Conference PaperPDF Available

Modeling and simulation of individual user behavior for building performance predictions


Abstract and Figures

User behavior is a key to the correct prediction of building performance. Performance is meant here not only in relation to energy consumption, but more importantly to user satisfaction and, in commercial buildings, job performance. The contributions of this paper are static and dynamic models of individual users and user activities in an office environment in relation to building control systems. The models are based on a well structured general model of five different domains of building entities, including user activities. Models for specific simulation projects are refinements of this general model. It is shown how multi-agent based simulations can be automatically derived from such models. A case study of an university office with irregular occupancy is introduced as an example.
Content may be subject to copyright.
Keywords: User behavior, building performance simula-
tion, multi-agent simulation, event driven simulation, simula-
tion software engineering
Abstract User behavior is a key to the correct prediction of
building performance. Performance is meant here not only in
relation to energy consumption, but more importantly to user
satisfaction and, in commercial buildings, job performance.
The contributions of this paper are static and dynamic models
of individual users and user activities in an office environ-
ment in relation to building control systems. The models are
based on a well structured general model of five different do-
mains of building entities, including user activities. Models
for specific simulation projects are refinements of this gener-
al model. It is shown how multi-agent based simulations can
be automatically derived from such models. A case study of
an university office with irregular occupancy is introduced as
an example.
Contemporary buildings are becoming more complex in
structure, control, and operation. At the same time, demands
for energy savings, user comfort, and security for example are
increasing. Therefore, simulations during the design, con-
struction, commissioning, and operation phases are becoming
more and more important.
So far, building simulations are primarily performance
simulations. Most simulators assume simple algorithms for
the control of internal air temperature and humidity, illumi-
nance and other human comfort parameters and also assume
fixed schedules for occupancy. With today’s flexible work
hours and more sophisticated occupation based and interac-
tive controls these assumptions may lead to erroneous simu-
lation results, which do not reflect the situation of the
buildings during its operation.
Therefore, the behaviors of individuals and user groups
have to be modeled and integrated into building performance
simulators to get more realistic results. This is the goal of our
research work. Applications that we envision are for example
in the design and test of control systems for heating, cooling,
air quality, and natural lighting that react to individual user
preferences and interactions. Post-occupancy-evaluations
show that these features are considered as most important for
the user satisfaction [Hewitt et al. 2005]. Control systems are
designed today following engineering principles with as-
sumptions about user preference averages. This may lead to
theoretically optimal solutions, but in practice users can be-
have very differently from the assumed average. It is the ex-
treme user behavior that has to be considered more than the
We model all users and user groups as individual agents
with different behaviors. Independently of the users, we mod-
el different roles and functional units such as work places.
All three define user activities over time. The assignment of
individuals to roles to functional units can also change over
time. Even with a small number of types in each of the three
areas we can create a large number of behaviors to map the
At the current state-of-the-art we do not try to create a uni-
versal model of user behavior. User behavior is defined in a
huge feature space, with large areas still unexplored or not
understood. Our approach is to define domain specific mod-
els, start with limited scenarios for simulation applications,
and model as many features of users as necessary for each
scenario. For each scenario we will experimentally compare
simulation results with reality to verify the partial model and
then extend and generalize it.
In order to make such experiments feasible and efficient
we use a new software engineering approach. Based on a se-
mantically well defined modelling language we create specif-
ic models for each project that on the one hand are derived
from more general models, on the other hand can be automat-
ically translated into executable code. We call such models
executable models. To increase efficiency, reuse of model
components is mandatory.
To make this possible, a well structured, homogeneous
modeling environment is necessary. Therefore, not only indi-
viduals, roles, and functional units are modeled as communi-
cating agents, but also building components, service units,
and control system components. This approach eliminates the
problem of integrating different models and simulators.
Modeling and Simulation of Individual User Behavior for Building Performance
Gerhard Zimmermann
University of Kaiserslautern
SCSC 2007 913 ISBN # 1-56555-316-0
As a case study we use a real scenario of an open-plan uni-
versity office which faculty members, visiting scientists, and
graduate students share. The office is used 24 hours a day
with strongly varying occupation patterns. Currently a new
HVAC system is installed to individually control and condi-
tion work spaces based on occupancy. The goal of the simu-
lation project is to design and test control algorithms that
minimize energy consumption while maintaining user com-
User activity modeling is an emerging field in architec-
ture. A large body of work exists in urban planning in order
to predict pedestrian movements. Models and simulators
based on cellular automata and multiple agents are used, e.g.
[Dijkstra and Timmermans 2002]. Emergency evacuation
simulation is an important subfield. One very interesting
branch is the evacuation simulation of sinking ships [Meyer-
König et al. 2005], because real life experiments prohibit
themselves. Extensions that include user decision making
processes have been applied to household migration ques-
tions in cities as well [Devisch et al. 2006]. Urban planners
also investigate pedestrian behavior in buildings, for example
shopping centers and office buildings [Tabak et al. 2005].
Because not much is known about user activities in build-
ings, several experiments observe the motions of real users
with cameras and others means of locating people [Tabak et
al. 2005]. In another experiment user activities regarding
light control have been monitored, providing useful statistics
to improve lighting control [Mahdavi 2007]. Although
such experiments provide realistic user patterns, it is not easy
to extract the reasons for activities and thus derive abstract
dynamic models.
Early on, static building models have been developed that
incorporate user activities [Eastman and Siabiris 1995] and
[Eckholm and Fridquist 2000]. An extension of these models
can be found in [Zimmermann 2001]. The industry founda-
tion classes model [ifc2x3] has some entities that look like us-
ers and roles, e.g. ifcActor, ifcOccupant, ifcActorRole, but
they describe persons and data related to the building project
and management, not to users in the building. Also, all de-
scriptions are static. Therefore, ifc is not applicable for this
Agent models are also well suited for modeling intelligent
control systems, but it is not self-evident that the physical be-
havior of buildings can be mapped onto multiple-agents as
well. We showed in several case studies that a correct simu-
lation behavior can be achieved [Zimmermann 2002] and that
the simulators can reach a high degree of run-time efficiency
in accelerated real-time [Vityaz and Zimmermann 2005].
