Building and Environment 44 (2009) 651656

Contents lists available at ScienceDirect

Building and Environment

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

Optimization of ventilation system design and operation in ofce environment,

Part I: Methodology
Liang Zhou*, Fariborz Haghighat
Department of Building, Civil and Environmental Engineering, Concordia University, 1455 de Maisonneuve Blvd. W, Montreal, Quebec H3G 1M8, Canada

a r t i c l e i n f o

a b s t r a c t

Article history:
Received 30 January 2008
Received in revised form 6 April 2008
Accepted 7 May 2008

Ventilation principles that integrate exible and responsive elements have grown in popularity in ofce
buildings due to increasing concerns about the impact of indoor environment quality on ofce workers
well-being and productivity, as well as concerns over the rising energy costs for space heating and
cooling in the ofce building sector. Such advanced elements as underoor air distribution (UFAD), passive
swirl diffusers, and demand controlled ventilation have posed challenges to system design and operation. This paper is concerned with the development and implementation of a practical and robust
optimization scheme, aiming to assist ofce building designers and operators to enhance thermal
comfort and indoor air quality (IAQ) without sacricing energy costs of ventilation. The objective function
was constructed in a way attempting to aggregate and weight indices (for thermal comfort, IAQ, and
ventilation energy usage assessment) into one indicator. The path taken was a simulation-based optimization approach by using computational uid dynamics (CFD) techniques in conjunction with genetic
algorithm (GA), with the integration of an articial neural network (ANN) for response surface approximation (RSA) and for speeding up tness evaluations inside GA loop.
2008 Elsevier Ltd. All rights reserved.

Keywords:
Ofce indoor environment
Ventilation system
CFD simulation
Articial neural network
Genetic algorithm

1. Introduction
Ventilation systems satisfy air quality and thermal comfort
requirements in ofce spaces by delivering conditioned air (return
air plus outdoor fresh air). Despite the fact that modern ventilation
systems strive to offer the capacity for subtle indoor climate control,
thermal conditions in ofce spaces are still far from satisfactory.
Every year, the international facility management association (IFMA)
announces survey data, identifying that predominant ofce occupants complaints are its too hot and too cold, simultaneously [1].
Triggered by the advent of the energy crisis in the early 1970s,
emphases have been put on the energy efciency of ofce buildings. Space conditioning contributes substantially to energy
consumption in ofce buildings. In 2004, approximately 60% of the
total energy consumed by Canadian commercial/institutional
buildings was used for space heating and cooling [2]. To reduce this
high energy cost, ofce building design has shifted towards high
levels of insulation and air tightness, and minimal ventilation rate
as well. However, this led to the deterioration in IAQ and raised the
problem of sick building syndrome (SBS) [3,4]. Previous research also
revealed that the risk of SBS and associated sick-leave rates were
strongly correlated with ventilation rates in ofce buildings [3,4].

* Corresponding author. Tel.: 1 514 848 2424x7257; fax: 1 514 848 7965.
E-mail address: grace_zhouliang@yahoo.com (L. Zhou).
0360-1323/$ see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.buildenv.2008.05.009

In summary, trade-offs always exist between the energy used for

ventilation and the benets of ventilation to occupants comfort
and health, therefore, further investigations into indoor environment quality in ventilated ofce spaces should orient to a holistic
evaluation of thermal comfort, IAQ and system energy efciency.
Also, the above statements demonstrate that plenty of room
remains for adapting ventilation system design and operation in
ofce spaces to achieve better IAQ and comfort and a broader effort
should be put on the promotion of more energy effective measures
during the design and operating stages of ventilated ofce buildings. In response to this, many building designers and researchers
have considered advanced ventilation systems that integrate exible and responsive elements such as underoor air distribution
(UFAD), passive swirl diffusers, and demand controlled ventilation.
Such systems can supply air directly to where it is required and
provide local control, which poses challenges to energy analysis due
to the non-uniform temperature distributions created within the
space. Under such circumstances, highly resolved methods are to be
used for performance evaluation and a systematic optimization
approach is to be devised to guide ventilation design and operation.
Although the accomplishments of previous studies on heating,
ventilation, and air conditioning (HVAC) system control optimization
have been signicant, very limited work has been done to optimize
the design congurations and operational states of ventilation
systems in ofce environment, by integrating thermal comfort and
air quality together with energy efciency into the objective

