Occupancy Driven Uncertainty in Electric Lighting Use: Modeling a

Case Study
Todd Otanicar, Michael Eribez, and Jeff Robertson, Loyola Marymount University
Jennifer Scheib and Elaine Hale, National Renewable Energy Laboratory

ABSTRACT

Uncertainty in building occupancy can lead to significant variations in electric lighting

use. In its first six months of operation, the National Renewable Energy Laboratory’s Research
Support Facility (RSF) exceeded its modeled electric lighting energy budget because of
unanticipated occupancy patterns in the evenings. This experience highlights the potential
usefulness of modeling occupancy uncertainty and its impact on loads such as lighting energy
use, while at the same time pointing out the difficulty of appropriately bracketing and
quantifying all reasonable scenarios. In this paper we build on the RSF example to develop
models for uncertainty in electric lighting use caused by uncertainty in building occupancy and
interactions with lighting controls, especially in office buildings with open work space. The
models so developed may be used by engineers and architects to understand how lighting control
strategies affect energy usage under occupancy uncertainty. We also use our models to reflect on
the RSF modeling process, and subsequent experience with the physical building and lighting
use mitigation strategies.

Introduction
Building energy simulations are typically relied upon to provide a design-point
calculation of the building energy consumption, but rarely are the simulations utilized to assess
real-world performance variation. Most building operators recognize that there is some level of
uncertainty in building use, operations, and weather conditions, often leading to a significant
difference in predicted and actual energy consumption (Clevenger & Haymaker 2006).
Additional uncertainties and inaccuracies arise from the simplifications necessary to model
complex building geometries, equipment, and controls systems in an energy simulation program
(Wit & Augenbroe 2002). These two types of uncertainty lead to challenges in the design
process, particularly in the case of design optimization and decision making (Eisenhower et al.
2011; Hopfe 2009; Jacob et al. 2010; C. Struck & P. Kotek 2009; Christian Struck et al. 2007).
Much of the current literature uses uncertainty as a means to assess the sensitivity of the
building’s energy performance to variation in a large number of variables (Eisenhower et al.
2011; Wit & Augenbroe 2002), or to a limited subset of variables, such as construction materials
(C Struck & J Hensen 2006), and schedules (Clevenger & Haymaker 2006; Zhao et al. 2011).
Occupant behavior, which in energy simulations is modeled using schedules, presents a
particularly interesting challenge since occupants not only impact energy use directly, but also
interact with lighting and other control systems. Recent experience at the National Renewable
Energy Laboratory’s Research Support Facility (RSF) highlights the importance and difficulty of
capturing such model uncertainties. Capturing and appropriately modeling uncertainty within a
building will allow for better prediction of actual energy usage as well as allowing designers to
incorporate systems to help mitigate uncertainty in energy usage.

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-212

 The National Renewable Energy Laboratory is home to the Research Support Facility
(RSF), which began operating in mid 2010. The RSF was designed with an extremely demanding
energy target (35 kBtu/sf/yr normalized including a data center, approximately half of typical
office buildings in the Denver metro area (Guglielmetti 2011)). According to the 2003
Commercial Building Energy Consumption Survey (CBECS), lighting directly accounts for 25%
of an office building’s site energy use, and 39% of its electricity use, on average (EIA 2003).
Thus, detailed attention was paid to the design of the RSF’s lighting systems, particularly in the
open offices and corridors, where very high levels of daylighting are used to minimize this
energy end use. Additionally, an aggressive electric lighting control sequence was implemented
to try to limit use outside typical work hours by automatically sweeping off a majority of lights
(with the exception of a single light over entrances, exits, and stairwells).
Within the first few months of operation it was observed that the lighting energy use was
significantly higher than expected, primarily on weekdays outside of prime working hours (8am
– 5pm). Part of this was attributed to cleaning and maintenance crews working from 6-10pm,
turning the lights on, but not switching them off so that they stayed on until the next sweep at 8
or 10pm. This pattern is evident in Figure 1, which shows the deviation between the expected
and actual lighting use broken down into day types (weekday and weekend/holiday), hour and
month. The median values in Figure 1a demonstrate that the predicted weekday lighting use was
accurate from hours 1-16 and 23-24, but in hours 17-22 a significant unexpected shift was
observed. The large variances in hours 7-17, and to some extent, the additional energy use seen
in those hours (a moderate positive shift of the median value) is attributable to variations in
daylight availability, which are particularly pronounced in January. In contrast, there is little
variance in electric lighting use in hours 17-22, but rather a more deterministic shift associated
with the unanticipated presence and behavior of occupants during these hours. As a fraction of
the averaged metered load for January the variance can be significant particularly during hours 7-
17 while the variance outside of these hours is small in relation to the average load. The amount
of variance in relation to the average is even larger for the weekends but it is a much smaller
average load.
Overall, the data shown in Figure 1 shows that the lighting energy use of open-plan,
daylit office buildings varies day-to-day and hour-to-hour, in response to both weather and
occupant interactions. It is also easy for even a highly engaged design team to miss entire
occupant populations and their likely behaviors. This case study highlights the need for a better
understanding of the (stochastic) relationships between weather, occupants, controls, and electric
lighting use; and also the need to somehow capture possible occupancy patterns outside of the
primary use of the space. Understanding these relationships further will help both in the design
phase and the actual building use phase. By developing better understanding, including
stochastic relationships, building designers will be able to incorporate uncertainty into the design
phase and create designs which help mitigate future energy use uncertainty. Additionally the
understanding will be useful when a building is in operation for the ability to test out new
lighting control strategies to reduce unwanted variation not initially captured or controlled after
the design phase.