An initial case study using user activity modeling and
simulation has been made with the test of different lighting
control strategies in an office building with different user in-
teractions [Zimmermann 2006b]. It could be shown that user
interactions together with semi-automatic controls result in
drastic energy savings under study assumptions. The current
paper is an extension of this work to heating and cooling and
a much more sophisticated user activity model.
Executable models are composed of static and dynamic
descriptions. In relation to agents the static model defines
agents, subagents and communication channels between
agents. The dynamic models describe the dynamic behavior
of the agents and the messages.
The general domain for our models is that of buildings
and building environments. In order to structure the model,
we define five subdomains:
user activity domain
functional unit domain
control domain
service domain
building domain
The control domain models all control modules and func-
tions, structured by the spheres they control. The service do-
main models the zones for heating, cooling, lighting etc. and
the related equipment. The building domain models all geo-
metric spaces and the related building fabric components
with their physical properties and dynamic behavior. Details
of these domains can be found in [Zimmermann 2003]. The
first two domains will be described in more detail in the fol-
lowing chapters.
3.1. Static Models
Several levels of refinement are used to reduce the com-
plexity of the model, as shown in Figure 1. Two different re-
lations are used between these levels: specialization
(inheritance) and instantiation. The three top levels define a
modeling language or ontology. The system level introduces
the five subdomains and relations between entities in differ-
Figure 1: Modeling specialization levels. The triangle
stands for generalization, the straight line for instantiation
ISBN # 1-56555-316-0 914 SCSC 2007
ent domains. The system level also introduces the distinction
between immaterial (space) and material (matter) entities.
The first specialization is the domain level, giving a specific
structure to each sub domain, as for example Figure 2. A fur-
ther specialization, the application domain level, introduces
entities and relations typical for an application domain, for
example office buildings, hospitals, or homes. This partition-
ing reduces the modeling complexity and introduces a termi-
nology for communication with field experts.
From the application domain model project level models
are derived by instantiations of agent types and relations that
are parts of a specific application, as for example the case
study. By keeping the number of types small and by another
instantiation into agents with individual parameters, the mod-
eling effort can be kept reasonable. This last step is done au-
tomatically by compilers and creates the run-time code for
execution. Instance parameters can be introduced at run-time.
One common feature of all project level models is that no
inheritance is used. The reason is that the modeling language
SDL-92 [Olsen et al. 1994] which we are using incorporates
inheritances, but it is not very obvious what it means in the
behavioral models, specifically in state transition graphs. In
order to keep the meaning of the models clear to the modeler
and also to domain experts, we only use composition as struc-
turing relation.
Composition relations create tree structures of agent types
which are also projected into the instance structure at run-
time. A strict model architecture has been introduced to sim-
plify the construction of agent type structures: all agent types
have the same basic structure. They are composed of a pro-
cess that defines its behavior and of 0..n subagents. Each
agent has one input and one output port and all the ports of the
subagents are only connected to the parent process. This pro-
vides a simple communication structure along the branches of
the tree. Such tree structures have the advantage that subtrees
can be easily replaced by other subtrees without changing the
basic communication structure and thus support model reuse
and modifications. A disadvantage is that direct communica-
tions between agents without direct composition relations is
not possible. This is compensated by a second communica-
tion network based on remote procedure calls between agents
which can be made visible in the structure diagrams by addi-
tional relations.
Figure 2 shows the user activity domain model. All enti-
ties are types of agent types, marked by a trailing T. On the
immaterial left side of the diagram roles of user groups and of
individuals are composed of activities. Correspondingly, the
roles are realized by groups of individuals and by individuals.
This distinction of roles and persons is an important means to
be able to let individuals or groups take or quit roles dynam-
ically. For example, a person may have different jobs or a job
is done by different persons during a day. An individual has
personal properties and preferences, whereas a role deter-
mines the workflow of a job.
In the application domain, roles can be further special-
ized, for example in university professor, student, or secretary
roles. Also, the activities can be refined to lecturing, desk
work, moving, meeting. The case study will show how these
entities are used.
The functional unit domain model in Figure 3 is of similar
simplicity. This is of great importance for our software engi-
neering approach to produce specific simulators for each ap-
plication with the smallest possible effort. By keeping the
number of different types of agents small, the modelling, im-
plementation, and testing effort is also small.
3.2. Dynamic Models
The dynamic models are the key to simulation. We use
different representations to describe the models during the
development process. On the one hand, the models have to be
descriptive enough for modelers and domain experts to un-
derstand and communicate with each other. For example, the
Figure 2: user activity domain model. The trailing T indi-
cates that all entities are types of agent types
Individ ualT
Figure 3: Functional unit domain model
SCSC 2007 915 ISBN # 1-56555-316-0
modeling expert may be a computer scientist, the domain ex-
pert an architect, a work-flow expert, or a control engineer.
Therefore, the basic representation is always textual, al-
though in a structured way to be able to manage text compo-
nents in a structured software development process (see
[Zimmermann and Metzger 2004] for more details).
A more formal representation is based on Message Tran-
sition Charts (MTC) as shown in Figure 4. In order to explain
the need for MTCs we have to explain the behavior of the
used model architecture. Multi-agent simulations consist of
many independent agents that exchange messages between
themselves and with the environment. Agents have internal
states and functions, react to and produce messages. There-
fore, a perfect model of an agent is a process with states, each
executing on an individual digital processor. This would
achieve full concurrency of processes and is a good abstrac-
tion during modeling. Since this hardware effort is not neces-
sary, many or all processes are mapped onto one processor by
a scheduler that simulates the behavior of many processors.