652

L. Zhou, F. Haghighat / Building and Environment 44 (2009) 651656

function. The reason is threefold. First of all, in pursuit of

improvement in indoor air quality (IAQ), thermal comfort, and
ventilation energy efciency in ofce spaces, it would be necessary
to acquire detailed information about the indoor airow, the
pollutant dispersion as well as the temperature variations resulting
from various types of ventilated systems. Therefore, it is necessary
to employ a highly resolved approach to the problem domain.
Second, the ventilation performance in a particular ofce is highly
dependent on a variety of geometric and thermal factors, such as
ventilation approaches (overhead mixing system or UFAD), air
supply/return terminals conguration, exterior and interior
thermal states, contaminant source location and emission rate,
ofce equipment and furniture conguration, supply air conditions,
etc. Consequently, a exible modeling method has to be used to test
a large design space. In addition, since the objective function of
such an optimization problem is a nonlinear mixed-integer one
with multi-optimum, care should be taken to select an appropriate
optimization algorithm tting into the context.
The current study was targeted at addressing such needs by
devising and developing an optimization approach that encompassed two essential components: the rst one was a high-resolution indoor airow and heat transfer investigation so as to
capture the distribution of assessment indices pertaining to
thermal comfort, IAQ, and energy use; the other one was the
integration of an economical optimization scheme. Such a simulation-based optimization approach was developed with the ultimate
goal of providing practical aid to conceptual ventilation design and
regulation. In addition, it should offer exibility to predict, evaluate,
and compare a wide range of objectives and constraints. This paper
(Part I) commences with a description of the numerical optimization scheme. Following this, issues such as the specication of
assessment criteria and prescription of objective function are discussed. Part II of the paper will focus on presenting the optimization results and demonstrating the application of such
a simulation-based optimization approach.
2. Simulation-based optimization approach in the current
study
2.1. Integration of CFD
Ever since Nielsens application of computational uid dynamics
(CFD) techniques to model room airow driven by a diffuser [5],
CFD techniques have been used routinely in research to predict
detailed room airow patterns, highly resolved temperature
distributions, and pollutant transportation indoors for more than
a quarter century. Even in design practice, CFD is a useful tool for
the verication and comparison of tentative building design alternatives. In contrast to physical measurements, the CFD method is
relatively inexpensive, is applicable to any existing or conceived
scenario, and can provide complete information. Consequently, the
optimization approach in this study was based on the use of CFD
techniques to evaluate various ventilation system design congurations and operation states, with the hope that the near-optimal
solution could be found thereafter. The platform for implementing
and demonstrating CFD simulations in this study was the Airpak
package from Fluent Inc. [6].
2.2. Genetic algorithm (GA) as optimization engine
The other important aspect of this simulation-based optimization work is the selection of appropriate optimization algorithms.
Generally, optimization algorithms can be sorted into two main
categories: the conventional gradient-based method and the
gradient-free direct search. Building phenomena are very often
nonlinear mixed-integer, which may lead to discontinuous outputs

