Energy and Buildings 59 (2013) 111122

Contents lists available at SciVerse ScienceDirect

Energy and Buildings

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

Energy analysis of chilled water system congurations using

simulation-based optimization
Muzaffar Ali a,b, , Vladimir Vukovic a , Mukhtar Hussain Sahir b , Giuliano Fontanella a
a
b

Energy Department, Austrian Institute of Technology, Gienggasse 2, 1210 Vienna, Austria

Mechanical Engineering Department, University of Engineering and Technology Taxila, Pakistan

a r t i c l e

i n f o

Article history:
Received 26 March 2012
Received in revised form 24 July 2012
Accepted 3 December 2012
Keywords:
Chilled water systems
Optimization
Modeling and simulation
Dymola/Modelica
HVAC system congurations
Energy performance

a b s t r a c t
The paper presents an incremental development of the methodology for chilled water system design
optimization. Initially, the system conguration parameters are varied with xed design conditions
to conrm the established best practice design criteria, followed by a comprehensive system design
optimization. The implemented simulation-based optimization approach couples the Dymola/Modelica
dynamic modeling and simulation program with GenOpt generic optimization program to nd optimal
system conguration. A dynamic system model is developed to vary and simulate different chilled water
system congurations. Optimization of the chilled water system is achieved at both design and conguration level using ve design variables. Two discrete variables are related to system conguration: number
of chillers and number of cooling towers and three continuous variables are related to system design:
building load demand, temperature difference across condenser, and cooling tower fan speed. The strategy of varying system design and conguration variables together for overall system optimization proved
to be the most energy efcient. For a xed building load demand, power consumption of the considered
system can be reduced 1743% by selecting an optimal system conguration.
2012 Elsevier B.V. All rights reserved.

1. Introduction
Energy performance of buildings strongly depends on selection
of heating, ventilation, and air-conditioning (HVAC) system conguration as HVAC systems account for approximately 3060% of
the total building energy consumption, depending on the building type [13]. As buildings are responsible for 2040% of the total
energy consumed [1], climatization systems consume 1020% of
total energy in the developed countries. It has been observed that
chilled water system consisting of primary HVAC system components, e.g. chillers, cooling towers, pumps accounts for most of
the electricity use in the HVAC system [3]. Thus, selecting optimal
chilled water system conguration can result in substantial power
savings. In practice, however, selection of optimal system conguration is a complex task in terms of effort and time as multifaceted
chilled water system congurations need to be analyzed. Therefore
HVAC practitioners select the system congurations and design

Corresponding author at: Energy Department, Austrian Institute of Technology,

Gienggasse 2, 1210 Vienna, Austria. Tel.: +43 050550 6484; fax: +43 050550 6613;
mobile: +923005316356.
E-mail addresses: muzaffar.ali@uettaxila.edu.pk, ali.muzaffar@ait.ac.at (M. Ali),
vladimir.vukovic@ait.ac.at (V. Vukovic), mukhtar.sahir@uettaxila.edu.pk
(M.H. Sahir), giuliano.fontanella@ait.ac.at (G. Fontanella).
0378-7788/$ see front matter 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.enbuild.2012.12.011

parameters at the initial design stage based on their experiencea

practice that could result in sub-optimal system operation.
In the current paper, simulation-based optimization methodology for optimal selection of chilled water system conguration
at initial design stage is presented. The methodology is applied in
three different strategies: (1) use xed system design conditions to
validate the methodology in comparison to the real system, (2) in
addition, vary the number of cooling towers according to the cooling tower ow turndown ratio to validate the best practice design
criteria, and (3) overall system optimization approach varying the
system design and conguration parameters.
2. Overview of modeling, simulation, and optimization
tools
Modeling and simulation can play a crucial role throughout
the system lifetime: (1) initial design stages, when the control
strategies are decided, (2) validation of system performance, (3)
commissioning, and (4) operation support. Object-oriented modeling approaches are particularly suited to this task due to their
exibility, inherent openness, and reusability features [4].
Modelica [5] is an equation-based, object-oriented, acausal
modeling language developed for multi-domain modeling of hybrid
systems. Component models are described by differential, algebraic, and/or discrete equations. To enhance reusability, models