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-213

 Figure 1. Difference between Metered Load and Modeled Load, a) January 2011
Weekdays, b) January 2011 Weekends/Holidays.
(The whiskers show the minimum and maximum values, the second quartile is red, and the third quartile is green.)
a)
75
Average Metered Load 29.6 kW
Difference in Lighting Load, Metered ‐Modeled (kW)

65

55

45

35

25

15

‐5

‐15

‐25
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour of Day

b)
50

Average Metered Load 9.1 kW

Difference in Lighting Load, Metered‐Modeled (kW)

30

10

‐10

‐30

‐50

‐70
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour of Day

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-214

 The need for modeling occupant behavior is well recognized in the literature (Bourgeois
et al. 2005; Wang et al. 2005; Clevenger & Haymaker 2006), and many models have been
proposed. Furthermore, the potential impact on lighting use has also been investigated
(Clevenger & Haymaker 2006; Bourgeois et al. 2005). Only two of these works have
investigated the uncertainty with occupant behavior (Zhao et al. 2011; Clevenger & Haymaker
2006); One of these is primarily focused on the development of advanced models to incorporate
uncertainty into the occupant behavior while the other takes a much simpler high/medium/low
approach to understand the sensitivity of total energy use. Occupant behavior has been
statistically investigated by a variety of authors under very specific conditions not representative
of the open office environment. Wang et. al looked into developing a statistical occupancy model
for a single office and determined that vacancy, not occupancy, could best be modeled with an
exponential distribution (Wang et al. 2005). Others have taken multiple sets of commercial
building office data and tried to link occupancy and lighting usage statistically but are limited to
manual controls (Rubinstein et al. 2003; Mahdavi & Pröglhöf 2009) but have shown that
occupancy and lighting may be correlated further through the total illuminance available upon
arrival (Mahdavi & Pröglhöf 2009). While these studies have shown that occupancy and lighting
can be potentially modeled with statistical models they have all been limited to manual controls
and primarily private offices.
In this paper, we model, measure, and study the interaction of occupant behavior and
electric lighting energy use in open office spaces typical of the RSF. Detailed occupancy and
lighting use data was collected from several RSF spaces, post the interventions (training of the
custodial staff operating during the evening hours) taken to address the initial measured to model
deviations seen in the evening hours. The next section describes the spaces in which the data
were collected. We then go on to analyze the data in an initial attempt to gain understanding into
the possibility of developing probabilistic models. Additionally we note the challenges and
limitations of the current data set and the data needs to develop more detailed models. While
uncertainty in daylight availability should ultimately be folded into these models as well, that is
outside the scope of this paper, although we hope to take up that question as future work.

Modeling Occupancy Effects on Electric Lighting Use

In this section we focus on describing the space both monitored and modeled. We first
present the space and describe the lighting design and operation, followed by a description of the
data collection that was performed. Finally we statistically investigate two primary spaces to try
to further understand the interaction of occupancy with lighting use and develop statistical
models for quantifying the occupancy effects on electric lighting use.