and thus cause problems for gradient-based methods [7,8]. These

issues have discouraged the adoption of gradient-based optimization methods, such as the sequential quadratic programming
method and the HookeJeeves algorithm in building studies. Also, in
most cases, gradient-based methods are generally prone to nd
a local optimal, and the convergence speed and the value of nal
results are strongly dependent on the initial guess values [9].
In contrast, gradient-free methodsdalso referred to as global
optimization strategiesdturn out to be better suited for building
applications. GA, as one of the most common global optimization
methods, has gained momentum in the optimization of building
thermal system design [10], HVAC system control [9,11], and green
building design [12]. Previous studies have demonstrated that GA is
capable of dealing with discontinuous variables and multi-modal
problems, and is also tolerant of noisy objective functions [10,11]. It
can nd a sufciently acceptable solution (near-optimal solution)
using less computing time, in comparison to other algorithms such
as mixed-integer programming method [13], and thus, it can be
incorporated into online optimal control. In addition, since no
derivative information is needed during the search, GA was also
proven to perform well in conjunction with some non-differentiable response surface approximation (RSA) methods [8,14]. Most
importantly, GA is essentially stochastic, thus it has better chances
to explore the entire design space and reach the global optimum.
Accordingly, GA was chosen as the optimization engine in this
work. GA optimization search in the current study was implemented by programming in Matlab 7.1 [15]. The GA parameters
chosen in the current study are summarized in Table 1.
2.3. Articial neural network (ANN) for RSA
One major negative attribute of GA is the substantial amount of
tness evaluations involved throughout the search process. In the
current study, the value of objective function was calculated based
upon CFD estimates. The full-scale ofce CFD simulation (with the
RNG k3 model) performed in the current study typically took
1720 h of CPU time on a Pentium IV desktop computer with dualprocessor (3.4 GHz speed) before reaching convergence. Meanwhile, with the GA setup used (the population size was 50 and the
number of generation limit was 100), 5000 runs of the CFD program
could have been required for a GA search. Therefore, it was
prohibitively expensive to directly invoke CFD simulation inside the
optimization loop for tness evaluation.
One possible remedy to this problem is to establish a relatively
inexpensive intermediate model for RSA. Such a surrogate model,
once pre-trained and validated using the inputs and outputs obtained
from the high delity model (CFD simulation in this case), can
be thenused in the place of CFD inside GA loops to provide approximations of the objective indices and thus reduce the computational
cost.
Though other RSA strategies (e.g., statistical regression and
simple curve-t) exist, the ANN technique has been extensively
Table 1
GA parameters in the current study.
Population size
Variable representation
Scattered crossover rate
Uniform adapted-feasible mutation rate
Number of elite
Parents selection
Survivor selection
Stopping criteria

50a
Real value
0.8
0.2
2
Stochastic uniform
Fitness-based replacement
Reach 100 generationsb/exceed 50
stall generations

a
Population size of 50 was found to be sufcient in terms of maintaining population diversity and avoiding excessive computational cost.
b
After 100 generations, no improvement was found in the cases studied here.

L. Zhou, F. Haghighat / Building and Environment 44 (2009) 651656

used in building-related studies as a global RSA method. No

attempt is made here to cite all the previous studies. The word
global denotes that such an approach can represent the response
of the system over the entire design space. ANN technique was
employed in the current study to provide RSAs of the objective
indices in response to the variations in the input variables. In order
to facilitate the automated optimization task, inputoutput data
sets extracted from CFD simulations were used to establish the ANN
model. The platform for implementing and demonstrating the
ANN-based GA optimization was Matlab 7.1 [15].
A multilayer feed-forward ANN was used in this study. Trade-off
always exists between the network accuracy and structural
complexity (and thus the computing/training time). The most
appropriate numbers of the hidden layers and the nodes in a hidden
layer are case specic and can only be determined by exploratory
calculations. The selection of input and output variables will be
addressed later on; the specication of these variables is left open
temporally. The feed-forward ANN model built contained 30 hidden
neurons (one hidden layer), 8 inputs, and 7 outputs (as shown
in Fig. 1), and it incorporated hyperbolic tangent sigmoid transfer
function in the initial/hidden layers and linear transfer function in
the output layer. LevenbergMarquardt (in conjunction with early
stopping technique) and Bayesian regularization [15] were selected
as the algorithms when implementing back-propagation training,
to improve the generalization ability of the network.
It is worth noting that such a simulation-based optimization
scheme was not devised from scratch in the present work. Mengistu [16] adopted a CFD simulation-based GA optimization
approach to improve the turbomachinery blade shape, with the
goal of maximizing the adiabatic efciency and pressure ratio of
the turbine. He also employed ANN techniques for RSA to cut down
the computational expenses required by tness evaluation. In
addition, Chow et al. [14] implemented an ANN-based GA approach
to nd the optimal set of parameters (the chilled water mass ow
rate and outlet temperature and condensed water temperature), to
minimize the fuel and electricity consumed by an absorption
chiller. These studies have conrmed that numerical optimization
combining ANN and GA could be an effective approach without
explicitly formulating the objective function and calculating the
derivatives to guide the optimization engine.
2.4. Framework of the current numerical optimization
In the simulation-based optimization approach developed, the
problem under consideration was divided into three sequential
steps:

653

1. Perform sufcient CFD simulations to predict the system

performance assessment indices in response to the variations
in the input data. This part of work can be considered the ANN
training/test data preparation by varying the inuential variables in the pre-validated template CFD model.
2. Train and test an ANN model for quick response approximation
by exploiting the input/output data pairs obtained in the
previous step. Such a model can be used in the place of CFD
simulation inside the GA search loops for approximating the
value of the objective function with less computational cost
and comparable accuracy.
3. Apply ANN-embedded GA to search for the near-optimal set of
controlled variables (parameters representing ventilation
system design congurations and operational states) in an
ofce space with particular geometrical conguration and
thermal conditions, with the goal of achieving satisfactory
comfort and IAQ without sacricing the energy efciency of the
ventilation systems.
The rst step was carried out independently beforehand, while
the ANN and GA were integrated into one numerical scheme to
facilitate automated optimization search. Consequently, the nearoptimal combination of controlled variables obtained in Step 3 was
supplied to the CFD program to perform a verication case study;
the simulation results can be used to illustrate the optimization
results preponderance and the accuracy of the low delity ANN
model and thus can justify the validity of the optimization
approach.
Fig. 2 briey outlines the implementation plan of the above
numerical optimization methodology. One may notice from this
ow chart that the topic of Latin hypercube sampling (LHS) is left
open temporally, which will be presented later on together with
ANN training/testing data preparation. Also, the validation treatment of CFD simulation is dropped here. It is straightforward that
the reliability of the optimization result is critically dependent on
the accuracy of the prediction of objective indices, that is, the
validation of CFD simulations against measurements is required to
lay the foundation for further optimization. The CFD simulations in
this study were pre-validated using experimental data from baseline cases with both UFAD system and ceiling mounted mixing
system (MS) [17,18]. Good agreement between the measured and
the predicted air velocities, temperatures, and contaminant
concentration proles was found, which justied the current choice
of turbulence models (the RNG k3 model for UFAD systems and the
indoor zero-equation model for MS) and the specication of
boundary conditions.

Fig. 1. Topology of the feed-forward ANN in this study.

654

L. Zhou, F. Haghighat / Building and Environment 44 (2009) 651656

Fig. 2. Implementation of the simulation-based optimization approach.

3. Assessment indices
Accurate and quantitative evaluation of indoor environment can be
only realized by considering reasonable independent variables and by
selecting appropriate criteria to assess the objectives of interest. Based
on a literature survey, the following indicators were selected in the
current study to cover all the issues under consideration. These indices
were calculated based on output quantities from CFD simulation and
were integrated into the objective function for optimization.
3.1. Predicted mean vote (PMV) for overall thermal comfort
assessment
The PMV model [19] is the most frequently used and bestunderstood model for quantitative thermal comfort analysis. PMV
reects the mean vote of a large group of occupants who are
exposed to a given combination of thermal parameters. PMV index
evaluates thermal environment in an indoor space by using
a thermal sensation range scale: 3 (cold), 2 (cool), 1 (slightly
cool), 0 (neutral), 1 (slightly warm), 2 (warm), 3 (hot) [19]. PMV
is dened as a function of six thermal variables related to the indoor
air conditions and human behaviors, including air temperature, air
humidity, air velocity, mean radiant temperature (Tmrt), clothing
insulation level, and human activity. The optimizer attempted to pull
the PMV value in adjacent to the occupant closer to neutral value.
3.2. Equivalent temperature for asymmetric thermal sensation
Despite the provision of acceptable whole-body comfort level,
UFAD systems are sometimes found to cause asymmetric thermal

