Automation in Construction 19 (2010) 9399

Contents lists available at ScienceDirect

Automation in Construction
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / a u t c o n

Review

Overview of HVAC system simulation

Marija Trka , Jan L.M. Hensen
Eindhoven University of Technology, PO Box 513, 5600MB Eindhoven, The Netherlands

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

Article history: The paper gives an overview of heating, ventilation and air-conditioning (HVAC) system modeling and
Accepted 30 November 2009 simulation. The categorization of tools for HVAC system design and analysis with respect to which problems
they are meant to deal with is introduced. Each categorization is explained and example tools are given.
Keywords: Further, the paper summarizes current approaches used for modeling (i) HVAC components, (ii) HVAC
Building system performance simulation
control and (iii) HVAC systems in general. After giving an overview of solution techniques for HVAC system
HVAC performance simulation
Integrated building performance simulation
simulation, the paper provides suggestions on how one should select an HVAC modeling approach relative to
Solution techniques for HVAC simulation the simulation objective at hand.
HVAC modeling and simulation approaches 2009 Elsevier B.V. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2. Tools for HVAC system design and analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3. Modeling approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.1. Modeling approaches for HVAC components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.2. Modeling approaches for HVAC control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.3. Modeling approaches for HVAC systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.4. Solution techniques for HVAC system simulation models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4. Integration of building and HVAC system models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5. Issues in selecting HVAC modeling approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

1. Introduction partial integration of different performance aspects of buildings, e.g.

thermal energy, visual, and acoustical. The current fourth generation
Forty-year long development of building performance simulation tools tend to be fully integrated with respect to different building
(BPS) tools resulted in a wide range of currently available products performance aspects, with new developments concerned with intelli-
[1,2]. These products range (complexity-wise) from spread-sheet gent knowledge-based user interfaces, application quality control and
tools to more advanced special-purpose simulation tools, and user training. The current tools can capture reality much better than
(integration-wise) from tools that handle a single aspect of the earlier tools, but are more complex to use.
building design, to tools that integrate multiple aspects of the building The number of currently available BPS tools, diversity of aspects
design [3]. taken into account in those tools and modeling approaches used by
A brief historical overview BPS tools is given in [4]. First generation of those tools makes writing a general overview of the eld a difcult task.
BPS tools is based on simplied methods found in handbooks Thus, this paper is restricted to an overview of modeling and simulation
(calculations based on analytical formulations that embody many sim- developments in one of the more important subsystems in buildings:
plifying assumptions). Second generation tools are based on methods heating, ventilation and air-conditioning (HVAC) systems.
that assume simplied (still analytical) modeling of dynamics in
buildings. Third generation tools use numerical methods and provide 2. Tools for HVAC system design and analysis

Tools for HVAC design and analysis can be categorized with respect
Corresponding author. to the problems they are meant to deal with. Although these problems
E-mail address: [email protected] (M. Trka). are not mutually exclusive, and some tools can handle several problems,

0926-5805/$ see front matter 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.autcon.2009.11.019
94 M. Trka, J.L.M. Hensen / Automation in Construction 19 (2010) 9399

