it is expected that the predictive models for adaptive behaviors and their reversals are different. For example, Reinhart and Voss [57] reported that blinds were manually closed at external facade illuminance of 50 klux and opened at 25 klux. On the other extreme, it was observed that some occupants did not raise their blinds for multiple months once they were closed [54,55]. Similarly, it was observed that occupants do not notice, or perhaps simply ignore, the availability of daylight and failed to switch-off their lights at the indoor daylight level that is equivalent to the artificial lighting [57]. Sutter et al. [50] confirmed these observations that occupants raise their blinds at illuminance levels lower than at which they are lowered and defined this as a “hysteresis phenomenon”. This concept was later adopted by others [10,39,42,46,65]. These studies observed the window, window shading, or light switch actions, not as mere adjustments. They rather classified these adjustments as opening/closing, lowering/ raising, and switching on/off. Therefore, observations that involve adaptive occupant behaviors should be distinguished as actions to mitigate discomfort and as the reversal of these actions after the source of discomfort fades. H.B. Gunay et al. / Building and Environment 70 (2013) 31e47 37

H.B.Gunay et aL Building and Ervironment 70 (2013)31-47 Behavior Windows Blinds Lights {2838-471 3 49,52,611 29,32,48.50.51.52 169-721 69-724 .581 Wind/ View Visibility of Thermal Acoustic Rain IAQ Visual Privacy Health Energy-Use 2334.42,48.49 128,38-471 2334.47 58.60.611 32.48.52,6166 164,67,68 681 169-721 69-724 Reversal Windows Blinds Lights Fig.3.Behavior to adaptive system and reversal of adaptive system. 2.1.10.Predictability of the occupants:active and passive ASHRAE Standard 55-2010 [14]recommends to use clothing levels The adaptive occupant behavior has been deemed as individual of 0.5 clo and 1.0 clo,in summer and winter,respectively.Most but not arbitrary [57.suggesting that individual occupants,despite recently.Schiavon and Lee [107]reported that median clothing the fact that their accumulated response is stochastic,undertake insulation for summer as 0.59 clo and for winter as 0.69 clo, control actions consciously and consistently [51-54,80-82,97-991 respectively.This suggested that seasonal clothing level variations For example,Lindsay et al.52]reported that the frequency of are significantly smaller than it is assumed by ASHRAE [141.Haldi manual blind adjustments varies from never to daily even in the and Robinson[108]reported that seasonal clothing level variations same facade,however,adjustments remain predictable for the in- have an amplitude notably larger than those observed within a day. dividual office level.Therefore,it was suggested that there are as shown in Fig.5.They also studied the behavioral adaptation by "passive"and "active"occupants according to their interaction with drinking hot or cold beverages and by changing the activity level. the windows [10,31,78,86].lights and blinds [51-54,80-82,98,99]. From their results it can be inferred that occupants consume more Boyce [82]reported that manual light control in shared offices are hot drinks during the heating season and more cold drinks during consistently performed by the same group of people.Passive oc- the cooling season.No seasonal dependence was reported in the cupants were observed to be consistently less reactive to thermal reviewed literature about the adjustment of the activity level. variations,while active occupants undertook control actions more frequently.Rijal et al.[10]asked subjects how frequently they 2.2.2.Observation techniques,time and state discretization adjust their windows.Fig.4a shows that subjects who responded as Studying personal adaptive behaviors,unlike the environmental "never"have a distinct way of interacting with their windows adaptive behaviors is challenging due to its invasive nature[1091.It compared to those who responded as "sometimes/seldom"and is challenging to place sensors or to carry out a time-lapse "often".In line with these findings,Reinhart and Voss [57]reported photography survey in order to study personal adaptive behav- models for manual lighting use for single or doubled occupied of- iors.Two techniques were observed in the reviewed literature:(1) fice spaces (see Fig.4b)and Haldi and Robinson [65]reported observational field surveys [100,102.110]and (2)self-reported models for manual blinds use for single or double occupied office questionnaire surveys [39,108,111.112].In the observational field spaces(see Fig.4c).These models indicate that occupants under- surveys,the researchers simply observe a group of subjects and take adaptive behaviors at wildly different indoor conditions.This their personal adaptive behaviors in their daily routines for a suggests that some occupants prefer darker indoor environments, certain period of time.In the self-reported questionnaire surveys, while some prefer brighter indoor environments.Some had more subjects are recruited