RSF Space Descriptions

We investigated two spaces: a large open office and a kitchen serving the open office
(Figure 2). The RSF is divided into east and west wings. All the data collected in this study
comes from the east wing. During the design-build contract period, the open office space lighting
was modeled, designed, and commissioned for aggressive electric lighting energy savings
through the use of coordinated daylighting, ambient and task electric lighting, and ambient
lighting control schemes. The integrated design of the building relies heavily on space
programming and the details of the interior finish. All office surfaces are light in color, and most

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-215

 occupants are located in cubicle workstations with low walls. Semi-private offices are located on
the north wall to minimize their impact on the daylight saturation of the rest of the space. Along
with daylight redirection technologies, the overall design gives very high daylight saturation that
conforms to the LEED IEQc8.1 daylighting requirements. The open office was iteratively
designed and modeled to maximize daylight saturation. The final daylighting design, control
equipment, and sequences were modeled in detail to predict actual, required ambient lighting
power density and usage schedules for different space types.
The lighting controls ensure that the space lives up to its potential in terms of electric
lighting energy use savings. The open office wings are sixty feet wide, and are divided in two
electric lighting switching zones. The perimeter zones use a bi-level switching control that turn
the lights on, half-on, or off to daylight availability. These systems account for one third of the
installed lighting power density. In the core of the open office space, a continuous dimming
system takes advantage of available daylight. These lights are never fully turned off by the
dimming controls, but only in response to manual switching. The specific control design is for
lights to be turned on manually by occupants if needed as they enter the space in the morning,
guided by always-on night lights near entrances and light switches. As daylighting increases
throughout the day, electric lighting in all offices and corridors will dim or switch off in
response. Manual on/off capability by occupants is always enabled. At 6:30 pm, a timed event
cause all open office and corridor lighting to blink, warning occupants that the lights will turn off
in five minutes. Occupants can override the sweeps, which occur every two hours throughout the
night, that is, at 6:30pm, 8:30pm, etc. All enclosed spaces such as kitchens and closed offices are
controlled based on local occupancy control. The kitchens are true occupancy sensors in that if
the light switch is in the on position then the lights will turn on when occupancy is detected. The
closed offices use vacancy sensors that must be manually turned on at each arrival.
The kitchen lighting design accounted for daylight availability and occupancy-triggered
controls, but was not modeled aggressively from a daylighting point of view because of both
location and size. There are no daylighting controls, and occupancy sensors were modeled with a
simple 10% LPD reduction as recommended by ASHRAE 90.1. The kitchen is comprised of
undercabinet and overhead lighting that is controlled via line voltage, manual switching and
auto-off overrides triggered by the switch being on but the space vacant. The occupancy sensor
turns lights on automatically only if the manual switch is in the on position.

RSF Occupancy and Lighting Data Collection

A challenge in creating occupancy-lighting models is that underlying any data set is a

lighting control scheme that influences occupant interaction with a lighting system. This
underlying occupancy-lighting system interaction limits the applicability of a model, built upon
building specific real-world data, that can be used for effective prediction of a new lighting
control scheme. Although this underlying challenge is significant, we can still use data to drive
the development of the models while keeping in mind how the data factors into the model
development. Due to the large space size of the open office space it is not possible to achieve
full coverage occupancy logging. Rather, occupancy loggers are used to track whether people are
entering or exiting the space. Although this limits the true knowledge of occupancy it is still of
relevance for determining the interactions of people with the lighting system.
The kitchen space is small enough to have full coverage with the occupancy loggers
allowing for direct capture of lighting use and occupancy. The monitoring of lighting usage in

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-216

 this space provides interesting insight into the impact of daylighting on occupant behavior. As
previously mentioned the kitchen spaces were not designed with aggressive daylighting but do
receive some low level daylighting (~5 fc) that may impact the observations seen with logged
data. The location of the occupancy and lighting sensors are highlighted in Figure 2 for the open
office space type considered. The occupancy and lighting sensors are not displayed within the
kitchen space due to the relatively small size. All spaces were monitored from January 7 through
February 12, 2012 using WattStopper IT-200 occupancy and light loggers.