they do tend to be investigated in isolation from each other. The The system simulation models that belong to this category are
categories are as follows. expected to predict system performance accurately. Thus, they need
Tools for pipe/duct sizing are system design tools that consider ow to be able to treat the departures from ideal behavior that occur in real
distribution and sizing of liquid/air distribution system (AFT Fathom, systems and to realistically model controls and HVAC system
DOLPHIN, Duct Calculator, DUCTSIZE, Pipe-Flo, PYTHON, etc.). dynamics. The tools for energy performance analysis can be used as
Tools for equipment sizing and selection offer HVAC equipment tools for real-time optimization of system performance [21,22], but
sizing (Carrier HAP, Trane TRACE 700, EnergyPlus, etc.). Most sizing models of a building and its systems need to be well calibrated [23]. In
tools are based on standard procedures and algorithms established by general, well calibrated rst-principle models can be used [24], but
e.g. American Society of Heating Refrigeration and Air-conditioning simpler and precise empirical (e.g. neural network models) models
Engineers (ASHRAE), but many are proprietary software products can be used as well [16].
distributed or sold by equipment manufacturers [5]. Digital catalogues
that are provided by equipment manufacturers can be used to locate a 3. Modeling approaches
suitable component model for the given design criteria. They can be
further linked to the equipment sizing tools, e.g. Carrier's HAP tool can 3.1. Modeling approaches for HVAC components
be linked to their chiller selection tool by importing performance data
for the actual chiller. According to Zeigler [25], the majority of models in building and
Tools for energy performance analysis are designed to predict the system performance simulation are:
annual energy consumption of an HVAC system. Based on a system of
Continuous in state, as the range of model variables is represented
equations that dene thermal performance of buildings and sys-
by real numbers or intervals. However, some models assume a
tems, and with given boundary conditions, operation strategy and
discrete set of values and are thus discrete state models.
controls, these tools perform (hourly or sub-hourly) simulations
Discrete in time, as time is specied to proceed in discrete steps. If
(Carrier HAP, Trane TRACE 700, DOE-2, eQUEST, EnergyPlus, ESP-r,
the model is continuous in state and discrete in time, it is then
IDA ICE, TRNSYS, HVACSIM+, VA114, SIMBAD, etc.). These tools are
described by a (system of) difference equation(s).
typically used to calculate and analyze the full- and part-load perfor-
Deterministic. However, stochastic models are used as well, e.g. in
mances, to analyze system operation strategy, to compare different
predictive control applications [20].
design alternatives, etc. [69].
Time varying, since the rules of interaction are different at different
Tools for system optimization are used in conjunction with tools for
times.
energy performance analysis. In multiple simulation runs, a set of
Both steady state and dynamic.
parameters is optimized according to a given objective function. An
Forward, as they are used to predict the response of output variables
example is the generic optimization tool GenOpt [10].
based on a known structure and known parameters when subjected
Tools for control analysis and control optimization (see also
to input and forcing variables. Backward (data-driven) models1 tend
Section 3.2). The level of HVAC system control modeling and simu-
to be much simpler but are relevant only for cases when system-
lation in the available tools varies:
specic and accurate models of specic building components are
Controllers can be associated with high abstraction system models, required, e.g. for fault detection and diagnosis [16].
such as in ESP-r.
There is a distinction between primary and secondary HVAC
Controllers can be represented explicitly either
system components. The former are sometimes referred to as plant,
- as models of supervisory control, such as in EnergyPlus, or
and the latter are referred to as system. A primary system converts
- as simple models of local control, such as in ESP-r and TRNSYS.
fuel and electricity and delivers heating and cooling to a building
More advanced representation of controllers, such as fuzzy logic, are through secondary systems. Examples are: chillers, boiler, cooling
available in e.g. MATLAB based tools (SIMBAD), Dymola and tools towers, thermal storage systems, etc. Secondary systems include air-
coupled to MATLAB (ESP-r [11], TRNSYS [12]). These tools are handling equipment, air distribution system and liquid distribution
efcient for design and more comprehensive testing of controllers in system between the primary system and the building interior.
a simulation setting [13], as well as for testing and validation of In both primary and secondary systems there are two types of
controller design in real time [14]. components: distribution components and heat and mass balance
Simulation tools for real-time performance optimization. Benets of components. The distribution components are: pumps, fans, dampers,
using simulation tools in the building operational stage are still valves, ducts and pipes. They affect the energy ow in buildings by [26]:
insufciently explored. Simulation tools could be used for: consuming electrical energy which drives pumps and fans, and
Commissioning diagnostics (initial commissioning): i.e. to verify the transferring thermal energy to/from the working uid in all
performance of the whole building, its subsystems and components distribution components.
[15]; The distribution component models should satisfy energy and mass
Monitoring diagnostics (continuous commissioning) and fault balance equations. Most of the BPS tools model distribution components
detection diagnostics: i.e. to detect, analyze, locate and/or predict in a simplied way [26], which eliminates the need to calculate the
problems with systems and equipment occurring during everyday pressure drop through distribution system at off-design conditions. In
operation [1619]; general, this approach is sufciently accurate for studying temperatures
Emulating a building and its HVAC systems: i.e. simulating the in the system. For detailed analysis of e.g. fan/pump control loops and for
response of a building and its HVAC systems to building energy answering questions related to the placement of the return/exhaust fan,
management system (BEMS) commands. Emulators can also be type and size of dampers/pipes, ow and pressure balancing between
used for control product development, training of BEMS operators, the components is necessary [18].
tuning of control equipment and imitating fault situations to see The above heat and mass transfer components are usually
how the BEMS would cope [20]; described by fundamental engineering principles rst-principle
Simulation assisted control: i.e. to execute a simulation model
(encapsulated within the BEMS) as part of the control task in order 1
In data-driven models the input and the output variables are known and
to evaluate several possible control scenarios and make a choice in measured, and the objective is to determine the mathematical description and to
terms of some relevant criteria [20]. estimate the system parameters.
 M. Trka, J.L.M. Hensen / Automation in Construction 19 (2010) 9399 95