sensation, if not designed properly. That is, even an individual

expresses global comfort, he/she may still experience discomfort at
a particular part of his/her body. An indicator named as equivalent
temperature (ET) was proposed to appraise thermally non-uniform
environments. ET was originally introduced to study the highly
non-uniform micro-climate encountered in automobiles [20]. ET
integrates the independent effects of air temperature, air velocity,
mean radiation, and solar load on heat loss/gain from occupant
body into a single physical quantity. It is dened as the temperature
of a uniform enclosure in which a human body would experience
the same rate of heat loss as in the actual thermally non-uniform
environment. It was concluded that if the variations of ET over the
entire body are controlled in the range of 2  C to 2  C, there is no
excessive thermal non-uniformity over the body. For each part of
the human body, ET can be calculated using the following equations
[21],

ET Ts  Icl  Q t

(1)
( C),

where Ts is the skin temperature

Icl is the thermal resistance of
the clothing (m2  C/W), and Qt is the local heat loss rate from skin
surface (W/m2), which was calculated by CFD in this study.
3.3. Head to ankle temperature difference and local air velocity
UFAD strategies supply cooler air into the lower region of the
ofce and produce distinct temperature stratication, however,
excessive temperature gradient along vertical direction may result
in complaints. According to ASHRAE standard 55-2004 [22], the
temperature difference between the head and the ankle level
should not exceed 3  C, which corresponds to 5% of dissatisfaction.

L. Zhou, F. Haghighat / Building and Environment 44 (2009) 651656

It is necessary to maintain this vertical temperature difference at or

below the recommended value to avoid cold feet and warm head
complaints. In addition, it was suggested by ASHRAE standard 552004 that the local air speed near an ofce worker should be
controlled at or below 0.25 m/s to avoid annoyance and distraction
[22].
The variant of ET on both sides of an individual occupant, the
head to ankle temperature difference (DTHeadToAnkle), and the local air
velocity near the occupant were integrated into the penalty term in
the objective function for optimization. When these three indices
exceeded the recommended values, the penalty term was set to
a relatively large positive number (since GA intended to minimize
the objective function here).
3.4. CO2-based ventilation effectiveness for IAQ assessment
In the baseline experimental cases [23], CO2 was injected into
the ofce space as tracer gas. The calculated CO2 concentration
distribution was further integrated into a dimensionless indexdventilation effectiveness (3v) for IAQ assessment. 3v was calculated
from the equation below, which is slightly different from the
original denition [24]. The CO2 concentration at breathing level
was used here instead of the average concentration throughout the
workstation.

3v

creturn  csupply
cbr  csupply

(2)

where creturn is the CO2 concentration in the return air (ppm), csupply
is the CO2 concentration in the supply air (ppm), and cbr is the CO2
concentration at breathing level near the occupant (ppm).
The transportation of CO2 in room air was predicted by the CFD
simulations; furthermore, the optimizer intended to search for
better system design and operation parameters, so as to decrease
the CO2 concentration at breathing level and improve 3v as much as
possible.