Figure 2. Approximate Location of Lighting and Occupancy Sensors in Open Office Space
and Location of Spaces within RSF Building

RSF Occupancy and Lighting Data Analysis

In this section the occupancy and lighting use variations are investigated as a function of
day type and time of day. Specific concern is paid to what type of statistical model will best fit
the data and how well occupant level variation can be used to explain lighting use variance.

Kitchen space analysis. The first step was a simple breakdown of the variations observed for
weekdays and weekend/holidays. The data are grouped by hour and utilize a method where an
occupancy of 1 for a given hour means the space was fully occupied for the hour, if 0.5 then it
was occupied 50% of the time which may be exaggerated due to the timeout settings (five

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-217

 minutes) of the loggers. The same structure applies to lighting for the kitchen space. While
observing the kitchen space on weekend days only limited occupancy was observed with zero
lighting usage. Figure 3 displays the average weekday occupancy and lighting usage for the
kitchen space. The data shows that from 11PM-6AM the occupancy is zero and lighting usage is
zero or very low. When the typical work day starts both occupancy and lighting increase in the
morning, decrease until a lunchtime peak, then in the afternoon it is observed that occupancy is
generally very small but lighting usage is seen to spike at 7PM, most likely associated with
custodial staff presence. The results also indicate a change in behavior with the presence of
daylighting. As noted the kitchen spaces do have a small amount of dayligthing allowing users
to most likely perform quick tasks (opening refrigerator, throwing out trash, etc.) without turning
on the lights but more lengthy tasks (preparing coffee and meals) will likely require turning on
some form of lighting. Because of this we investigated specific hours to look at a more detailed
level of occupancy lighting interactions at key hours.

Figure 3. Average Hourly Weekday Occupancy and Lighting Usage for the Kitchen Spaces
(solid line – occupancy, dashed squares– undercabinet lights switching profile, dashed
triangles – overhead lights switching profile)
0.9 0.3

0.8
0.25
0.7
Occupancy Profile

0.6 0.2

Switching Profile
0.5
0.15
0.4

0.3 0.1

0.2
0.05
0.1

0 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour of Day

Further investigation into the data used in figure 3 reveals additional interesting
conclusions about occupancy interactions with lighting. For key selected hours the occupancy
profile for each weekday is plotted against the switching profile on weekdays for both the
overhead (two-thirds of the total kitchen fixture light output) and undercabinet lights (one-third
of the total light output). The switching profile represents the fraction of the time within a given
hour that the lights are on to full power output (i.e. if the switching profile is point 0.5 the lights
were on 50% of the time for that hour, if at full power, or were on the full time at 50% power).
Figure 4a and 4d exhibit what initially appears to be a relatively strong correlation of low
occupancy to low lighting but upon investigating other hours this breaks down. Figure 4b and 4c

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-218

 are the hours of 12PM and 8PM respectively both have similar mean lighting levels but
significantly different mean occupancy profiles. It should be noted that for any given hourly or
total weekday set of occupancy lighting data, no simple linear correlation could be found with an
R2 value greater than 0.5. This result significantly differs than that reported in (Mahdavi &
Pröglhöf 2009) where a much more significant correlation could be found. Our initial conclusion
is that two underlying effects are occurring that are not captured here: daylighting and task.
With respect to daylighting it can be seen in Figure 3 with the significant decreases in lighting
without significant decreases in occupancy. The task effect can be seen at the prime lunch hours
12 and 1 PM where the lighting also peaks while the daylighting is also at a maximum indicating
that even though there is some daylighting more complex tasks in the kitchen like preparing
lunch leads to more lighting usage. This result is similar to what was previously reported where
a strong correlation was seen between illuminance level and relative frequency of switching on
lights (Mahdavi & Pröglhöf 2009).