models (if equations are derived from fundamental principles but Pure conceptual system modeling approach represents the case
require some empirical input these models are also referred to as where only room processes are considered, while all other processes
quasi-rst-principle models [27], e.g. most of the component models in primary and secondary systems are idealized, with a possibility to
in [28] and [29]), or by empirically obtained equations, i.e. by using impose a capacity limitation upon them. An example application is to
regression analysis of design data published by a manufacturer, or by use the predicted room cooling/heating peak loads to determine the
simply specifying look-up tables. The former modeling approach is required HVAC system size. Most state-of-the-art BPS tools can be
usually used for secondary system component descriptions, while for used to model systems using this approach. Some, like ESP-r,
primary system components, due to their complexity, the latter introduce certain complexity by modeling conceptual systems
approaches are more often used, but exceptions exist [29]. thermal zone interactions through control algorithms. Thus, even
though the pure conceptual system model is used, system processes
3.2. Modeling approaches for HVAC control are not completely idealized. Their interaction with the building is
more realistically modeled since their characteristics can be included
HVAC controllers can be divided into two categories as follows. in terms of aspects such as heat injection/extraction point, ux limit
Local controllers are low-level controllers that allow HVAC systems values, response time, and convective/radiant split. In [32] the authors
to operate properly and to provide adequate services. Local controllers state that this method of system simulation is often misunderstood
can be further subdivided into two groups [30]: and under-rated.
Sequencing controllers dene the order and conditions associated System-based modeling approach represents the case with pre-
with switching equipment ON and OFF. Typical sequencing controllers congured common system types, such as variable air volume system
in HVAC systems are chiller sequencing controller, cooling tower and constant-volume variable-temperature system. This modeling
sequencing controller, pump sequencing controller, fan sequencing approach is implemented in DOE-2, eQUEST, Building Energy
controller, etc. Analyzer, BLAST, DesignBuilder, HAP, etc. The user has exibility in
Process controllers adjust the control variables to meet the specifying capacities, system ow rates, efciencies and off-design
required set point in spite of disturbances and considering the system system component characteristics, but is restricted to the system
dynamic characteristics. The typical process controllers used in the congurations and control strategies that are pre-dened in the tool.
HVAC eld are P, PI, PID, ON/OFF, step controller, etc. Component-based system modeling approach represents the case
Supervisory controllers are high level controllers that allow where a system is specied by (a) network(s) of interconnected
complete consideration of the system level characteristics and component models. This approach is more exible in terms of possible
interactions among all components and their associated variables. system congurations and control strategies compared to the
For example, a supervisory controller sets operation modes and sets previous approach.
points for local controllers. Component-based multi-domain system modeling approach repre-
From a modeling point of view, controllers are represented by sents the case where component representation is further partitioned
equations that must be satised in every simulation step. The into multiple interrelated balance concepts, e.g. uid ow, heat and
controllers direct the interaction between building and system as electrical power balance concepts. Each balance concept is then solved
well as interactions between components within the system. simultaneously for the whole system. Thus, the overall system of
In reality, the closed-loop local-process control includes a sensor equations is broken into smaller systems of equations. Different
that samples a real-world (measurable) variable. The controller, based solvers, well adapted for the equation types in question, can be used
on the set point value and measured value, and according to the for different problem partitions. It is also possible to easily remove
controller-specic control algorithm, calculates the control signal that partitions as a function of the problem at hand.
feeds the real-world actuator. However, in the simulation tool the As an addition to the above four categories dened by Hensen [31],
user can address variables that cannot be sensed or actuated in reality, this paper lists a fth category: the equation-based system modeling
as well as apply control algorithms that do not exist in reality. For approach. This modeling approach represents the case where a
example, a modeler can directly actuate the heat ux in a model system is represented by a basic modeling unit that is physically
where in reality this could only be done indirectly by changing a smaller than a component and that is in the form of an equation or a
valve/damper position. low-level physical process model. It has evolved from the need to
Furthermore, due to the accessibility of many variables not directly improve the BPS tools that had been based on technology available in
known in the real world, such as the zone cooling/heating load, in the early seventies [33]. Equation-based simulation tools are [34,35]:
simulation the concept of ideal (local process) control becomes
feasible. An ideal local-process controller means that the actuated inputoutput free (all models are declarative in nature) as opposed
variable will be adjusted to satisfy the set point requirements for the to the traditional procedural,
controlled variable, without specifying the explicit control algorithm modular (supported by object-oriented programming languages),
and by numerically inverting the (forward) simulation models (from hierarchical (enable incremental modeling, i.e. models can consist of
the required output calculate the input needed to satisfy this). sub-models in multiple levels), which helps in managing the
Possibilities to simulate different (advanced) controllers in state- complexity of large systems,
of-the-art BPS tools are limited. Some tools offer pre-dened control universal (model denition in a generic form, e.g. using NMF and
strategies (system-based simulation tools), some offer exibility in Modelica).
specifying only supervisory controllers (EnergyPlus) and some even They provide separation of modeling the physics from numerical
in specifying local controllers (TRNSYS, ESP-r). The domain-indepen- solution algorithms.
dent environments, such as MATLAB and Dymola, are efcient tools They provide faster developments of simulation models, etc.
for designing and testing of controllers in a simulation setting, but lack
the models of all other physical phenomena in buildings. Examples of equation-based tools are:
SPARK (Simulation Problem Analysis and Research Kernel),
3.3. Modeling approaches for HVAC systems formerly EKS/US and SPANK, is developed by the Lawrence Berkeley
laboratory [36]. The primary goal of the EKS/US was improvement of
Hensen [31] denes four categories of HVAC system representa- the modeling and solution processes which resulted in SPARK. It is an
tion in BPS tools, ranging from purely conceptual towards more object-oriented simulation environment, of which the fundamental
explicit, as follows. object is an equation.
96 M. Trka, J.L.M. Hensen / Automation in Construction 19 (2010) 9399