655

by relating the efciency to overall supply ow rate. Since the

pressure rise via the supply fan in this study was assumed to be at
a constant level, its effect on fan efciency is thus dropped here.
Accordingly, the overall efciency of the supply fan within the
allowable ow rate range (80160 L/s) was calculated using the
following correlation:

hfan

V_ air;total
0:25
400

(4)

This correlation was established based on the assumption that

when the total supply airow rate is at 160 L/s the resulting
combined efciency (motor plus fan) was at 65% while, when the
supply airow rate is reduced to 80 L/s, it corresponds to
a combined energy efciency at 45%. The intermediate values were
approximated using a linear interpolation between the two
extremes.
2. Cooling energy consumption
Cooling coils remove the sensible heat load produced within the
conditioned space and offset the humidity and temperature in the
outdoor fresh air. In light of the previous methods used for energy
usage prediction with alternate ventilation systems [26,27], it is
clear that the cooling energy requirement can be subdivided into
two portions:



_ air;total cp Treturn  Tsupply
Q cooling Q space Q vent m
_ fresh hout  hreturn
m

(3)

where Qcooling is the total cooling load (W), Qspace is the cooling
energy portion used to remove sensible heat load in the indoor
space (W), Qvent is the cooling energy portion used to condition the
_ air;total is the total mass
outdoor fresh air to return air states (W), m
ow rate of supply air (outdoor fresh air plus the re-circulated
portion of return air) (kg/s), cp is the specic heat of air (J/kg  C),
Treturn is the temperature of return air ( C), Tsupply is the tempera_ fresh is the mass ow rate of outdoor fresh
ture of supply air ( C), m
air (kg/s), and hout and hreturn are the specic enthalpy of the
outdoor air and return air (J/kg), respectively.
Accordingly, the total energy consumed by cooling can be
determined provided that the operating states of ventilation
system, the return air conditions, and the outdoor air states are
_ air;total , the
specied. Thermo-physical quantities such as Tsupply, m
mean moisture content indoors (dreturn), Treturn, and the moisture
content (dout) and the temperature (Tout) of outdoor air would be
treated as the independent variables here. One point need to be
explained is the relative humidity throughout the ofce space was
assumed to be maintained at a constant level around 40% and the
outdoor relative humidity was assumed to be at 70%. Based on the
relative humidity and air temperature, the moisture content and
enthalpy in return/outdoor air can be determined accordingly.
As a supplement to the above statement, the following
assumptions were imposed on the energy analysis for the ventilation system:

where DP is the pressure rise through the supply fan (Pa), V_ air;total is
the overall volumetric ow rate of supply air (L/s), and hfan is the fan
efciency. It is straightforward and understandable that the value
of V_ air;total can be directly extracted from the CFD model; whereas
the pressure rise and fan efciency remain to be determined.
Based on previous studies on UFAD system [25,26], the static
pressure rise via the central fan with the MS was assumed to be at
750 Pa in this study while, with the UFAD system, 562.5 Pa was the
static pressure rise assigned to the central fans (25% reduction). The
efciency of central supply fan in current study was characterized

1. The energy used by condenser water pump, chilled water

pump, and fans in cooling tower was excluded.
2. The cooling coil and chilled water heat exchanger were
presumed to work under perfect condition, that is, the heat
transfer efciency in each component was assumed to be 1.
3. The ducts were assumed to be perfectly tight and airow
through the ductwork was adiabatic; there was no temperature
gradient along the duct when there was no other energy
generation equipment; the return air temperature was equal to
the local air temperature in the vicinity of the return grille.