112

M. Ali et al. / Energy and Buildings 59 (2013) 111122

can be symbolized graphically via icons [6]. Modelica features, such

as acausal, declarative modeling, code transparency, encapsulation
and modularity, inheritance, and reusability, are summarized in a
number of studies [79]. Modelica Standard Library (MSL) as a basis
for more enhanced model development is freely distributed, while
numerous other libraries can be downloaded or purchased [1013].
Models developed in the Modelica language cannot be executed
directly. Rather, a simulation tool translates a Modelica model into
an executable program [14]. The current study uses Dymola [15]
that executes symbolic manipulations to reduce the dimensionality
of the linear and non-linear systems of equations dened by the
Modelica model. The execution of a Modelica model also performs
automatic differentiation followed by generation and compilation
of C/C++ code [6,16].
Optimization tools are used in combination with energy performance simulation software to optimize a set of parameters for
a given cost function and achieve simulation-based optimization
of HVAC systems [7]. Coupling of TRNSYS-based simulation [17]
and evolutionary programming demonstrated effective potential
of HVAC system optimization for energy management [18]. IDAICE [19] was coupled with GenOpt to show temporal advantages of
simulation-based optimization in nding the optimal values of the
building assembly and HVAC system design variables [20]. A combination of EnergyPlus [21] and GenOpt was used for optimization
of energy consumption in a school building with hydronic heating
system [22].
GenOpt is a generic optimization program for single or multi
parameter optimization with efcient search techniques [23].
GenOpt is developed for optimization problems where the cost
function is computationally expensive to evaluate and its derivatives are not available or may not even exist. GenOpt library consists
of local and global multi-dimensional optimization algorithms.
In addition, new optimization algorithms can also be added by
extending the superclass Optimizer [24,25].

3. Chilled water system optimization

Performance optimization of chilled water system equipment
and control was achieved in different studies through a wide variety of optimum strategies and operating sequences. A theoretical
optimal plant performance (TOPP) model was applied to select optimal control of a chilled water system consisting of three chillers
and ve cooling towers [26]. A simplied model was developed for
evaluation of system congurations of a multiple-chiller system
consisting of a maximum of ten equally sized chillers [27]. With
this model, the designer could easily complete initial evaluation of
system congurations. Similarly, an optimization model was developed for a chilled water system with three chillers and four cooling
towers for deciding on the best operating mode of the entire system
[28]. In another study, four design options were studied to decide
the optimal number and size of chillers operating with maximum
system performance. The study estimated that electricity savings
of 10.1% could be achieved with six chillers of three different sizes
instead of four equally sized chillers [29]. Selection of the optimal
conguration after analyzing all the possible alternatives is a complex task in terms of efforts and time. However, such a task can be
efciently accomplished through a simulation-based optimization
strategy [30].
Design parameters of individual system components play
crucial role for the selection of optimal chilled water system congurations. Therefore, it is essential to operate a system with
optimal values of the design parameters during the evaluation of
different system congurations. Effect of the design parameter values and control on the performance of chilled water systems has
been extensively studied. Load-based speed control integrated with