Figure 4. Relationship of Occupancy to Lighting Switching Profile for Hours: a) 7AM, b)

12PM, c) 8PM, d) 10PM (Mean And Standard Deviations for Occupancy, Overhead, and
Undercabinet Lights respectively)
a) 7 AM b) 12 PM
1.00 1.00
0.90 Mean: 0.23, 0.02, 0.14 0.90 Mean: 0.71, 0.11, 0.16 Overhead
Standard Deviation: 0.07, 0.03, 0.09 Standard Deviation: 0.19, 0.09, 0.12 Undercabinet
0.80 0.80
Switching Profile
Switching Profile

0.70 Overhead 0.70

0.60 Undercabinet 0.60
0.50 0.50
0.40 0.40
0.30 0.30
0.20 0.20
0.10 0.10
0.00 0.00
0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00
Occupancy Profile Occupancy Profile

c) 8 PM d) 10 PM
1.00 1.00
Mean: 0.09, 0.08, 0.19 Mean: 0.08, 0.05, 0.06
0.90 0.90
Standard Deviation: 0.10, 0.14, 0.21 Standard Deviation: 0.06, 0.08, 0.08
0.80 0.80 Overhead
Switching Profile

Switching Profile

Overhead
0.70 0.70 Undercabinet
Undercabinet
0.60 0.60
0.50 0.50
0.40 0.40
0.30 0.30
0.20 0.20
0.10 0.10
0.00 0.00
0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00
Occupancy Profile Occupancy Profile

While the results shown above demonstrate that more complex interactions may be
needed further study was taken to investigate how a joint occupancy lighting model, a
probabilistic model that includes uncertainty in occupancy that drives the uncertainty in lighting
usage, may be developed. The first step was to assess the probabilistic distribution of both
occupancy and lighting. Selected hours are plotted in figure 5 for the occupancy distributions
which display a strong dependence on time of day. Such histograms show that occupancy profile

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-219

 may best be represented by a beta distribution where the two shape parameters would be time
dependent.

Figure 5. Kitchen Weekday Occupancy Histograms: a) Hours 20, 21, and 22, b) Hours 9,
11, 12 & 13
a)
20
Hour 20
18 Hour 21
16 Hour 22
Frequency( Days)

14
12
10
8
6
4
2
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Occupancy Profile

b)
14
Hour 9
12 Hour 11
Hour 12
10
Frequency (Days)

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Occupancy Profile

Similarly we investigated the lighting profiles and it was revealed that all followed an
exponential type distribution for the kitchen space. We stopped short of developing the
distributions proposed due to the need for further investigation of additional effects.

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-220

 Open office space analysis. Similar to the kitchen space we investigate the monitored lighting
usage within the open office space. As noted the occupancy sensors cannot provide full
occupancy coverage but rather provide an indication of someone entering the space, deemed the
entrance profile. Again it should be mentioned that the open office space type in the RSF was
designed to achieve high levels of daylighting throughout the space. Figure 6 presents the
weekday variation in switching profile and entrance profile for the open office space. From
Figure 6 the impact of daylighting can mildly be observed due to the less than 100% switching
profile during near maximum peak occupancy. It also becomes clear that the late afternoon,
early evening hours have significant lighting use potentially reflecting occupant behavior
differences between groups who arrive early versus stay late at work, although this could also be
linked to daylighting and highlights the need for additional monitoring of daylighting
illuminance levels. As mentioned previously this is also compounded with cleaning and
maintenance crews working primarily in the evening hours.

Figure 6. Average Hourly Weekday Entrance and Switching Profiles for the Open Office
Spaces (solid line – entrance profile, dashed – swithcing profile)
1 1
0.9 0.9
0.8 0.8

Switching Profile
Entrance Profile

0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
1 3 5 7 9 11 13 15 17 19 21 23
Hour of Day

In addition to observing the hourly trends a similar investigation into the frequency
distributions of the switching profiles to that of the kitchens was performed. The impact of
daylighting is strongly observed in Figure 7 by observing the frequency distributions associated
with various times of the day. In the early hours the switching profile is typically low while
during the middle of the day (peak daylighting) the switching profile follows almost a unform
distribution, with the late evening hours having high levels of switching profiles. Figure 7
reveals a novel progression that seems to be previously unrecognized in the literature of a highly
time and occupant dependent statistical distribution that changes distribution type depending on
the level of daylighting but also on the occupant. The first effect, daylighting, is linked to the
time of day where it can be observed that switching profiles frequency distributions vary as a
function of the time of day. Particularly of interest are the potentially uniform distributions
during hours of peak daylight. The second effect, the occupant, which is just primarily
qualitative at this point is based upon the assumption that office workers working later into the
evening may have different lighting desires than those arriving early in the morning. This effect

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-221

 is certainly intertwined with the first effect and further justifies the need for further study. Based
upon these results further investigations are warranted to also try to track illuminance levels in
the space as well as come up with a way to get a more realistic grasp on the actual occupancy.