3.5. Energy demand for ventilation

CFD simulation has been widely used for the prediction of airow
in ofce environment; in contrast, issues with regards to heat
transfer and energy usage in ofce spaces have not been successfully
addressed using CFD techniques. Based on the survey of previous
experimental and numerical work on quantifying energy efciency
of ventilation systems, the current study divided the energy usage by
ventilation into two main partsdfan power input and cooling
energy consumption, both can be derived from the CFD estimations.
1. Fan energy consumption
Principally, we know that fan power input (W) can be determined using the following expression:

Efan

DP  V_ air;total
1000hfan

656

L. Zhou, F. Haghighat / Building and Environment 44 (2009) 651656

4. The effect of fan power on the supply air temperature was

neglected.
5. Air density and specic heat were set to constant values, at
1.205 kg/m3 and 1005 J/kg  C, respectively, as 20  C is taken as
the
reference
temperature
(available:
http://www.
engineeringtoolbox.com/air-properties-d_156.html).
4. Specication of objective function for optimization
PMV, CO2 concentration-based 3v, and energy input for cooling
load offset and fan driving were selected as the indices to measure
comfort, IAQ and energy usage, respectively. Also, the thermal and
airow states of room air had to comply with the constraints
imposed on DTHeadToAnkle, the variation of ET over the occupants
body and the local air velocity. Accordingly, the objective function
was prescribed by aggregating and weighting the above indices
into one equation:

"
JX Min wtc

!
XABSPMVi 
i

wfan

Efan
Efan max

PMVmax
!
wcooling

wiaq

3v max
3v

Ecooling
Ecooling max

#
PT

(6)

where X is the input vector, which consists of all the controlled

variables including the design parameters and operating settings of
ventilation systems (e.g., system type, supply air temperature, and
supply airow rate, distance from occupant to diffuser, distance
from return grille to contaminant source, etc.). Subscript i denotes
the number of occupants, as PMV is evaluated separately for individual occupant. wtc, wiaq, wfan, and wcooling represent the weighting
factors for thermal comfort index, IAQ index, fan power index, and
cooling power index, respectively.
It should be mentioned that in the present study GA was used in
order to minimize the cost function. Therefore, the rst term in the
cost function represents proximity of occupants thermal sensation
(represented by PMV index) to a neutral value; the second term
reects the removal of indoor contaminants; the third and fourth
terms are targeted at minimizing the energy consumed by space
cooling and supply fan; the last term of the cost function is
a penalty term, which accounts for the aforementioned constraints
imposed on the ow and thermal conditions. PMVmax, 3v max, Efan max,
and Ecooling max are the maximum values of corresponding objective
variables that can be observed from the training data, which were
used to scale the objective variables into a usable range ([0,1] in the
current case).
The magnitude of the weighting factors is to be specied by the
user according to personal preference and/or based on sensitivity
analysis. Different decision makers will not have the same goals in
system design. For example, when a user is more concerned about
the overall comfort level than IAQ, wtc would be set to a higher
value relative to wiaq; for maximum energy efciency design, wfan
and wcooling can be set to 1, whereas wtc and wiaq can be set to 0; etc.
Sensitivity analysis results by varying individual weights are presented in Part II of this paper.
5. Closing remarks
This part of the paper has set the technical stage for the optimization of ventilation system design and operation in ofce
environment. Issues regarding the numerical methods, performance evaluation, and objective function construction have been
covered. In Part II of this paper, results from LHS (for data base