variable condenser water ow and optimal tower fan speed was

introduced to achieve optimum chilled water system performance
and economical benets [31,32]. It was analyzed that load-based
speed control could reduce the annual system electricity use by
5.3% and operating cost by 4.9% as compared to using constant
speed fans and pumps. Selection of the optimal condenser water
design temperatures is a crucial and complex task at the initial
design stage due to the signicant impact on the system energy consumption. Substantial rst-cost savings can be achieved using high
condenser water temperature differences [33]. Therefore, the current study considers optimal selection of the design temperature
difference across the condenser at various building load demands.
4. Methodology
The current study uses automated simulation-based optimization approach to select the optimal chilled water system
conguration. In this approach, the equation-based objectoriented (EOO) chilled water system model is developed in
Dymola/Modelica and coupled with GenOpt. The model uses hybrid
optimization algorithm with total system power consumption as
an objective function. Finally, the minimum power consumption
with respect to the specic building load demand determines the
optimal system conguration.
4.1. Description of the investigated chilled water system
The considered chilled water system serves a large ofce building in Southern California [26]. As Table 1 illustrates, the system
contains three equally sized chillers rated at a cooling capacity of
2725 kW (775 tons) each. The system has ve draw-through crossow equally sized cooling tower cells, designed for approximately
76 l/s (1200 gpm). Three equally sized condenser water pumps,
designed for 111 l/s (1760 gpm) with 19 kW (25 hp) motors, are
piped together in a headered arrangement so that each pump can
serve any of the chillers or tower cells. Similarly, three equally sized
chilled water pumps, designed for 57 l/s (900 gpm) with 30 kW
(40 hp) are also headered. In the current study, the design conditions are based on the typical operating conditions specied
by the Air Conditioning and Refrigeration Institute (ARI) standard
550/590-2003 [34] and Cooling Tower Institutes (CTI) test conditions [35]. Design wet bulb temperature of 17 C (63 F) and dry
Table 1
Details of the investigated chilled water system.
Total cooling capacity kW [tons]
(Three identical sets of chillers, pumps, and cooling towers)
Chillers
Compressor type
Nominal cooling capacity kW [tons]
Nominal compressor power kW [tons]
Minimum cooling capacity kW [tons]
Design COP
Design chilled water supply/return temperature C [ F]
Design chilled water ow rate l/s [gpm]
Design condenser water entering temperature C [ F]
Design condenser water ow rate l/s [gpm]

8175 [2325]

Centrifugal
2725 [775]
446.4 [127]
457 [130]
6.1
6.7/12.2 [44/54]
57 [900]
29.4 [85]
111[1760]

Cooling towers
Type
Water ow rate l/s [gpm]
Fan motor power kW [hp]
Design wet bulb temperature C [ F]
Design dry bulb temperature C [ F]
Design approach temperatures C [ F]
Design range temperature C [ F]

Draw-through
76 [1200]
18.65 [25]
17 [62.6]
26 [78.8]
8.3[15]
5.56 [10]

Pumps
Rated power of each chilled water pump kW [hp]
Rated power of each condenser water pump kW [hp]

30 [40]
19 [25]

M. Ali et al. / Energy and Buildings 59 (2013) 111122

113

Fig. 1. Graphical representation of the chilled water system model in Dymola.

bulb temperature of 26 C (78.8 F) are used for cooling tower model

according to the ASHRAE standard 90.1-2004 climate data for San
Francisco [36].
4.2. Model development
Component models for the investigated chilled water system
were used from the Modelica standard and LBL Buildings libraries.
Graphical representation of the chilled water system, modeled in
Dymola, is shown in Fig. 1. Centrifugal chiller component model,
developed by AIT (Austrian Institute of Technology) is based on
TRNSYS centrifugal chiller model TYPE 68 [37]. The pump model
with prescribed mass ow rate was used to model three chilled
water pumps and three condenser water pumps. Model of a steady
state cooling tower with variable speed fan using a York cooling
tower performance curve to compute the approach temperature
was employed for heat rejection from the chillers. The applied

pump and cooling tower models were part of the Buildings library.
All component models were operating in parallel arrangement
under the design conditions specied in Table 1.
Centrifugal chillers were operated with xed chilled water supply and return temperatures at 6.7 C (44 F) and12.5 C (54 F),
respectively. Chiller sequencing allowed the running chillers to
operate at the same part load conditions. The demand loads, Qload ,
were applied from an external text le with an increment of
35.16 kW (10 tons) up to the peak load demand. Each applied load
determined the evaporator mass ow rate assuming xed design
temperature difference of 5.55 C (10 F) across the evaporator.
Thus, load-based mass ow rates were applied through the prescribed ow chilled water pumps on the evaporator side of the
chillers. Similarly, the condenser water ow rate was determined
to satisfy the required amount of heat rejection by the condenser.
Such heat rejection was computed from the chiller model based
on the applied load, Qload . The condenser water ow rates were