participants such that they are aware of the consistent light level preferences,while some had less predictable scope of research and are being observed.The observational studies preferences. give the opportunity to monitor the state in a very small scale.For example,Wyon and Holmberg [110]carried out an observational 2.2.Behaviors that adapt the personal characteristics survey on primary school students and revealed small scale diurnal clothing adjustments such as an opening collars or rolling up Based on a summary of 30 field studies,Humphreys [100]in sleeves.In a self-reported survey,the questionnaire size and fre- 1975 observed a relatively small variation in the reported thermal quency should be kept short for practical reasons.For example, sensation over a range of indoor temperatures from 17 to 30C.This Haldi and Robinson [108]discretized the clothing level in the observation,despite not being completely applicable in today's questionnaire response in eight different clothing assemblies(e.g. fully-conditioned office buildings,is a clear indication of occupants' shirt with short sleeves,sandals,and shorts)that would result in adaptation of their personal characteristics to accommodate to between 0.3 and 0.95 clo.It was also acknowledged that this dis- their environment.This is typically achieved by adjusting their cretization was a deliberate compromise and this might have also clothing and activity levels(e.g.siestas)and by drinking colder/ caused a loss of information about the small scale diurnal clothing hotter beverages. level changes.For example a small,yet significant variation of the clothing level of the order of 0.1 clo might have occurred more 2.2.1.Seasonal effects frequently than relatively large level variations.A similar approach Most of the early research efforts [101-106]have shown that was also used in discretization of the activity level with six possi- clothing level of the occupants vary notably from season to season bilities such as seated relaxed or sedentary activity.Drinks were by observing the same group of subjects.In line with these studies, discretized as a binary state,which was a reasonable assumption

2.1.10. Predictability of the occupants: active and passive The adaptive occupant behavior has been deemed as individual but not arbitrary [57], suggesting that individual occupants, despite the fact that their accumulated response is stochastic, undertake control actions consciously and consistently [51e54,80e82,97e99]. For example, Lindsay et al. [52] reported that the frequency of manual blind adjustments varies from never to daily even in the same facade, however, adjustments remain predictable for the in￾dividual office level. Therefore, it was suggested that there are “passive” and “active” occupants according to their interaction with the windows [10,31,78,86], lights and blinds [51e54,80e82,98,99]. Boyce [82] reported that manual light control in shared offices are consistently performed by the same group of people. Passive oc￾cupants were observed to be consistently less reactive to thermal variations, while active occupants undertook control actions more frequently. Rijal et al. [10] asked subjects how frequently they adjust their windows. Fig. 4a shows that subjects who responded as “never” have a distinct way of interacting with their windows compared to those who responded as “sometimes/seldom” and “often”. In line with these findings, Reinhart and Voss [57] reported models for manual lighting use for single or doubled occupied of- fice spaces (see Fig. 4b) and Haldi and Robinson [65] reported models for manual blinds use for single or double occupied office spaces (see Fig. 4c). These models indicate that occupants under￾take adaptive behaviors at wildly different indoor conditions. This suggests that some occupants prefer darker indoor environments, while some prefer brighter indoor environments. Some had more consistent light level preferences, while some had less predictable preferences. 2.2. Behaviors that adapt the personal characteristics Based on a summary of 30 field studies, Humphreys [100] in 1975 observed a relatively small variation in the reported thermal sensation over a range of indoor temperatures from 17 to 30 C. This observation, despite not being completely applicable in today’s fully-conditioned office buildings, is a clear indication of occupants’ adaptation of their personal characteristics to accommodate to their environment. This is typically achieved by adjusting their clothing and activity levels (e.g. siestas) and by drinking colder/ hotter beverages. 