construction), ANN training/testing, sensitivity analysis (for objective function specication), and GA optimization search will be
presented.
Acknowledgements
This work was nancially supported by the funding of the
Innovations & Solutions Directorate at Public Works and Government Services Canada (PWGSC). The authors would like to thank Dr.
E. Morofsky and Dr. S. Rastan of PWGSC for their support and for the
many inspiring suggestions.
References
[1] International Facility Management Association. Too cold and too hot rank at
top of workplace complaints; 2003. Available online: http://www.ifma.org/
tools/monthly_media/insider/2003/03_may_16.html.
[2] Natural Resources Canada. Energy use data handbook tables (Canada): Tables
2 and 3; 2005. Available online: http://oee.nrcan.gc.ca/corporate/statistics/
neud/dpa/handbook_totalsectors_ca.cfm?attr0.
[3] Pejtersen J, Brohus H, Hyldaard CE. The effect of renovating an ofce building
on occupants comfort and health. Proceedings of Indoor Air, Edinburgh, vol. 2;
1999. p. 1605.
[4] Wargocki P, Wyon DP, Sundell J, Clausen G, Fanger PO. The effects of outdoor
air supply rate in an ofce on perceived air quality, Sick Building Syndrome
(SBS) symptoms and productivity. Indoor Air 2000;10:22237.
[5] Nielsen PV. Flow in air conditioned rooms. Ph.D. Thesis. Technical University of
Denmark, Copenhagen; 1974.
[6] Airpak documentation. Version 2.1. Fluent Inc; 2002.
[7] Wetter M, Wright JA. Comparison of a generalized pattern search and
a genetic algorithm optimization method. Proceedings of Building Simulation,
Eindhoven, vol. 3; 2003. p. 14018.
[8] Lu L, Cai WJ, Xie LH, Li SJ, Soh YC. HVAC system optimizationdin building
section. Energy and Buildings 2005;37:1122.
[9] Wang SW, Jin XQ. Model-based optimal control of VAV air-conditioning
system using genetic algorithm. Building and Environment 2000;35:4718.
[10] Wright JA, Loosemore HA, Farmani R. Optimization of building thermal design
and control by multi-criterion genetic algorithm. Energy and Buildings
2002;43:95972.
[11] Huang W, Lam HN. Using genetic algorithms to optimize controller parameters
for HVAC systems. Energy and Buildings 1997;26:27782.
[12] Wang WM, Zmeureanu R, Rivard H. Applying multi-objective genetic algorithms in green building design optimization. Building and Environment
2005;40:151225.
[13] Sakamoto Y, Nagaiwa A, Kobayasi S, Shinozaki T. An optimization method of
district heating and cooling plant operation based on genetic algorithm.
ASHRAE Transactions 1999;105(2):10415.
[14] Chow TT, Zhang GQ, Lin Z, Song CL. Global optimization of absorption chiller
system by genetic algorithm and neural network. Energy and Buildings
2002;34:1039.
[15] Matlab documentation. Version 7.1. MathWorks Inc; 2006.
[16] Mengistu T. Aerodynamic design and optimization of turbomachinery blading.
Ph.D. Thesis. Concordia University, Montreal; 2005.
[17] Zhou L, Haghighat F. Simplied method for modelling swirl diffusers.
Proceedings of IAQVEC conference, Sendai, Japan; 2007.
[18] Zhou L, Wan MP, Chao CYH, Haghighat F. A comparison of an underoor air
distribution system and a mixing system: measurements and CFD simulations.
Proceedings of Healthy Buildings conference, Lisbon, Portugal, vol. 4; 2006. p.
299304.
[19] Fanger PO. Thermal comfort. Copenhagen: Danish Technical Press; 1970.
[20] Wyon DP. Standard procedures for assessing vehicle climate with a thermal
manikin. SAE Technical Paper Series 1989.
[21] Bauman FS, Inkarojrit V, Zhang H. Laboratory test of the ARGON personal airconditioning system (APACS); 2000. Available online: http://www.argonair.
com/pdf/Argon%20Test%20Report.pdf.
[22] ASHRAE. ANSI/ASHRAE standard 55-2004: thermal environmental conditions
for human occupancy, Atlanta; 2004.
[23] Cermak R. Design strategies for personalized ventilation. Ph.D. Thesis. Technical University of Denmark, Copenhagen; 2004.
[24] Awbi HB. Ventilation of buildings. London: Spon Press; 2003.
[25] Webster T, Bauman FS, Ring F. Supply fan energy use in pressurized underoor
air distribution systems. Berkeley: Center for the Built Environment, University of California; 2000.
[26] Bauman FS. Underoor air distribution (UFAD) design guide. Atlanta: ASHRAE;
2003.
[27] Xu HT, Niu JL. Numerical procedure for predicting annual energy consumption
of the under-oor air distribution system. Energy and Buildings 2006;38:
6417.