EKS (Energy Kernel System) was researched in the UK [37]. The associated with those entities. The lack of the agreement upon the
objective of the EKS/UK was to place tool development on a task- above issue has resulted in a limited presence of object-oriented
sharing basis in order to ensure integrity and extensibility of future programming in the domain of BPS.
systems. The primary goal of EKS/UK, i.e. to improve the tool
development process, was later researched via primitive part 3.4. Solution techniques for HVAC system simulation models
modeling in the ESP-r simulation environment [34].
NMF (neutral model format) was designed to bring the power of The differences in solution techniques employed by different
differential algebraic equation (DAE)-based modeling to the building simulation tools are based on the distinction in the way the integrator
simulation community and yet be compatible with major BPS tools is employed [48].
such as TRNSYS, IDA and SPARK. The basic objective of NMF is to Simultaneous modular solution, where the various components are
provide a common format of model expression for a number of integrated simultaneously by a common integrator. In general, the
existing and emerging simulation tools, e.g. TRNSYS, HVACSIM+, IDA tools that employ this solution technique use model equations that
and SPARK [38]. are based on rst principles [48]. Each component is described with
IDA is one of a few equation-based efforts that have been pursued time-averaged discretized heat and/or mass conservation statements,
beyond the stage of prototyping [38]. The NMF initiative continues to which are combined to form a system matrix, and which are solved
live with IDA, since most of the IDA models are written in NMF, simultaneously for each simulation time step using either an implicit,
besides a few written in Modelica [33]. explicit or mixed numerical scheme.
Modelica [39] is an ambitious modeling language that has shown Independent modular solution, where each module is provided with
potential to bring order to the fragmented world of DAE-based individual integrator routines. In general, the tools that employ this
simulation. It draws on the collective experience of a large number of solution technique use model equations that can be based on rst
rst generation languages. Since the rst Modelica based tool, principles but can also be empirical input/output correlations [48].
Dymola, appeared in 1999, several large industries such as Toyota, The component's modules encapsulate all information relevant to the
Ford, United Technologies, Caterpillar, ABB, Alstom, TetraPak, etc. component's simulation model setting and execution. Each compo-
have adopted it [38]. Efforts to develop building and HVAC system nent is executed sequentially and the system solver iterates until a
simulation models resulted in various Modelica libraries, such as convergent solution has been found.
ATPlus [40], UTRC Modelica library [41] and Building Informatics Equation-based solution using formula manipulation, which has
Environment [42]. emerged in recent years with the developments of equation-based
SimScape [43] is a new development by MathWorks that extends tools. Models composed with these tools cannot be executed directly.
Simulink with tools for modeling and simulating multi-domain To be executed, a model needs to be transferred into a programming
physical systems, such as those with mechanical, hydraulic, and language that can be compiled. Tools employ different techniques to
electrical components. SimScape can be used for a variety of reduce the dimensionality of the linear and non-linear systems
automotive, aerospace, defense, and industrial-equipment applica- dened in the model in order to increase the execution efciency of
tions. Together with other MatLab toolboxes, SimScape allows the compiled program. For example, in SPARK [49], mathematical
modeling of complex interactions in multi-domain physical systems. graph algorithms are used for problem decomposition and reduction,
There appears to be no evidence yet of using SimScape in BPS. greatly reducing solution time for wide classes of problems [50].
Based on object-oriented programming approach, the above
projects were aiming to introduce modern concepts from computer 4. Integration of building and HVAC system models
science and software engineering in the BPS eld to make available to
developers basic software modules and supporting framework that The integration of building and HVAC system models is accom-
could be used to construct new BPS software [44]. But, as Sahlin et al. plished at different levels. The models can be (i) sequentially coupled
[33] notes that nothing much has happened in recent years to change (many duct/pipe sizing tools, BLAST, DOE-2, etc.) without system
the direction of fundamental reasoning. The authors also state model feedback to the building model or (ii) fully integrated (ESP-r,
several factors that contributed to this apparent lack of progress, as EnergyPlus, IDA ICE, TRNSYS, etc.) allowing the system deciencies
follows: to be taken into account when calculating the building thermal
conditions. Levels of detail of both building and system models can
Some exploratory projects did not deliver as expected.
vary from simple (e.g. the bin method and pure conceptual
Leading research groups have reverted back to existing solutions
representation for system model) to complex (numerical model of
and organic evolution.
physical processes).
Multi-domain simulation is being attempted by coupling of existing
domain specic simulators (co-simulation).
5. Issues in selecting HVAC modeling approach
Driven by product model research, attention has shifted from new
tool development to improved integration of existing modeling and
Different HVAC system modeling approaches demand different
simulation tools into the design process.
levels of user skills, different modeling resolutions and details, and
Sahlin [45] states that the primary cause of the lack of success is different levels of user customization capability. Higher explicitness in
unwillingness by BPS developers to learn other engineering elds. It system representation requires more knowledge about the system
seems that the equation-based tool development has not shifted away because of the increasing number of model parameters for system
attention from existing tools. Due to the difculty in obtaining funding specication, often difcult to obtain as they are not supplied by
for work other than incremental improvements of BPS tools [46], manufacturers. In addition, for higher explicitness in system repre-
many research teams continued to improve the integration of sentation the computational requirements become more intensive
traditional simulation tools into the design process. and the analysis of the results more complicated.
The major motive for the adoption of object-oriented software Most design analyses do not require detailed system modeling and
engineering approaches has been its support for modularity in simulation as the energy consumption can be estimated by using
modeling. However, a model for the simulation of a complex system, simpler modeling approaches. The conceptual system representation
such as a building, in object-oriented languages is not trivial [47]. One shows its advantages (lower required user expertise, lesser input
of the main questions is to what the objects should correspond. data, less intense computations, easier results analysis, etc.) when
Should they correspond to real-world entities, or to the equations only load predictions are considered, and/or when energy saving
 M. Trka, J.L.M. Hensen / Automation in Construction 19 (2010) 9399 97