114

M. Ali et al. / Energy and Buildings 59 (2013) 111122

Table 2
Summary of estimated initial costs (costs in D are estimated based on the 1.23$/D exchange rate).
Description

Cost/ton (D [$])

Total unit cost (D [$])

Water-cooled centrifugal chiller (nominal 775 tons each)

Chiller installation cost
Cooling tower (nominal 400 tons each)
Cooling tower installation cost
Piping/tting/valve
chilled water side
a
condenser water side
b
adding another chiller in the system
c
adding another cooling tower in the system
d
Estimated baseline cost for 1 chiller and 1 cooling tower
Contractor markup (25%)
Total baseline cost
Estimated cost for adding each chiller
Contractor markup (25%)
Total cost of an additional chiller
Estimated cost for adding each cooling tower
Contractor markup (25%)
Total cost of an additional cooling tower

147 [180]
37 [45]
106 [130]
4.1 [5]

113,925 [139,500]
28,370 [34,875]
42,400 [52,000]
1640 [2000]

13,015 [16,000]
19,523 [24,000]
4393 [5400]
2440 [3000]
218,873 [268,375]
54,718 [67,094]
273,591 [335,469]
146,688 [179,775]
36,672 [44,944]
183,360 [224,719]
46,480 [57,000]
11,620 [14,250]
58,100 [71,250]

calculated keeping the xed design temperature difference across

the condenser according to the calculated heat rejection rate for
the rst and second strategies, while the third strategy additionally varied the temperature differences across the condenser. The
number of cooling towers was decided based on the calculated condenser water ow rate. The cooling towers operated with xed
design conditions in terms of design range and approach temperatures given in Table 1. In the rst strategy, the number of cooling
towers varied according to the real considered system but in the
remaining strategies cooling towers were varied according to the
ow turndown ratio up to one-third of the design [38]. According
to such criterion, maximum number of cooling towers in the current study could be 25 provided with an appropriate temperature
difference across the condenser at higher loads. Such high number
of cooling towers is modeled as an array of parallel components
with identical design parameters. Consequently, Dymola/Modelica
model was capable of varying the number of chillers, number
of cooling towers, temperature difference across the condenser,
cooling tower fan input signal, and load, from an external text
le in order to nd optimal chilled water system conguration
setup.

[33,38,41,42]. The initial and installation costs of water-cooled centrifugal chillers and cooling towers are estimated based on the unit
capacity, i.e. cost per ton. The piping costs, including ttings and
valves, are estimated using Pipe Size Optimization tool [33,43]. The
tool calculates the initial piping costs based on the ow rate for
specied piping segments. In addition, the tool also accounts for
the number and types of various valves and ttings typically used
in chilled water systems. The costs of different valves and ttings
are also provided based on the respective pipe sizes. In the current study, the number and type of valves and ttings are decided
from the piping schematic of the reference system [26]. The initial and installation costs of water-cooled centrifugal chillers and
cooling towers along with the piping/tting/valve costs are summarized in Table 2. The estimated value of 25% contractor markup
is considered in the calculations [38].
As the current study involves different congurations of the
chilled water system varying the number of chillers and cooling
towers, the nal estimated initial cost of each conguration is
calculated from Eq. (2).
Ctotal = CH CPchiller (Cchiller + Cinst.CH ) + CT CPtower
(Ctower + Cinst.CT ) + Cpiping/fittings/valves

(2)

4.3. Objective function

The current study denes the objective function for system
design optimization as the total system power consumption. For
the optimal conguration design, the objective function is minimal while satisfying the building load demand. Eq. (1) computes
the objective function within the Dymola/Modelica model for various system congurations considering the power consumption
of specic number of chillers (chiller.PTot ), chilled water pumps
(CHPump.PElc), condenser water pumps (CTPump.PElc), and cooling
towers (CoolingTower.PFan) against the xed load demand.
Ptotal = CHPump.PElc + CTPump.PElc + CH Chiller.PTot + CT
CoolingTower.PFan