2.2.1. Seasonal effects Most of the early research efforts [101e106] have shown that clothing level of the occupants vary notably from season to season by observing the same group of subjects. In line with these studies, ASHRAE Standard 55-2010 [14] recommends to use clothing levels of 0.5 clo and 1.0 clo, in summer and winter, respectively. Most recently, Schiavon and Lee [107] reported that median clothing insulation for summer as 0.59 clo and for winter as 0.69 clo, respectively. This suggested that seasonal clothing level variations are significantly smaller than it is assumed by ASHRAE [14]. Haldi and Robinson [108] reported that seasonal clothing level variations have an amplitude notably larger than those observed within a day, as shown in Fig. 5. They also studied the behavioral adaptation by drinking hot or cold beverages and by changing the activity level. From their results it can be inferred that occupants consume more hot drinks during the heating season and more cold drinks during the cooling season. No seasonal dependence was reported in the reviewed literature about the adjustment of the activity level. 2.2.2. Observation techniques, time and state discretization Studying personal adaptive behaviors, unlike the environmental adaptive behaviors is challenging due to its invasive nature [109]. It is challenging to place sensors or to carry out a time-lapse photography survey in order to study personal adaptive behav￾iors. Two techniques were observed in the reviewed literature: (1) observational field surveys [100,102,110] and (2) self-reported questionnaire surveys [39,108,111,112]. In the observational field surveys, the researchers simply observe a group of subjects and their personal adaptive behaviors in their daily routines for a certain period of time. In the self-reported questionnaire surveys, subjects are recruited participants such that they are aware of the scope of research and are being observed. The observational studies give the opportunity to monitor the state in a very small scale. For example, Wyon and Holmberg [110] carried out an observational survey on primary school students and revealed small scale diurnal clothing adjustments such as an opening collars or rolling up sleeves. In a self-reported survey, the questionnaire size and fre￾quency should be kept short for practical reasons. For example, Haldi and Robinson [108] discretized the clothing level in the questionnaire response in eight different clothing assemblies (e.g. shirt with short sleeves, sandals, and shorts) that would result in between 0.3 and 0.95 clo. It was also acknowledged that this dis￾cretization was a deliberate compromise and this might have also caused a loss of information about the small scale diurnal clothing level changes. For example a small, yet significant variation of the clothing level of the order of 0.1 clo might have occurred more frequently than relatively large level variations. A similar approach was also used in discretization of the activity level with six possi￾bilities such as seated relaxed or sedentary activity. Drinks were discretized as a binary state, which was a reasonable assumption Fig. 3. Behavior to adaptive system and reversal of adaptive system. 38 H.B. Gunay et al. / Building and Environment 70 (2013) 31e47

H.B.Gunay et al.Building and Environment 70(2013)31-47 39 (a) (b) 0.8 G Onen △ -△Sometimes/Seldom 5 0.6 Q 号 0.6 0.4 △ A 0.4 0.2 只 22 24 26 子 0 200 400 600 800 Temperature (C) Workplane illuminance (lux) (c) 0,8 0.6 0.4 0.2 2000 4000 6000 8000 Workplane illuminance (lux) Fig.4.(a)Window opening behavior of participants that defined themselves as passive,medium,and active occupants in a survey 40].(b)manual light use behavior in ten single or double occupied offices [56].(c)manual blinds use behavior in ten single or double occupied offices [64]. given that the occupants would finish that cold drink at once in a level adjustments based on the outdoor temperature was a pre- short period of time.However,drinking as a behavioral adaptation dictive measure.For example,on a warm day that is followed by a to the environment can be further investigated by monitoring the cold day,occupants would likely wear heavier clothes based on beverage size and temperature. their experience.This type of behavior was distinguished with the Time discretization of the self-reported questionnaire ranged diurnal clothing level variations such as taking off a jacket or a from one per hour [112]to two per day [108.111.113].This is also sweater.Therefore,people's long-term clothing selections are restrained by the practical reasons associated with participants.For based on the short term history of the outdoor conditions and the example,sometimes volunteers were given the option to select the short-term clothing level adjustments are based on the current frequency (i.e.either 2 or 3 times a day)of the pop-up question- indoor temperature [108.In a more recent study,Schiavon and Lee naire on their computer [108].This questionnaire in Haldi and [107]elaborated this by suggesting two models.Of these,one used Robinson[108]asked occupants whether or not they changed their the outdoor temperature (at 6 am)as a predictor variable and the clothing ensemble,activity type and drank a cold/hot drink in the other included both the outdoor temperature and the operative previous hour.Another problem associated with this type of temperature as predictor variables.However,these models could questionnaire surveys is the bias of the self-reported data.Veitch only predict a small portion of the clothing level variances and it et al.