Each process is modeled as an extended finite state ma-
chine (EFSM) with a message input queue. Messages and
timers cause state transitions. A state transition can perform
computations, changing the states of local variables, and send
new messages. State transitions can depend on conditions.
We use the graphical modeling language SDL-92 [Olsen et
al. 1994] and a tool environment “Tau SDL Suite” for editing,
syntactic and semantic analysis, and debugging [Telelogic].
This environment also provides a code generator that trans-
lates the static and dynamic models into executable code and
a run-time environment for virtual- or real-time simulations.
SDL is sufficient to describe models in all details, at the
structural level it also uses an abstraction of the composition
and communication structure, but it lacks the ability to show
the interaction of messages and state transitions in an abstract
form. Message Sequence Charts (MSC) are supported by the
“Tau SDL Suite” for this purpose, but we found MTCs a bet-
ter abstraction to develop and document complicated scenar-
ios. We will show an example in Chapter 5.2.
Let use start with user groups and individual users. We
only model the begin and the end time of the work day and
personally induced activities a user performs during the work
day. Times off work are of no interest for office building do-
main simulations.
Start and stop times of all activities are modeled as sto-
chastic variables with individual parameters. Typically some
parameters can change for every day of the week, others are
constant. Since, at the moment, we do not have data about
measured probability distributions, we use uniform distribu-
tions with average and range as parameters.
Typical personal activities during a work day are: get a
drink, go to toilet, take a break. All job or role related activi-
ties are modeled in the role agent.
Other individual specific parameters are preferences re-
lated to comfort. They are sent from the individual to the plac-
es and are typically temperature and humidity ranges and
light levels for different activities. These values may change
with the season of the year.
Roles of groups and individuals are defined by tasks.
Each day the dynamics of roles are determined by schedules
of task related activities. We distinguish:
continuous activities: typically the work done between
other activities, e.g. desk work
regular activities: executed at fixed times and dura-
tions, e.g.lectures, lunch break
irregular activities: executed spontaneously or trig-
gered from the environment, e.g. ad hoc meetings
secondary activities: induced by the above activities,
e.g. copying, filing
These activities are modeled only to the degree of detail
that influences the experiment. For example, if we test occu-
pancy detector controlled service systems, it does not matter
what is done at the desk or at a meeting. It only matters if a
place is occupied or not, at most the number of persons to
control air quality. But if task specific automatic lighting is
tested, reading, manual drawing, or computer terminal usage
should be distinguished, if possible with sensors.
Scheduling is based on task priorities. Regular activities
can be scheduled for a work day when the role is taken by an
individual. Irregular activities are also scheduled at this point
randomly in between activities with higher priority, based on
a total number per workday or a frequency of events. Contin-
uous activities just fill the time that is left over. Typically, an
Figure 4: Simple MTC example with state transitions (top)
and abstractions (bottom). Thick arrows denote messages
agent1 agent2
ISBN # 1-56555-316-0 916 SCSC 2007
individual performs only one continuous activity for the job,
for example desk work. This may be the main task of the role,
but in practice it gets a low priority and is often interrupted.
If a minimum duration for the continuous task is defined by
the role, this may lead to extending the work day beyond the
time planned by the individual or group.
Secondary activities are not scheduled. They are created
randomly by an activity, based on role specific frequencies.
Personal activities interrupt role activities by messages from
the individuals, if the priority is set high enough. If the prior-
ity is too low, two things can happen: the interrupt is forgot-
ten, because the activity is no longer necessary (drinks are
served at meetings) or handled on the way (go to toilet com-
ing back from a meeting), or the personal activity is post-
poned until after the current activity.
For one important aspect of user behavior we have no so-
lution yet. Users react to changes in the work environment in
different ways that are very difficult to predict. Depending on
past experience, origin, age, sex or other factors, people will
open or close windows and shades, change thermostat or dim-
mer settings and do all sorts of things that the control strategy
would like to avoid. Systems that prevent such interactions
produce user dissatisfaction. Also, users adapt to the environ-
ment and change their comfort parameters. Therefore, the
generally accepted indoor comfort functions do not provide
reliable user satisfaction predictions. More analyses in real
work environments are necessary before such aspects of user
behavior can be modeled for predictions. The NEAT project
[Loftness et al. 2006] is a good source of data, but these are
not yet developed into an abstract model.
Before project domain models are introduced, the case
study is shortly explained. The office example is a part of the
Robert L. Preger Intelligent Workplace (IW) that is used by
the Center of Building Performance and Diagnostic, Archi-
tecture Department, Carnegie Mellon University, Pittsburgh,
for staff and student workplaces. It also serves as a lived-in
laboratory for HVAC and lighting experiments for teaching
and research purposes.
The office is in principal open-plan, but due to the regular
roof structure, 5m wide parallel bays partition the space. In
the south end of the IW, 5 bays are further partitioned into 2
office spaces and a central hall space, providing 10 offices.
The partitions are created by furniture of different height, not
reaching the ceiling. These separations provide some visual,
acoustic, and thermal isolation, but some air can still circulate
between adjacent spaces.
3 offices are occupied by one professor each, 1 office is a
meeting place, 6 offices have desks for 3 graduate students
and visiting scientists each. Due to lectures, meetings, and
personal work hour preferences the occupation pattern is very
irregular, extending throughout most of the nights and week-
ends. Existing occupation patterns have been documented in
interviews with occupants.
Air conditioning is currently provided by forced air from
ducts under the raised floor, provided by a central air han-
dling unit. In addition, water mullions, that are vertical water
pipes between the window elements, provide radiation above
or below air temperature to compensate the window radia-
tion. The mullions also act as radiators.