(1)

Here CH is the number of chillers and CT is the number of cooling

towers.
Initial costs of each system conguration are also important
aspect for system selection. The initial costs involve the cost of
chillers, cooling towers and piping/ttings/valves required for efcient hydronic design. However, the costs of such equipment
vary country-to-country and region-to-region. Therefore the costs
are estimated based on the data published by the selected manufacturers [39,40] and cost studies related to such equipment

The piping/tting/valve cost is calculated from Eq. (3).

Cpiping/fittings/valves = CCHWside + CCWside + CH Cadd.chiller + CT
Cadd.tower

(3)

Here Ctotal is the total initial cost, CPchiller , CPtower are the chiller
and cooling tower nominal capacities (tons), Cchiller , Ctower are the
unit initial the chiller and cooling tower costs (cost/ton), Cinst.CH ,
Cinst.CT are the unit installation costs of the chiller and cooling
tower (cost/ton), respectively, Cpiping/ttings/valves is the total cost of
piping/ttings/valves, CCHWside , CCWside are the baseline costs of piping/ttings/valves for the chilled water and condenser water sides,
and Cadd.chiller , Cadd.tower are the costs of piping/ttings/valves for
adding each chiller and cooling tower in the system, respectively.
The estimated initial costs of each conguration are used for
economic considerations related to the automatically chosen system congurations based upon the dened optimization objective
function.
4.4. Optimization algorithm
As previously mentioned, overall optimization of the chilled
water system involves ve design variables with the total system

M. Ali et al. / Energy and Buildings 59 (2013) 111122

115

Table 3
Design variables and bounds.

Minimum
Maximum
Step
Initial

Tower fan
speed, F (%)

Temp. difference
condenser side, T ( C) [ F]

No. of
chillers, CH

No. of cooling
towers, CT

Building load Qload

(kW) [Tons]

0.3
1
0.01
1

3 [5.4]
15 [27]
0.01
3

1
3
1
1

3
18
1
3

1055 [300]
7032 [2000]

power consumption as an objective function. Two discrete design

variables are the number of cooling towers CT, and the number
of chillers CH. Three continuous design variables are building load
demand Qload , temperature difference across the condenser T, and
cooling tower fan speed F. Table 3 shows bounds of the considered
variables.
GenOpt optimization algorithms are categorized with respect
to the problems they are meant to solve. Thus, various algorithms
are available for continues, discrete, and mix of continuous and
discrete independent variables. For problems including both continuous and discrete independent variables, such as the current
optimization, GenOpt contains Particle Swarm Optimization (PSO)
algorithms, and a hybrid Generalized Pattern Search Particle Swarm
Optimization with Constriction Coefcient HookeJeeves (GPSPSOCCHJ) algorithm.
The hybrid global optimization algorithm GPSPSOCCHJ consists
of a stochastic population-based constriction coefcient PSOCC
algorithm and a direct search HookeJeeves (HJ) algorithm. The key
advantage of this algorithm is the global PSO search which increases
the possibility of getting close to the global minimum rather than
achieving only a local minimum, while the HJ algorithm renes the
search locally [44]. The GPSPSOCCHJ algorithm parameters chosen
in the current study are summarized in Table 4 [25].
4.5. Simulation-based optimization
After the development of the chilled water system model in
Dymola/Modelica the simulation model is coupled with GenOpt to
nd optimal values of design parameters, which ultimately decide
the optimal system conguration. Fig. 2 shows the implementation
of the simulation-based optimization approach. The optimization
process repeats iteratively until a minimum of the objective function is found [25,30].
5. Results and discussion
A wide range of design and conguration factors inuences the
chilled water system performance. At a system design level, the
selection of suitable design conditions plays a crucial role, e.g. nominal water ow and temperature difference across the evaporator
Table 4
Optimization algorithm input parameters.
Parameters

Value

Neighborhood topology
Neighborhood size
Number of particles
Number of generations
Seed
Cognitive acceleration
Social acceleration
Max velocity gain continuous
Max velocity discrete
Constriction gain
Mesh size divider
Initial mesh size exponent
Mesh size exponent increment
Number of step reductions

von-Neumann
5
20
5
1
2.8
1.3
0.5
4
0.5
2
0
1
4

and condenser, cooling tower approach and wet bulb temperature.