[66]reported that people perceived themselves more active was concluded that indoor and outdoor climate can only account than they actually were.Moreover,so-called "Hawthorne effect" for a small portion of human clothing behavior.This may be an suggests that occupants behave differently when they know that indication of non-adaptive,perhaps habitual and task based nature they are monitored [114].Therefore,attention must be paid while of human clothing behaviors which underlines the inherent diffi- interpreting the self-reported activities and these should be sup- culty in predicting them.Similarly,beverage temperature selec- ported with other observations carried out by the researchers. tions (e.g.cold/hot drink)were observed to follow the long-term outdoor temperature variations [108.The adjustment of the ac- 2.2.3.Physical variables tivity rate was observed with respect to the indoor and the outdoor Indoor[100,102,108,1121 and outdoor[100,102.110.1151tem- temperature [108,113].However,given that in offices,activity was peratures were monitored as variables to be utilized in the restrained with the task (e.g.sedentary).no strong relationship modeling stage.Haldi and Robinson [108]reported that clothing could be proposed based on these observations [108.113]

given that the occupants would finish that cold drink at once in a short period of time. However, drinking as a behavioral adaptation to the environment can be further investigated by monitoring the beverage size and temperature. Time discretization of the self-reported questionnaire ranged from one per hour [112] to two per day [108,111,113]. This is also restrained by the practical reasons associated with participants. For example, sometimes volunteers were given the option to select the frequency (i.e. either 2 or 3 times a day) of the pop-up question￾naire on their computer [108]. This questionnaire in Haldi and Robinson [108] asked occupants whether or not they changed their clothing ensemble, activity type and drank a cold/hot drink in the previous hour. Another problem associated with this type of questionnaire surveys is the bias of the self-reported data. Veitch et al. [66] reported that people perceived themselves more active than they actually were. Moreover, so-called “Hawthorne effect” suggests that occupants behave differently when they know that they are monitored [114]. Therefore, attention must be paid while interpreting the self-reported activities and these should be sup￾ported with other observations carried out by the researchers. 2.2.3. Physical variables Indoor [100,102,108,112] and outdoor [100,102,110,115] tem￾peratures were monitored as variables to be utilized in the modeling stage. Haldi and Robinson [108] reported that clothing level adjustments based on the outdoor temperature was a pre￾dictive measure. For example, on a warm day that is followed by a cold day, occupants would likely wear heavier clothes based on their experience. This type of behavior was distinguished with the diurnal clothing level variations such as taking off a jacket or a sweater. Therefore, people’s long-term clothing selections are based on the short term history of the outdoor conditions and the short-term clothing level adjustments are based on the current indoor temperature [108]. In a more recent study, Schiavon and Lee [107] elaborated this by suggesting two models. Of these, one used the outdoor temperature (at 6 am) as a predictor variable and the other included both the outdoor temperature and the operative temperature as predictor variables. However, these models could only predict a small portion of the clothing level variances and it was concluded that indoor and outdoor climate can only account for a small portion of human clothing behavior. This may be an indication of non-adaptive, perhaps habitual and task based nature of human clothing behaviors which underlines the inherent diffi- culty in predicting them. Similarly, beverage temperature selec￾tions (e.g. cold/hot drink) were observed to follow the long-term outdoor temperature variations [108]. The adjustment of the ac￾tivity rate was observed with respect to the indoor and the outdoor temperature [108,113]. However, given that in offices, activity was restrained with the task (e.g. sedentary), no strong relationship could be proposed based on these observations [108,113]. Fig. 4. (a) Window opening behavior of participants that defined themselves as passive, medium, and active occupants in a survey [40], (b) manual light use behavior in ten single or double occupied offices [56], (c) manual blinds use behavior in ten single or double occupied offices [64]. H.B. Gunay et al. / Building and Environment 70 (2013) 31e47 39