Because of the irregular occupancy, the whole office is
heated or cooled 24 hours 7 days per week to guarantee user
comfort at all times. This situation will be changed in the fu-
ture by inserting fan coil units in the floor and by controlling
each office separately using occupancy sensors. Central air
supply shall be reduced just to provide air quality and to con-
trol humidity. Several control strategies shall be compared to
find the best compromise between energy savings and com-
The purpose of the case study is to test control strategies
with realistic individual user behavior by simulation before
the hardware is installed to be able to select and program con-
trol algorithms to be implemented when the hardware be-
comes operational. Only in such a way the office can be used
with minimal disruption.
As further experiments it is planned to experiment with
different office separation heights, with different occupancy
detection schemes, with different combinations of radiation
and air temperature to provide comfort, and with natural
lighting control. Also, different sets of individual comfort
preferences will be tested. Simulation outputs are satisfaction
ratings (based on thermal comfort and air quality) and energy
consumption. l
For the purpose of the case study, a static agent type mod-
el and a dynamic model have been derived from the applica-
tion domain level as introduced in Chapter 3.
Figure 5: Agent type structure with instance names and car-
FctUnitDomain ControlDomain ServiceDomain BuildingDomain
Individu al
IndivRole ProfPlace
SouthS paceusg1
ppl1.. 3
udm1 fdm1 cdm1 sdm1 bdm1
fcu1.. 10
rom1 ..10
FCUSuppl yCtrl
SCSC 2007 917 ISBN # 1-56555-316-0
5.1. Static Model
Figure 5 shows the agent structure of the complete simu-
lator for the case study. The components can be easily recog-
nized by descriptive names. Each domain is represented by a
common node, for example on the left the UserActivityDo-
main. These nodes have data distribution and collection func-
tions. The node IWSouth represents simulation control
functions such as time synchronization and file handling. The
root node is the instance management node GenSys that at
system start propagates unique instance names to all agents
and retrieves unique process ids for remote procedure calls.
Unique names of nodes are constructed by concatenating all
instance names of the tree branch above the node. GenSys
also provides typical functional extensions to SDL that are
used in simulations and control systems. The two top nodes
are part of a universal simulation environment, the five do-
main nodes are reused in all building system simulations, all
other nodes are project specific, but can be modified and re-
used in different projects. In our process, reuse means reuse
of models, since code is automatically generated.
5.2. Dynamic models
In Figure 6 the main state transitions and messages are
shown in an MTC. At first sight this looks rather confusing
and we will try to follow the main path of events for a typical
scenario from left to right. Depending on individual prefer-
ences and role requirements, startTimer triggers the begin of
a work day. Each individual takes up a role which starts with
scheduling the regular and irregular tasks for the work day.
Typically, the individual moves from the entrance to its
workplace, using CircPlace.CircPlace knows the distances
and sends a message back to Move to set a timer accordingly.
CircPlace also counts the individuals for lighting control.
Once at the workplace, e.g. StudentPlace, the continuous
activity is executed until IndividualRole triggers another
move to another place according to the schedule or the indi-
vidual takes a break. In the example breaks can only be taken
while at the workplace. Breaks at other places are assumed to
be either skipped or combined with the other activity. While
at the workplace, secondary activities are scheduled random-
ly according at role induced frequencies. These interruptions
are handled in the same way as breaks and are not explicitly
shown in the diagram.
Workplace occupancy changes environment require-
ments, such as air temperature, light setpoints, and fresh air
volume in the RoomCtrl. This results in changes of valve and
fan speed settings of the fan coil units and other parameters.
The scenario shows how many agents are involved in the con-
trol system of an office, if user activities are considered.
MTCs are a good means to create a dynamic model that can
be refined and translated into an SDL model. Group meetings
are different scenarios, not shown here. Additional MTCs
have been created for the interaction of the control agents
with service and building domain agents. It has to be pointed
out that MTCs are only an intermediate step in the modeling
process and do not have to be complete. Completeness is re-
quired from SDL models only.
As already introduced in Chapter 3, user activity models
are so far based on stochastic variables that govern entry and
exit times of all activities. The parameters of the distribution
functions are different for each experiment setup. In experi-
igure 6: MTC of the domains User Activity, Functional Unit Domain, Control Domain for the main workplace scenario
Reaction Chain
acti ng
acti ng
IndividualRole Move DeskWork
acti ng
acti ng
StudentPlace RoomCtrl
ISBN # 1-56555-316-0 918 SCSC 2007
ment 4 the data from interviews with occupants in the IW are
used, the other experiments use constructed parameters.
Scheduling is performed in a very simple way, based on
the personal experience of the author. Coming into work in
the morning, only the regular tasks are scheduled, everything
else develops during the day. Attempts for optimal schedules
are not the reality. Priority for the main continuous activity is
normally low, only increased in the case of strict deadlines.
Any more sophisticated scheduling strategies would not pro-
vide better experimental results in the case study.
As already mentioned, the formal SDL-92 models are ex-
ecutable. This means that they contain all information neces-
sary to execute a simulation. Because SDL semantics are well
defined, models can be translated by compilers into execut-
able code.
SDL uses two types of graphical representations: one de-
scribes the static component hierarchy of each agent type and
its instances, thus representing structures as in Figure 5. The
other describes the dynamic model of each agent as processes
and communication interfaces, as explained in Chapter 3.2. In
this way SDL is used as a very high level graphical program-
ming language that is translated automatically into a high lev-
el language, in this case into C. This is the basis of our
software engineering approach.