At a system conguration level, decision about the number, type,
and size of chillers, pumps, and cooling towers extensively affect
the energy performance of chilled water systems.

5.1. Baseline system congurations: 1st strategy

In the rst strategy, the number of chiller(s) varied between 1
and 3, while the number of cooling tower varied between 3 and 5 to
nd optimal system conguration of the considered chilled water
system. The temperature difference across the condenser was xed
at 4.45 C (8 F) according to ARI standard design conditions with
xed design chilled water supply and return temperature and full
tower fan speed.
The chiller component model was operational in three modes:
(1) shutdown mode, if the applied Qload is less than the minimum
load, Qmin , specied in the model, (2) normal mode, if applied load
is more than the specied minimum and less than the specied
maximum load, and (3) overload mode, if Qload is more than the
specied maximum load, Qmax . The modes enable the chiller model
to properly react during the simulation subject to the applied load.
In the shutdown mode, the leaving condenser and evaporator water
temperatures become equal to the corresponding entering temperatures and total power consumption, condenser heat transfer rate,
and COP are set to zero. In the overload mode, Qload is set to Qmax
and new leaving evaporator water temperature and COP are determined. As such, the modes are needed for the simulation control,
rather than a real system, in order to prevent the chiller model from
coming up with unrealistic chilled water supply temperatures to
meet the cooling demand.
In the current study, the chiller is always operated in the normal
mode. Therefore, the number of chillers was varied satisfying the
normal operating mode requirements. Fig. 3 shows the total system
power consumption and initial cost of different congurations at
critical loads where system conguration changes to yield energy
benet. The power consumption of each conguration shown in
Fig. 3 is based on xed design conditions throughout the load variation. In such cases the variable loads correspond to the part load
operation of the system due to variable building occupancy.
In Fig. 3(A), the optimal conguration requiring minimum total
power consumption at Qload of 1054.8 kW (300 tons) has 1 chiller
(CH) and 3 cooling towers (CT). Similarly, optimal conguration
at 1582.2 kW (450 tons) is 1CH4CT, at 2461.2 kW (700 tons) is
1CH5CT, at 3516 kW (1000 tons) is 2CH5CT, and at 5274 kW
(1500 tons) is 3CH5CT as shown in Fig. 3(B)(E), respectively. The
results show that adding cooling towers is benecial for two reasons: (1) reduction of the entering condenser water temperature
reduces chiller power consumption due to decrease in chiller lift,
and (2) decrease in pressure drop across the condenser water pump
reduces the pump power consumption.
The power consumption is also affected by the number of
chillers, depending upon the chiller coefcient of performance
(COP) at part load conditions. For example, consider the case when
2461.2 kW (700 tons) Qload is applied on chiller(s) with nominal
load of 2725 kW (775 tons) and ve cooling towers. If the system
operates with one chiller, the applied load is approximately 90% of

116

M. Ali et al. / Energy and Buildings 59 (2013) 111122

Fig. 2. Flow chart of simulation-based optimization approach.