options are investigated. However, for comparing HVAC system alter- model has a value only above certain modeling complexity for some
natives and evaluating different control strategies [18,51] detailed other simulation objectives, as illustrated in Fig. 1.
HVAC system models are required. In the system-based modeling The potential techniques that can be used to ease selection of
approach, the speed of system alternatives evaluation is much higher modeling complexity for a particular simulation objective can be
than in the component based modeling approach, but the investiga- described in following paragraphs (the use of the techniques has been
tion of innovative technologies is limited. reported previously [53,54]).
Matching the applicabilities of system modeling approaches with Denition of the minimum required modeling complexity can be
the design questions at hand, the user can benet from both ease of accomplished by using the checklist rationale from [55] represented
the former categories and exibility of the latter ones. However, in Fig. 2.
building a right model for a simulation task at hand is still more an art The stakeholder denes the simulation objectives and the relevant
than an engineering discipline. This issue is highly relevant when performance indicators. Based on this information, the checklist
there is no (measured) data which can be used for direct model framework can be used to identify the entities and variables to be used
accuracy evaluation. Thus, in this case the model adequacy for the in the simulation and thus estimate the initial modeling complexity.
particular simulation objective needs to be evaluated differently. The initial modeling complexity should be the lowest possible
Building a right system model for a specic purpose is to require complexity that satises the simulation objectives in terms of
that the modeling validity and data validity match as far as possible performance indicators. The quantication of validity of the initial/
the required validity [52]. The required validity is assessed only minimum required modeling complexity is achieved by specifying a
against those aspects of the real world that are of relevance for range for error tolerance, as a model deviation of the real world.
successful accomplishment of simulation objectives, represented by The error in a veried model is the sum of: (i) abstraction error, (ii)
performance parameters. input data error, and (iii) numerical errors. Here, only the former two
Model complexity can be expressed in terms of scope (dened by a are discussed while it is assumed that by decreasing the discretization
number of components in the model) and resolution (dened by a step the numerical errors can be controlled. The rst error is due to the
number of states per component in the model) of the model and modeling abstractions, i.e. using an incomplete model of a physical
interactions among components in the model. Abstraction is a general system, and the second is due to uncertainties in the parameters
process and includes various simplication approaches with regards themselves. Sometimes the distinction between the two is not clear.
to system boundaries considered, number of modeled physical Parameter uncertainty can be quantied and therefore the corre-
phenomena, the resolution of modeling of each considered phenom- sponding uncertainty of the model output as well. This uncertainty of
enon, etc. in increase in model complexity increases the cost of using the output is known as predictive uncertainty.
the model. Thus, the model should be of the lowest complexity while The modeling uncertainty is not easily quantiable and therefore
preserving its validity for the intended simulation objectives. The its inuence can be considered as a modeling bias. As illustrated in
required lowest model complexity depends on the simulation Fig. 3, with the increase of modeling complexity the predictive
objective. Also, increasing the complexity, for different simulation uncertainty rises as there are more parameters to consider. On the
objectives, has different implications for the value of the model to the other hand, the models approach reality and the bias decreases. The
user, as schematically shown in Fig. 1. For different simulation curve that denes predictive uncertainty depends on how much of
objectives the model cost exceeds the model value to the user at
different model complexities. For some objectives the cost of the
model will exceed its value even when the modeling complexity is
low, and for some, the simulation objective can justify the use of more
complex models. Moreover, the rate of change in the model value can
be different for different simulation objectives at different complex-
ities. On the one hand, a simple model can have a high value at low
modeling complexity for some simulation objectives; this value might
not be increased by increasing the complexity. On the other hand, a