On the basis of this, a software engineering process
PROBAnD [Metzger and Queins 2002] has been defined
with additional structured text documents to guide the design-
er from informal descriptions to formal models. The process
is also based on the model architecture shortly described in
Chapter 3. Together with a simple version management and a
development time recording tool the process is made as effi-
cient as possible. This allows the fast creation of models and
simulators for every simulation project from scratch if neces-
sary. This process is further supported by a tool PROTAGO-
NIST that translates structured textual descriptions into SDL
models [Metzger and Queins 2003]. In projects with known
simulation principles a development time of about 2 person-
hours per agent type could be achieved. In the current project
with 38 new agent types and some new principles, the time
had to be doubled. This effort is still in the time range neces-
sary to personalize commercial simulation environments, if
they existed for all five domains.
All experiments were conducted with weather files from
a station in the University of Kaiserslautern [Litz]. All data
are recorded every minute, providing a sufficient resolution
even for light and radiation on cloudy days. From 2003, days
1 to 5 of each month have been combined to 60 consecutive
days. The simulation is set up to compute each hour in one
second real-time on a laptop with an 850MHz Pentium pro-
cessor. One simulation run takes 24 minutes accordingly,
with set-up times about 30 minutes.
For the shown simulations all 10 office spaces were as-
sumed to be separated by walls with indefinite heat resistance
to give the largest possible differences between occupancy
patterns. Also, the AHU and the mullions have been turned
off. All controllers have been assumed to be simple P-con-
trollers. These simplifications have been used to concentrate
on the dynamics and the influence of individual user activi-
The simulations of all physical components are based on
first principles. Room dynamics consider the heat capacitanc-
es of walls and furniture. Hot and cold water dynamics are
modeled by time delays. Heating and cooling energy calcula-
tions are based on supply water temperature differences and
flows, integrated over time. The sampling time periods of all
different agents are adjusted to the individual time constants.
Table 1 shows some simulation results. Since heating and
cooling is provided by the university in the form of hot and
chilled grid water supplies and mixed to appropriate temper-
atures in local supply circuits, energy calculations are based
on water flow and the supply and return temperature differ-
ences. Exp 1 shows the worst case, 24h heating and cooling
with daytime setpoints because of the irregular work hours of
the occupants. In contrast, the best case in Exp 2 assumes no
occupation at any time and thus 24h nighttime setpoints.
Exp 3 shows the energy consumption of an office with
regular work hours, in the middle between best and worst
case. Under the assumption that the occupants in a university
office are less than 9 hours at their desk because of classes
and meetings, but at irregular times of the day, a good occu-
pancy controlled air conditioning system should reach the
same values as in Exp 3. Exp 4 happens to show this result.
Any other occupancy pattern will give different results. Also,
the energy consumption will grow with increased heat perme-
ability of the inter office walls.
Table 1: Experimental results
Exp Description Heating
1 24h/day 16.1kWh 1.8kWh 17.9kWh
2 0h/day 11.9kWh 0.8kWh 12.7kWh
3 8-17/day 13.9kWh 1.0kWh 14.9kWh
4 actual usage 13.8kWh 1.1kWh 14.9kWh
SCSC 2007 919 ISBN # 1-56555-316-0
Our main result is that user activities of individuals and
groups in office environments can be modeled on the basis
communicating agents. By separating persons from roles and
work places, the individual agents are relative simple entities.
They are personalized by parameters such that a small num-
ber of agent types has to be modeled and implemented. Inte-
gration into building performance simulations is achieved by
modeling the building fabric, service system components and
control subsystems as agents as well in the same modeling
and implementation environment.
Experiments show that this approach is feasible, provides
tangible results, and executes in acceptable run times despite
automatically generated simulation codes.
As further extensions, the building and service models
will be refined and different control strategies will be devel-
oped and tested, including user satisfaction models. Also, the
simulated user activity patterns will be compared with obser-
vations of real patterns and changed accordingly. It is planned
to use the observations in other environments to extend, re-
fine, and generalize the user activity model. Also, its applica-
tion in building space utilization simulations will be tested.
Another, more important goal is the modeling and simu-
lation of user reactions to new building control and service
strategies. Humans are very adaptable and often counteract to
the intentions of energy saving systems, thus letting the inten-
tions fail. Simulation could help to predict and avoid such
This research is based on results from earlier projects
funded by the Deutsche Forschungsgemeinschaft and is par-
tially supported by the Center for Building Performance,
CMU. The author also has to thank for the use of the experi-
mental facilities and informations of the Robert L. Preger In-
telligent Workplace.
Devisch, O.T.J et al. 2006, “Modeling Residential Search
and Location Choice.” In Proc. 8th International Design
and Decision Support Systems Conference, Einhoven,
Dijkstra, J. and Timmermans, H. 2002, “Towards a multi-
agent model for visualizing simulated user behavior to
support the assessment of design performance.” Automa-
tion in Construction, 11, 135–145.
Eastman, C. M. and Siabiris, A. 1995, “A generic building
model incorporating building type information.” Auto-
mation in Construction. 3: 283–304.
Eckholm, A.and Fridquist, S. 2000, “A concept of space for
building classification, product modelling, and design.”
Automation in Construction 9, 315–328.
Hewitt, D. et al. 2005, “A Market-Friendly Post-Occupancy
Evaluation: Building Performance Report.” Final Report-
Contract C 10091, New Buildings Institute.
Loftness, V. et al., 2006, “Workplace 20•20 Field Evaluation
of Baseline Environmental Quality & NEAT National
Environmental Assessment Toolkit”, ABSIC Meeting.
Mahdavi A. 2007, “People as Power Plants - Energy Impli-
cations of User Behavior in Office Buildings.” Proc.
Intern. Energiewirtschaftstagung (IEWT), Vienna.
Meyer-König T. et al. 2005, “Implementing Ship Motion in
ARENAS - Model Development and First Results.” In
Proc. Pedestrian and Evacuation Dynamics, Vienna, 429-
Metzger, A. and Queins, S. 2002. “Specifying Building
Automation Systems with PROBAnD, a Method Based
on Prototyping, Reuse, and Object-orientation.” GI-Edi-
tion, Lecture Notes in Informatics (LNI), P-5. Bonn:
Köllen Verlag, 135–140.