40 H.B.Gunay et aL Building and Environment 70(2013)31-47 Time (hour) presents a comparison of existing modeling methodologies with 810121416182022 their limitations and challenges. 3.1.Adaptive behavior models Seasonal vvariations Systematic observations on the system states (e.g.window 0.8 open/closed).once plotted with respect to the monitored variables (e.g.indoor temperature)resulted in a data scatter,as shown in Fig.6.Early researchers [118-120]and most of the current practi- tioners used deterministic models to predict the adaptive occupant variations behaviors.These models are simple enough to be easily incorpo- rated in the BPS-based design process.For example,in these models,the probability of an occupant being uncomfortable below .4 the defined threshold value is zero and the probability becomes one just after the predictor variable or variables reach the threshold value,as shown in Fig.6.A comparison between the data scatter JAN MAR MAY JUL SEP NOV and the probability curve,which is a step function,showed that a Time(month) deterministic model cannot predict the observed adaptive occu- pant behavior shown in Fig.6.Occupants'adaptive behaviors, Fig.5.Seasonal and diurnal variations in the mean clothing level reported by Haldi despite being influenced by the physical conditions,are governed and Robinson 106]. by a stochastic,rather than a precise,relationship [121].Stochastic models estimate an adaptive behavior by assuming a probabilistic 2.2.4.Non-physical parameters relationship with the predictor variable or variables.Numerous Morgan and De Dear [115]underlined social and cultural con- researchers [31.38,51,55]proposed using linear-response (e.g. straints that should be taken into account in predicting a model for linear or polynomial regression)models to estimate the probability clothing levels.For example,it was reported that women tend to of the adaptive occupant behavior as a function of a predictor wear less in summer and more in winter than men.However,a variable.Linear-response models assume a linear relationship be- recent work by Schiavon and Lee [107]reported that men and tween the response and predictor variables,as follows: women wear clothing at similar insulation levels.The dress codes in an office environment may restrain the ability to undertake Pi Bo +81x1i+82x2i++8mxmi (1) clothing adjustments.Haldi and Robinson [108]suggested that different models (e.g.strict dress code,casual working environ- where pi is the probability of success (e.g.probability of window ment,residential environment)can be established for offices with opening).B is the vector for the regression coefficients and x is the different dress codes.Similarly,drinking traditions(e.g.a cup of hot vector for the predictor variables(e.g.indoor temperature,outdoor coffee or tea in the morning of a summer day)cannot be directly temperature,CO2 concentration). associated with the physical variables.Researchers should Haldi and Robinson [108]reported linear regression as a sub- acknowledge these non-physical variables and try to incorporate optimal method to model the adaptive occupant behavior.It is their effects in the modeling process.For example,Haldi and Rob- evident that the linear regression model poorly predicts the upper inson [108]suggested to modify their stochastic clothing model for and the lower bounds of the observations as shown in Fig.6.This different offices with different dress-codes. can be explained since the linear-response models (e.g.linear or polynomial regression)are not appropriate to model response 2.2.5.HVAC system and operation variables that have a non-normal distribution.Generalized linear Newsham and Tiller [116]carried out a self-reported question- models(e.g.logistic regression or probit)cover such cases by letting naire survey in four fully conditioned offices during fall and winter. In this study,about 15%of the occupants reported that they adjusted their clothing in the previous hour.On the contrary.Haldi ++Observations and Robinson [108]surveyed office occupants in a naturally -----Deterministie Model ventilated office building and revealed that occupants rarely adjust Linear Regression Model Logistic Regression Model their clothing level during the day.This may imply that once the ”十 occupants are given the option to make changes in their environ- ment (e.g.opening a window),they undertake changes in their environment before they try to adapt to the environment (e.g. 0.8 clothing adjustments).This hypothesis opens a further discussion for researchers:whether or not the order at which the occupants 0.6 undertake adaptive behaviors can be stated in a statistically coherent way.In line with this,Andersen [117]reported that the order of the manual control sequence(e.g.thermostat-window 4 blinds-lights)may be responsible for up to 3.3 fold variation in the energy use predictions.This underlines the importance of 0.2 being able to predict the order of manual control actions. 3.Model prediction + Variable (0) Once researchers obtain their observations,they focus on Fig 6.Generic univariate deterministic,linear regression,logistic regression occupant establishing models that predict adaptive behaviors.This section models(data scatter is extracted from Nicol 119])