Fig. 2. Schematic representation of a checklist rationale [55]: 1. There must be a total

tracking of items in the requirements to the conceptual model. 2. There should be a
specic simulation element for every item (parameter, attribute, entity, task, state,
etc.). 3. As far as possible, there should be real world counterparts for every
simulation element. 4. The simulation elements should correspond to standard and
widely accepted decomposition paradigms to facilitate acceptance of the conceptual
model and effective interaction (including reuse of algorithms and other simulation
components) with other simulation endeavors. 5. Simulation elements required for
computational considerations that fail to meet any of the previously stated items should
be used only when absolutely essential. 6. There should be no extraneous simulation
Fig. 1. Cost and value to the user vs. complexity. elements.
98 M. Trka, J.L.M. Hensen / Automation in Construction 19 (2010) 9399

Although some design (operational) problems immediately exclude

the use of some tools, the user is still free to choose between a large
number of available tools for a particular case. So far, there is no
comprehensive guideline on how to make this choice relative to the
required accuracy of the predictions based on the model. The above
discussion regarding selection of the most appropriate modeling
approach could be a rst step towards such a guideline. More
research is needed in this area.
Simulation tools have been seen as promising tools for establishing
the baseline (or baseband) performance prediction which can be
used during building operation to monitor the performance and/or
to detect and identify abnormalities in the system behavior. How-
ever, the research is still in its early phases.

References
[1] DOE, Building Energy Software Tools Directory, U.S. Department of energy, 2009,
Available at: http://www.eere.energy.gov/ [Accessed: August, 2009].
[2] D.B. Crawley, J.W. Hand, M. Kummert, B.T. Grifth, Contrasting the capabilities of
Fig. 3. Model uncertainty vs. complexity. building energy performance simulation programs, in: Proceedings of 9th Interna-
tional IBPSA Conference, International Building Performance Simulation Association,
Montral, Canada, 2005, pp. 231238.
[3] J. Hensen, Simulation Tools for Energy Efcient Building Design, 2009, Available at:
system knowledge is available. If the modeled system is well known, http://www.bwk.tue.nl/bps/hensen/courseware [Accessed: August, 2009].
the input parameters are less uncertain and the rate of increasing [4] J.A. Clarke, Energy Simulation in Building Design, Second edition, Butteworth-
predictive uncertainty with model complexity is low. The modeling Heinemann, UK, 2001.
[5] WBDG, Energy Analysis Tools, Whole Building Design Guide, 2009, Available at:
complexity for which the model error has its minimum will closely be
http://wbdg.org/ [Accessed: August, 2009].
related to the available system knowledge. [6] L. Mei, D. Ineld, U. Eicker, V. Fux, Thermal modelling of a building with an
There is a certain modeling complexity after which the predictive integrated ventilated pv faade, Energy and Buildings 35 (6) (2003) 605617.
[7] Y. Jeong, F. Haghighat, Modelling of a hybrid-ventilated building using ESP-r,
uncertainty will be higher than the modeling bias. There is no sense in
International Journal of Ventilation 1 (2) (2002) 127139.
going beyond this complexity, as the overall error in the model [8] J. Jokisalo, J. Kurnitski, M. Korpi, T. Kalamees, J. Vinha, Building leakage, inltration
uncertainty will not decrease. Hence, whether the required validity and energy performance analyses for Finnish detached houses, Building and
will be met by the model depends not only on the system modeling Environment 44 (2) (2009) 377387.
[9] M. Stadler, R. Firestone, D. Curtil, C. Marnay, On-site generation simulation with
complexity, but also on the available system knowledge. EnergyPlus for commercial buildings, ACEEE Summer Study on Energy Efciency
in Buildings, PANEL 3 Commercial Buildings: Technologies, Design, Performance
Analysis, & Industrial Trends, Pacic Grove, California, 2006.
[10] M. Wetter, GenOpt a generic optimization program, in: Proceedings of 7th
6. Conclusion International IBPSA Conference, International Building Performance Simulation
Association, Rio de Jeneiro, Brazil, 2001, pp. 601608.