Metzger, A and Queins, S. 2003, “Model-Based Generation
of SDL Specifications for the Early Prototyping of Reac-
tive Systems.” Telecommunications and beyond: The
Broader Applicability of SDL and MSC. Springer Lec-
ture Notes in Computer Science, LNCS 2599. Heidel-
berg: Springer-Verlag, 158–169.
Olsen, A. et al. 1994. System Engineering using SDL-92.
Amsterdam: Elsevier.
Tabak, V. et al. 2005. “Model for Office Building Usage
Simulation.” In Proc. Pedestrian and Evacuation Dynam-
ics, Vienna 2005, 391-403.
Telelogic Tau SDL Suite,
Vityaz, O. and Zimmermann, G. 2005, “Real-Time Simula-
tion Using Graceful Degradation of Accuracy”, Energy
and Buildings, Volume 37, Issue 8, August, 795-806.
Zimmermann, G. 2001. “A new approach to building simula-
tion based on communicating objects.” Seventh Interna-
tional IBPSA Conference Proceedings. Vol. 2. Rio de
Janeiro, Brazil, 707–714.
Zimmermann, G. 2002. “Efficient creation of building per-
formance simulators using automatic code generation.
Energy and Buildings. 34. (2002): 973–983.
Zimmermann, G. 2003. “Modeling the building as a system.”
Eighth International IBPSA Conference Proceedings.
Eindhoven, Netherlands. (2003): 1483–1490.
Zimmermann, G and Metzger, A. 2004. “A software-genera-
tion process for user-centered dynamic building system
models.” Proc. ECPPM 2004, Istanbul.
Zimmermann, G. 2006a, “Multi-Agent Model to Multi-Pro-
cess Transformation - A housing market case study.” In
Proc. 8th International Design and Decision Support Sys-
tems Conference, Einhoven, 203-219.
Zimmermann, G. 2006b, “Modeling and Simulation of
Dynamic User Behavior in Buildings - a Lighting Con-
trol Case Study.” In Proceedings ECPPM'2006, Valencia.
ISBN # 1-56555-316-0 920 SCSC 2007
... O microclima local pode variar em todo o parque edificado, desde locais rurais altamente expostos à insolação e ventos, até centros urbanos afetados pelo efeito de ilha de calor. O comportamento dos ocupantes é uma complexidade adicional que envolve múltiplos fatores: número de pessoas, idade, padrões de ocupação entre outros (CARLUCCI et al., 2021;CASSOLA et al., 2022;FIRTH;LOMAS;WRIGHT, 2010;MANN;BECERIK-GERBER, 2012;PAONE;BACHER 2018;RAFSANJANI;AHN;ALAHMAD, 2015;SOOMRO et al., 2019;ZIMMERMANN, 2007). ...
Full-text available
A formulação de políticas e esquemas de incentivo para redução do consumo de energia nos edifícios demandam conhecimento das características dos estoques edificados para entender como o seu desempenho energético pode ser melhorado. O objetivo deste trabalho é identificar, com relação ao sistema de condicionamento de ar instalado, a forma, dimensões e percentual de abertura da fachada, edifícios representativos que possam ser utilizados como referência na análise de desempenho energético de edifícios de escritórios. O estudo concentrou-se na cidade de Florianópolis. Uma amostra de 197 edifícios de escritórios foi estudada, adotando a abordagem K-means como metodologia de classificação de 187 edifícios quanto à sua forma e dimensões e 179 edifícios quanto ao percentual de abertura das fachadas. Para o sistema de condicionamento de ar instalado, a análise baseou-se na frequência de ocorrência. Como resultado da análise é possível afirmar que os edifícios de escritórios de Florianópolis podem ser representados por plantas de duas formas, em quatro tamanhos distintos: (a) quadrado pequeno, medindo 16 m x 15 m (considera tolerância para a diferença entre as duas medidas) e (b) quadrado grande, medindo 23 m x 23 m; (c) retângulo pequeno, medindo 15 m x 24 m e; (d) retângulo grande, medindo 15 m x 34 m. Os edifícios da tipologia utilizam majoritariamente sistemas de condicionamento individual providos por aparelhos de janela ou split. O edifício de escritórios típico de Florianópolis possui 33% de área envidraçada na fachada norte, 30% na fachada sul, e 24% nas fachadas leste e oeste.
... A probabilistic model to simulate and predict occupancy in a single-person office was proposed by examining the statistical properties of occupancy (Wang et al., 2005). Taking the university office with irregular occupancy as a case study, a static and dynamic model of the individual occupant and occupant behavior in an office environment in relation to building control systems were validated (Zimmermann, 2007). A new, open-source modeling tool was set for stochastic simulation to predict occupant services demand in buildings (Rysanek & Choudhary, 2015). ...
Full-text available
The excessive simplification of occupant behavior is considered as the most important factor that affects the uncertainty of building performance simulation, thus affects the reliability and generalizability of simulation-based design and forecast. In this paper, occupant behavior in air-conditioned office buildings of the Pearl River Delta (PRD) region was investigated and defined. Copies of 873 questionnaires about the occupant behavior in air-conditioned office buildings in the PRD region were collected to study the relationship between indoor environment quality and adaptive behaviors. Eight typical office occupant schedules were defined via K-means clustering method. A probability prediction model of cooling temperature set-point was established by using the Ordinal Logistic Regression method. According to the different control modes of air conditioning, window, blind and lighting equipment, four types of typical behavior patterns were proposed using the K-prototype clustering method, which could be developed into 20 typical occupant behavior styles of office buildings in the PRD region.