2.2.4. Non-physical parameters Morgan and De Dear [115] underlined social and cultural con￾straints that should be taken into account in predicting a model for clothing levels. For example, it was reported that women tend to wear less in summer and more in winter than men. However, a recent work by Schiavon and Lee [107] reported that men and women wear clothing at similar insulation levels. The dress codes in an office environment may restrain the ability to undertake clothing adjustments. Haldi and Robinson [108] suggested that different models (e.g. strict dress code, casual working environ￾ment, residential environment) can be established for offices with different dress codes. Similarly, drinking traditions (e.g. a cup of hot coffee or tea in the morning of a summer day) cannot be directly associated with the physical variables. Researchers should acknowledge these non-physical variables and try to incorporate their effects in the modeling process. For example, Haldi and Rob￾inson [108] suggested to modify their stochastic clothing model for different offices with different dress-codes. 2.2.5. HVAC system and operation Newsham and Tiller [116] carried out a self-reported question￾naire survey in four fully conditioned offices during fall and winter. In this study, about 15% of the occupants reported that they adjusted their clothing in the previous hour. On the contrary, Haldi and Robinson [108] surveyed office occupants in a naturally ventilated office building and revealed that occupants rarely adjust their clothing level during the day. This may imply that once the occupants are given the option to make changes in their environ￾ment (e.g. opening a window), they undertake changes in their environment before they try to adapt to the environment (e.g. clothing adjustments). This hypothesis opens a further discussion for researchers: whether or not the order at which the occupants undertake adaptive behaviors can be stated in a statistically coherent way. In line with this, Andersen [117] reported that the order of the manual control sequence (e.g. thermostat / window / blinds / lights) may be responsible for up to 3.3 fold variation in the energy use predictions. This underlines the importance of being able to predict the order of manual control actions. 3. Model prediction Once researchers obtain their observations, they focus on establishing models that predict adaptive behaviors. This section presents a comparison of existing modeling methodologies with their limitations and challenges. 3.1. Adaptive behavior models Systematic observations on the system states (e.g. window open/closed), once plotted with respect to the monitored variables (e.g. indoor temperature) resulted in a data scatter, as shown in Fig. 6. Early researchers [118e120] and most of the current practi￾tioners used deterministic models to predict the adaptive occupant behaviors. These models are simple enough to be easily incorpo￾rated in the BPS-based design process. For example, in these models, the probability of an occupant being uncomfortable below the defined threshold value is zero and the probability becomes one just after the predictor variable or variables reach the threshold value, as shown in Fig. 6. A comparison between the data scatter and the probability curve, which is a step function, showed that a deterministic model cannot predict the observed adaptive occu￾pant behavior shown in Fig. 6. Occupants’ adaptive behaviors, despite being influenced by the physical conditions, are governed by a stochastic, rather than a precise, relationship [121]. Stochastic models estimate an adaptive behavior by assuming a probabilistic relationship with the predictor variable or variables. Numerous researchers [31,38,51,55] proposed using linear-response (e.g. linear or polynomial regression) models to estimate the probability of the adaptive occupant behavior as a function of a predictor variable. Linear-response models assume a linear relationship be￾tween the response and predictor variables, as follows: pi ¼ bo þ b1x1;i þ b2x2;i þ / þ bmxm;i (1) where pi is the probability of success (e.g. probability of window opening), b is the vector for the regression coefficients and x is the vector for the predictor variables (e.g. indoor temperature, outdoor temperature, CO2 concentration). Haldi and Robinson [108] reported linear regression as a sub￾optimal method to model the adaptive occupant behavior. It is evident that the linear regression model poorly predicts the upper and the lower bounds of the observations as shown in Fig. 6. This can be explained since the linear-response models (e.g. linear or polynomial regression) are not appropriate to model response variables that have a non-normal distribution. Generalized linear models (e.g. logistic regression or probit) cover such cases by letting Fig. 6. Generic univariate deterministic, linear regression, logistic regression occupant models (data scatter is extracted from Nicol [119]). Fig. 5. Seasonal and diurnal variations in the mean clothing level reported by Haldi and Robinson [106]. 40 H.B. Gunay et al. / Building and Environment 70 (2013) 31e47