This paper presents a review of available tools for HVAC system [11] A. Yahiaoui, J.L.M. Hensen, L. Soethout, Integration of control and building
performance simulation software by run-time coupling, in: Proceedings of 8th
design and analysis, modeling approaches and simulation techniques. International IBPSA Conference, International Building Performance Simulation
The numerous available tools range from simple spread-sheet tools to Association, Eindhoven, The Netherlands, 2003, pp. 14351442.
more advanced simulation tools. Even though they cover a wide range [12] CSTB, Type 155 A New TRNSYS Type for Coupling TRNSYS and Matlab, 2003,
online presentation available at: http://software.cstb.fr/articles/18.ppt [Accessed:
of design and operational problems, there is still an enormous amount
December, 2003].
of work to be done in this area. We nish by identifying some [13] D. Jreijiry, A. Husaunndee, C. Inard, J.G. Villenave, Control of ventilation in
requirements for further research and development: buildings using SIMBAD building and HVAC toolbox, in: Proceedings of 8th
International IBPSA Conference, International Building Performance Simulation
Association, Eindhoven, The Netherlands, 2003, pp. 661667.
Buildings are complex systems of which the real performance [14] P. Riederer, Matlab/Simulink for building and HVAC simulation state of the art,
usually deviates from the performance predicted in the design stage. in: Proceedings of 9th International IBPSA Conference, International Building
Recent studies (e.g. [56] show that the difference between the Performance Simulation Association, Montral, Canada, 2005, pp. 10191026.
[15] IEA-ANNEX40, Commissioning Tools for Improved Energy Performance, Tech.
predicted and real energy consumption can be up to 40%). For crude Rep., Program of the International Energy Agency, 2004, Available at: http://www.
analysis, including the relative comparison of the design alter- commissioning-HVAC.org [Accessed: August, 2005].
natives, this may not be a problem. However, to be able to correctly [16] J. Hyvikinen, Real time simulation of HVAC systems for building optimization, in:
Fault Detection and Diagnosis Technical Papers of IEA Annex 25, Finland, Technical
base design decisions on predictions, there is a need to understand Research Centre of Finland, VTT Building Technology, 1996.
where the above discrepancies come from and to include the [17] P. Haves, T. Salsbury, C. Claridge, M. Liu, Use of whole building simulation in on-
uncertainties in the system model. line performance assessment: modeling and implementation issues, in: Proceed-
ings of 7th International IBPSA Conference, International Building Performance
In general, the ongoing research that deals with uncertainty and Simulation Association, Rio de Jeneiro, Brazil, 2001, pp. 335342.
sensitivity analysis in building performance predictions does not [18] P. Haves, L.K. Norford, M. DeSimone, A standard simulation test bed for the evaluation
take into account the modeling bias, but only the predictive of control algorithms and strategies, ASHRAE Transactions 104 (1) (1998).
[19] E.H. Mathews, C.P. Botha, Improved thermal building management with the aid of
uncertainty. In order to assess the validity (delity) of the model,
integrated dynamic HVAC simulation, Building and Environment 38 (2003)
both the modeling bias and predictive uncertainty need to be taken 14231429.
into account. [20] J.A. Clarke, J. Cockroft, S. Conner, J.W. Hand, N.J. Kelly, R. Moore, T. O'Brien, P. Strachan,
Simulation-assisted control in building energy management systems, Energy and
The capability of most of the tools is limited to a set of predened
Buildings 34 (2002) 933940.
system congurations. To successfully continue the development of [21] M. Zheng, S. Pan, Application of Various HVAC Simulation Programs and Visual
BPS tools to accelerate innovation of building technologies and thus Tools to commissioning, Tech. Rep., Texas A&M University, 2007, Available at:
help in mitigating climate change, the focus should be on supporting http://hdl.handle.net/1969.1/6212 [Accessed: June, 2009].
[22] T. Reddy, Literature review on calibration of building energy simulation programs:
exible modeling environments that allow to analyze building uses, problems, procedures, uncertainty, and tools, ASHRAE Transactions 112 (2)
systems which are not yet covered in current BPS tools. (2006) 226240.
 M. Trka, J.L.M. Hensen / Automation in Construction 19 (2010) 9399 99