... At the same time, the study shows that available technologies are heavily underused in the industry. In addition, buildings are becoming increasingly complex in terms of their structure, operation and control [137], which means that even property owners with modest property portfolios have trouble keeping track of their assets and use ever-increasing amounts of digital building information. ...
Full-text available
The construction industry lacks a comprehensive and holistic way of utilising Building Information Modelling (BIM) throughout a building’s life-cycle, where the transition to Asset Management (AM) and Facility Management (FM) is particularly lacking. Enterprise BIM (EBIM) is an emerging, unexplored, holistic organisational concept that is designed to support and optimise business management throughout the entire life-cycles of buildings and infrastructure. However, current understanding and implementation of EBIM is rare. To fix this gap in knowledge, this paper defines the EBIM concept and assesses associated perspectives from research and practice in order to integrate BIM more strongly into the enterprise’s core processes and other needs at different levels within the organisation and externally. To achieve a higher and more efficient utilisation of building data, however, EBIMs need to develop a clear life-cycle-based information management strategy, including coordination and communication between all stakeholders. The paper highlights and discusses the importance of both available and missing standards related to the effective implementation of EBIM. Several existing open standards have been identified to help strengthen the EBIM concept. The paper identifies a need to develop standards to integrate BIM and IoT (the Internet of Things) and standards that can handle both structured and unstructured data. The authors have conducted a wide-ranging review of research, practice and existing standards and considers today’s potential for using EBIM, as well as discusses existing challenges and future research needs. Furthermore, the EBIM concept is contextualised by providing a case study with focus on cleaning. This study identifies a need to develop best practices in interdisciplinary collaborative projects. Such practical implementation requires integrated and standardised information and technology management and the exploitation of the available technologies in interdisciplinary interaction with those involved in the various processes and the flow of information throughout the life-cycle.
... Agent-based modeling uses controlled computational experiments and synthetic data to conduct research, a point of departure from the either inductive or deductive approaches typical in social science (Axelrod & Tesfatsion, 2006). It is widely applicable as an approach that can be used for making predictions about the results of parameter (or independent variable) changes through synthetic data generation (Bonabeau, 2002;Jackson et al., 2017;Malatesta et al., 2009;Zimmerman, 2007). For example, once an interaction is discovered, researchers can test the bounds of the interaction by changing the model's parameters. ...
Substantially advancing the study of teams will require a new research paradigm complete with methods capable of capturing the complex, dynamic process of teamwork. In this paper, we suggest studying teams with an integrated mixed methods approach (i.e., methods defined by an interconnected mix of quantitative and qualitative characteristics) can help address current methodological shortcomings of our science by promoting sufficiently contextualized research. Through a review of methods, we highlight exemplars of integrated mixed methods that have the potential to be more widely adopted; namely, interaction analysis, content analysis, cluster analysis, state space grids, and agent-based modeling.
... However, those models are more suitable for use in existing buildings. For new building projects, user simulation models (Zimmermann 2007;Goldstein, Tessier, and Khan 2010;Tabak, de Vries, and Dijkstra 2010) can display occupancy patterns over time, on the basis of hypothetical occupancy schedules. To generate these schedules automatically, user activities are often classified into several groups (e.g. ...
This empirical study sought to validate the spatial-choice modelling approach to user simulation. This validation is challenging, because spatial-choice behaviour is difficult to observe in real-life settings. To overcome this challenge, the study assembled 1137 activity cases from time-use diaries kept by university students and then extracted 74 space-rejection cases (i.e. events when students were unable to take up a particular space-option and so moved to another one). We assessed the attributes of the space-options in the rejection cases by means of a site investigation, predicted space-use probabilities using spatial-choice models, and assessed forecasting accuracy by a Brier score. The spatial-choice models scored 0.35, demonstrating higher forecasting accuracy than two alternative methods, namely space-use analysis (0.50) and the closest distance method (0.84). This study contributes to user simulation by validating the accuracy of the spatial-choice models based on revealed spatial-choice data in real-life settings and comparing it with other methods.
... Since activities have different properties, they cannot all be generated and scheduled with the same methodology. Thus, to make the activity-based framework possible, a classification of activities is created such that each category contains activities with similar properties whilst all categories together allow representing the full range of activities occurring in various situations [3,10]. These categories are further grouped according to their association with a time-of-day (time-dependent or time-independent) and these groups each have their own scheduler. ...
Full-text available
As the number of people working in office buildings increases, there is an urgent need to improve building services, such as lighting and temperature control, within these buildings to increase energy efficiency and well-being of occupants. A pedestrian behaviour model that simulates office occupants’ movements and locations can provide the high spatial and temporal resolution data required for the testing, evaluation, and optimization of these control systems. However, since most studies in pedestrian research focus on modelling specific actions at the operational level or target situations where movement schedules do not have to modelled, a pedestrian behaviour model that can simulate complex situations over long time periods is missing. Therefore, this paper proposes a tactical level model to generate occupant movement patterns in office buildings. The Markov-chain activity-based model proposed here is data parsimonious, flexible in accepting different levels of information, and can produce high resolution output. The mathematical properties of the methodology are analyzed to understand their impact on the final results. Finally, the tactical level pedestrian behaviour model is face validated using a case study of an imaginary office with a simple layout.
... This can be interpreted as follows. Since collision avoidance failure was determined after a normal navigational situation, both models had normal navigation situations [44][45][46]. In particular, the success model reflects more of the normal navigation situation compared to the failure model. ...