Figure 7. Variation in Open Office Switching Profile Frequency Distribution as a Function

of Hour of Day

14
8 11 12 13 17 18
12
Frequency (Days)

10

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Switching Profile

Conclusions
Through the detailed data collection and statistical modeling performed within it was
shown that using occupancy as the sole variable for describing lighting use variations was not
adequate. In particular the kitchen space where a small amount of daylighting is available
lighting use did not correlate with occupancy or with hours of maximum daylighting indicating
that additional factors are important. One potential factor is the task of the occupant in
conjunction with the available daylighting leads to different lighting use scenarios. For the open
office space type daylighting seems to be a much stronger factor in affecting the interaction of
occupants with a lighting system. Of particular interest is the changing frequency distribution of
the switching profiles which ranges from exponential to normal to uniform depending on the
time of day and daylighting. This dynamic suggests that further studies are needed for both
space types to get a full understanding how occupant uncertainty leads to lighting uncertainty
particularly in the presence of daylighting and spaces where occupants perform a variety of
different tasks.

References
Bourgeois, D., Reinhart, C. & Macdonald, I.A. 2005. "Assessing the total energy impact of occupant
behavioural response to manual and automated lighting systes." In 9th International IBPSA
Conference. Montreal, Canada, pp. 99-106.

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-222

 Clevenger, C.M. & Haymaker, J., 2006. "The Impact of the Building Occupant on Energy Modeling
Simulations." In Joint International Conference on Computing and Decision Making in Civil and
Building Engineering, Montreal, Canada. Citeseer, pp. 1-10.

[EIA] Energy Information Administration. 2003. No Title. Commercial Buildings Energy

Consumption Survey. Available at: http://www.eia.gov/emeu/cbecs/contents.html [Accessed
2012].

Eisenhower, B. et al., 2011. "Uncertainty and Sensitivity Decomoposition of Building Energy

Models." Journal of Building Performance Simulation.

Guglielmetti, R., Scheib, J., Pless, S.D., Torcellini, P.A., Petro, R. 2011. "Energy Use Intensity and
Its Influence on the Integrated Daylighting Design of a Large Net Zero Energy Office Building.
"ASHRAE Winter Conference, Las Vegas, Nevada.

Hopfe, C.J., 2009. Uncertainty and sensitivity analysis in building performance simulation for
decision support and design optimization Christina Johanna Hopfe. Eindhoven University of
Technology.

Jacob, D. et al., 2010." Optimizing building energy simulation models in the face of uncertainty" In
SimBuild2010. pp. 118-125.

Mahdavi, A. & Pröglhöf, C. 2009. "Toward empirically-based models of people ’ s presence and
actions in buildings." In Eleventh International IBPSA Conference. Glasgow, Scotland, pp. 537-
544.

Rubinstein, F. et al., 2003. Analyzing occupancy profiles from a lighting controls field study,
Available at: http://escholarship.org/uc/item/8t92v307.

Struck, C & Hensen, J, 2006. "Uncertainty analysis for conceptual building design a review of input
data." In 1st International IBPSA Germany/Austria Conference BauSIM. pp. 9 - 11.

Struck, C. & Kotek, P.. 2009. "On the Application of Uncertainty and Sensitivity Analysis with
Abstract Building Performance Simulation Tools." Journal of Building Physics, 33(1), pp.5-27.

Struck, Christian, Kotek, Petr & Hensen, Jan, 2007. "On incorporating uncertainty analysis in
abstract building performance simulation tools. "In 12th Symposium for Building Physics. pp.
193-205.

Wang, D., Federspiel, C.C. & Rubinstein, F. 2005. "Modeling occupancy in single person offices."
Energy and Buildings, 37(2), pp.121-126.

Wit, S.D. & Augenbroe, G., 2002. "Analysis of uncertainty in building design evaluations and its
implications." Energy and Buildings, 34, pp.951-958.

Zhao, Z.B., Xu, W.S. & Cheng, D.Z., 2011. "Uncertainty and Sensitivity Analysis of Building
Energy Demand Model Based on Occupant Behaviors." Applied Mechanics and Materials, 71-
78, pp.411-415.

©2012 ACEEE Summer Study on Energy Efficiency in Buildings 7-223