[23] T. Reddy, I. Maor, C. Panjapornpon, Calibrating detailed building energy simulation [40] F. Felgner, F.S. Agustina, R.C. Bohigas, R. Merz, L. Litz, Simulation of thermal
programs with measured data part ii: general methodology (RP-1051), building behaviour in Modelica, in: Proceedings of 2nd International Modelica
AHVAC&R Research Journal 13 (2) (2007) 221241. conference, Oberpfaffenhofen, Germany, 2002, pp. 147154.
[24] T. Reddy, I. Maor, C. Panjapornpon, Calibrating detailed building energy simulation [41] M. Wetter, Multizone building model for thermal building simulation in Modelica,
programs with measured data part II: application to three case study ofce in: Proceedings of Modelica Conference at arsenal research, Vienna, Austria, 2006,
buildings (RP-1051), AHVAC&R Research Journal 13 (2) (2007) 243265. pp. 517526.
[25] B.P. Zeigler, Theory of Modeling and Simulation, John Wiley & Sons, Canada, 1976. [42] M. Wetter, Building Control Virtual Testbed Bcvtb, Lawrence Berkeley National
[26] ASHRAE, ASHRAE Handbook, Fundamentals, American Society for Heating, Laboratory, 2009, Available at: https://gaia.lbl.gov/virBui [Accessed: August, 2009].
Refrigerating and Air-conditioning Engineers, 2009. [43] MathWorks, Simscape 3.2, 2009, Available at: http://www.mathworks.com/
[27] J. Wright, Building Performance Simulation for Design and Operation; Chapter: products/simscape/ [Accessed: August, 2009].
HVAC Performance Prediction, Taylor and Francis, in: J.L.M. Hensen, R. Lamberts [44] J.A.Clarke, D.F.MacRandal, The Energy Kernel System: form and content, in:
(Eds.), in press. Proceedings of 3rd International IBPSA Conference, International Building
[28] HVAC-Toolkit2, ASHRAE toolkit for secondary HVAC system energy calculation Performance Simulation Association, Adelaide, Australia, 1993, pp. 307315.
629-rp, Tech. Rep., American Society of Heating, Refrigerating and Air-condition- [45] P. Sahlin, The Methods of 2020 for Building Envelope and HVAC Systems Simu-
ing Engineers, ASHRAE, 1993. lation Will the Present Tools Survive?, in: CIBSE conference, Dublin, Ireland,
[29] HVAC-Toolkit1, ASHRAE toolkit for primary HVAC system energy calculation 665-rp, 2000.
Tech. Rep., American Society of Heating, Refrigerating and Air-conditioning [46] J. Spitler, School of Mechanical Engineering, Oklahoma State University, Stillwater,
Engineers, ASHRAE, 1994. OK, personal communication (2008).
[30] S. Wang, Z. Ma, Supervisory and optimal control of building HVAC systems: A [47] A.J. Wright, D. Bloomeld, T.J. Wiltshire, Building simulation and building
review, HVAC & R Research Journal 14 (1) (2008) 332. representation: overview of current developments, Building Services Engineering
[31] J.L.M. Hensen, Application of modelling and simulation to HVAC systems, in: Research and Technology 13 (1) (1992) 111.
Proceedings of 30th Int. Conf. MOSIS '96, Krnov, Technical University of Ostrava, [48] M. Hillestad, T. Hertzberg, Convergence and stability of the sequential modular
CZ, 1996. approach to dynamic process simulation, Computers & Chemical Engineering 12
[32] NRC, Dening the Methodology for the Next-Generation HOT2000 Simulator, Tech. (5) (1988) 407414.
Rep., National Resources Canada, 2004, Available at: http://www.buildingsgroup. [49] E.F. Sowell, M.A. Moshier, P. Haves, D. Curtil, Graph-theoretic methods in simulation
nrcan.gc.ca/docs/pdf/nal_report_survey.pdf [Accessed: April, 2005]. using SPARK, in: Proceedings of High Performance Computing Symposium of the
[33] P.Sahlin, L.Eriksson, P.Grozman, H.Johnsson, A.Shapovalov, M.Vuolle, Will equa- Advanced Simulation Technologies Conference, Society for Modeling Simulation
tion-based building simulation make it? experiences from the introduction of International, Arlington, VA, 2004, pp. 1822.
IDA indoor climate and energy, in: Proceedings of 8th International IBPSA [50] E.F. Sowell, P. Haves, Numerical performance of the SPARK graphtheoretic simulation
Conference, International Building Performance Simulation Association, Eindho- program, in: Proceedings of 6th International IBPSA Conference, International
ven, The Netherlands, 2003, pp. 11471154. Building Performance Simulation Association, Kyoto, Japan, 1999.
[34] T.T. Chow, Air-conditioning plant component taxonomy by primitive parts, Ph.D. [51] D.E. Miller, The impact of HVAC process dynamics on energy use, ASHRAE
thesis, University of Strathclyde, Glasgow, Scotland (1995). Transactions 86 (2) (1980) 535553.
[35] M. Wetter, C. Haugstetter, Modelica versus TRNSYS a comparison between an [52] DMSO, Fidelity, Validation, Verication and Accreditation; Recommended
equation-based and a procedural modeling language for building energy simulation, Practices Guide, Tech. Rep., Defense Modeling and Simulation Ofce, USA, 2000.
in: Proceedings of the SimBuild, 2nd National Conference of IBPSA-USA, International [53] M.Trka-(Radoevi), J.Hensen, Towards a guideline for selecting the appropriate
Building Performance Simulation Association, USA chapter, Cambridge, MA, USA, abstraction level for building systems simulation, in: Proceedings of the 17th Int.
2006. Air-conditioning and Ventilation Conference, Society of Environmental Engineer-
[36] LBNL, SPARK 2.0 Reference Manual Simulation Problem Analysis and Research ing, Prague, 2006, p. 6 pages on CD.
Kernel, Tech. Rep., Lawrence Berkeley National Laboratory, 2003. [54] E. Djunaedy, J.L.M. Hensen, M. Loomans, Selecting an appropriate tool for airow
[37] J.A.Clarke, D.Tang, K.James, D.F.MacRandal, Energy Kernel System, Tech. Rep. GR/F/ simulation in buildings, Building Services Engineering Research and Technology
07880, Engineering and Physical Science Research Council, Swindon, 1992. 25 (3) (2004) 269278.
[38] P.Sahlin, L.Eriksson, P.Grozman, H.Johnsson, A.Shapovalov, M.Vuolle, Whole- [55] D.K. Pace, Ideas about simulation conceptual model development, Johns Hopkins
building simulation with symbolic DAE equations and general purpose solvers, Apl Technical Digest 21 (3) (2000) 327336.
Building and Environment 39 (2004) 949958. [56] B. Elkhuizen, E. Rooijakkers, Visie op ontwikkelingen gebouwbeheersystemen,
[39] M.M. Tiller, Introduction to Physical Modeling with Modelica, Kluwer Academic Verwarming Ventilatie Plus (2008) 336339.
Publisher, Norwel, Mass, 2001.