Full-text available
In order to prevent ship collisions, it is important to understand the behavior of navigators that leads to these collisions. The main cause of marine accidents in the Republic of Korea is attributed to navigator error, particularly in collisions. Hence, reducing navigator error is a key issue that needs to be addressed to prevent accidents. However, the lack of objective measure to quantify navigator error remains a challenge. The purpose of this study is to develop an objective identification of a navigator’s behavior in a collision encountering situation. Two behavior models for the success and failure of collision avoidance are developed by collecting participants’ actions, using a ship maneuvering simulator within a given scenario. These maneuvering behavior models are validated in terms of their discrimination powers. The results show that maneuvering behavior is clearly identified in the data processing and model development phases. The proposed behavior models are expected to provide a better understanding of how navigators behave to help reduce collision accidents.
... The ability of users to control the use of space make user behavior become the determinant of optimal building design, both in terms of convenience and use of energy [14]. User behavior is used to predict the performance of buildings especially for the purpose of saving energy and efficient simulation [15,16]. User behavior can be assessed as culture, presenceaction-interaction and satisfaction. ...
Full-text available
This doctoral dissertation explores how computational modelling, simulation and sensitivity analysis techniques can be used to appropriately represent occupants and their behaviors when predicting the energy and comfort performance of buildings. More specifically, the aim of this work is to develop a computational methodology that can be used to facilitate the model selection of various OB aspects in order to achieve fit-for-purpose OB modelling. The methodology comprises four main steps. The first three steps (the Impact Indices Method, the Diversity Patterns Method, and the Mann-Whitney U test) are developed to identify which OB aspects are truly influential for the simulation output. These three steps can be seen as a sequence of increasingly complex sensitivity analyses. The fourth step concerns dealing with the influential OB aspects. The outcomes of this research show that the computational methodology presented here is able to produce BPS models that include an adequate (fit-for-purpose) representation of occupants and their behavior, while satisfying the purpose of the simulation in an efficient and effective way.
This paper, deals with the usually need of data to simulate behavior and efficiency of proposed solutions in several fields, and also knowing that personal data always bring privacy and security issues. This work wants to promote a balanced solution between the need of personal information and the user’s privacy expectations. We propose a solution to overcome these issues, and don’t compromise the balance between security and personal comfort based on generating real context data of users, that allow to test user dependent systems.
Conference Paper
Full-text available
We define building systems as systems that not only consist of the load bearing building structure and its environment, the service systems (installations), and the control systems, but also include building usage and users. Our ultimate goal is a virtual building labo- ratory for the concurrent simulation of all components of such building systems. This will be an extension of performance simulators. In this paper we focus on an efficient modeling process for deriving object-orient- ed structural models for virtual building laboratories, with the specialization towards the development, test, and maintenance of integrated control systems. We also focus on control systems that support individual user environments and that can easily be adapted to changes in the use of building systems. For the pur- pose of efficient modeling, we divide the building sys- tem domain into five domains, called building, service, control, functional unit, and user role domain. We present corresponding domain models that pre- define domain dependent modeling primitives for project specific building system models. All models are based on the notion of space and matter. In each domain, spaces have special semantics and build ag- gregation hierarchies, which we call " backbones". The backbones of the domain models are linked by require- ment and spatial relations, thus composing one inte- grated building system model.
Real-time computer simulation is required in a number of applications, where the simulated dynamic systems are directly coupled with systems in the real world or with other systems also running in real-time. Typical problems of real-time simulations are that the dynamics of the modeled system are not represented in enough detail, either by not sampling the inputs or not producing output values frequently enough or both. Increasing the sampling frequencies may lead, as an additional problem, to real-time violations caused by insufficient computing power. The latter problem is amplified if the real-time is accelerated to achieve a kind of fast motion and reduce experiment times. In this paper, we show how the first problem can be solved by automatic frequency adjustment, minimizing the number of computations and how the remaining second problem can be reduced by trading a graceful degradation of accuracy against the computing power requirements automatically. Experiments show that simulation times could be reduced by a factor of 100 without unacceptable simulation errors.
This paper presents a model for office building usage simulation which is part of the research project “User Simulation of Space Utilisation”. The aim of this project is to develop a model for the simulation of human movement to predict space utilisation in office buildings. The underlying model takes two main types of input: firstly the workflow of the organisation and secondly the design of the building in which the organisation is (or will be) housed. The latter refers to the spatial conditions. The model generates data about the activity the members of the organisation perform and their location in the building space for the specific time interval of the simulation run. Proceeding from these data, performance indicators can be deduced, like average/maximum walking distance per individual, time occupied per space and number of persons per space. These indicators can be used to analyse the performance of a design or can act as input for other engineering methods and models.
Since the first software programs for evacuation simulations were applied to ships, questions about the influence of the ship motions, and how these could be taken into account, came up. As the acceptance of simulation tools was pushed by the developments at the International Maritime Organization, research was undertaken by various institutes and the appropriate data was collected. The paper describes, how a model for the implementation of ship motions into the software AENEAS was developed and how the ship motions influence the results of the evacuation analysis. The implementation of ship motion is part of a research project, funded by the German Ministry of Transportation (BMVBW) and organised by the Hamburgische Schiffbau-Versuchsanstalt (HSVA).
Building performance simulation environments are very complex software systems and the result of many person years of development and testing. The instantiation of a specific building requires additional work, depending on the quality of the user interface. We will show that with today’s software engineering environments, the development time can be drastically reduced. Simulators for specific applications can be created in a few weeks and discarded after use. A prerequisite for our development method is the partitioning of the building and its installed systems into simple components that compute and interact at run-time autonomously, as in the real world. In contrast, classical building performance simulators are typically based on global systems of differential equations that model the physical reality and are numerically solved at run-time. Our approach is based on object-oriented modeling. All effects are modeled dynamically and are executed in real-time. New physical effects can be easily integrated. As calculations are triggered by value changes, very short real-time responses are achieved if necessary. Also, fast and accurate responses to external events can be guaranteed. These features are necessary for real-time tests of building automation systems. We will present data showing that our approach results in realistic simulations and show limits of real-time computations.