the nominal load, yielding a chiller COP of 7.68. Chiller COP reduces
to 7.57 and 6.83 when system operates with two and three chillers
at 45% and 30% of the nominal load, respectively.
Fig. 4 summarizes the optimal conguration options minimizing
the total system power consumption with respect to the number
of chillers and cooling towers at full range of load demand. The
optimal congurations shown in Figs. 3 and 4 are in agreement
with the real system design [26] and thus validate the proposed
methodology.
Moreover, due to the initial conguration costs shown in Fig. 3,
savings in total power consumption are more signicant at higher
than at lower demand loads per unit initial cost increment.
5.2. Modied system congurations: 2nd strategy
More cooling towers within the ow limit, i.e. ow turndown
ratio of one-third of the design ow [38], are generally benecial as
lower fan speed and power are required. The original chilled water
system design used ve cooling towers with 76 kg/s design ow.
In the second strategy, the number of cooling towers within the
ow constraint is based on the temperature difference across the
condenser. Such temperature difference determines the required
water ow rate through the condenser and cooling towers to satisfy
the condenser heat rejection for specic Qload . Eqs. (4) and (5) were
used to nd the lower and upper limits for the number of cooling
towers, respectively, satisfying the ow constraints.

Upper limit:
CT

Q
1
C1
load
C2
ml
T

(4)

Lower limit:
CT

C1
Q
1
load
C2
md
T

(5)

where CT is the number of cooling towers, C1 is the coefcient

between the applied load and condenser heat rejection, C2 is the
specic heat of water, Qload is the applied load, T is the temperature difference across the condenser, md is the cooling tower
design water ow rate, i.e. 76 kg/s, and ml is limiting water ow
rate, i.e. 25.3 kg/s. Integer output of the above equations was used
to determine the number of cooling towers. For example, if the Qload
is 2637 kW (750 tons), the temperature difference across the condenser could vary between 3 C (5.4 F) and 9.5 C (17.1 F) within
the cooling tower ow design limits. At 3 C (5.4 F), mass ow rate
to satisfy the condenser heat rejection was 236 kg/s (3741 gpm),
resulting in the chilled water system operation with a maximum of
9 cooling towers within ow limit. Thus, 39 cooling towers were
evaluated to nd the impact on system performance at the specic
load. Fig. 5 shows possible numbers of cooling towers for different
values of temperature difference across the condenser along with
the corresponding water mass ow rates at 2637 kW (750 tons).
Similarly, the feasible number of cooling towers is determined for
all load values.

M. Ali et al. / Energy and Buildings 59 (2013) 111122

117

Fig. 3. Total power consumption and costs of the Baseline system congurations at various demand loads (A) Qload = 1055 kW, (B) Qload = 1582 kW, (C) Qload = 2637 kW, (D)
Qload = 3516 kW, (E) Qload = 5274 kW, and (F) Qload = 7032 kW.

Fig. 4. Summary of optimal system congurations.

Fig. 5. Number of cooling towers within ow limit at Qload of 2637 kW (750 tons).

118

M. Ali et al. / Energy and Buildings 59 (2013) 111122

Table 5
Range of conguration parameters used for optimization at various loads with constant condenser side temperature difference (4.45 C [8 F]) at full fan speed (100%).
Load, Qload (kW [tons])

No. of chillers, CH

No. of cooling towers, CT

7032 [2000]
5274 [1500]
3516 [1000]
2461 [700]
1582 [450]
1055 [300]

3
23
23
13
12
12

617
412
38
26
24
23

Compared to the real system design operating with maximum

ve cooling towers, the investigated chilled water system was
operated with more cooling towers in accordance to the aforementioned relationships (4) and (5). The upper and lower limits of the
number of cooling towers depend on the temperature difference,
T across the condenser for each load value. In the second strategy,
the feasible number of cooling towers operating at full fan speed
is presented in Table 5 with xed T of 4.45 C (8 F) at various
Qload values. Number of cooling towers with varying T across
the condenser is considered in the third strategy. Results of the
optimization show that operating the chilled water system with
more cooling towers is benecial compared to the referent design
system. Even with full fan speed, increasing the number of cooling
towers will decrease the total system power consumption. Fig. 6

shows signicant decrease in the total system power consumption

due to the introduction of additional cooling towers at higher
loads. Although such effect diminishes as the number of cooling
towers rises, the second strategy conrms the best design practice:
maximize the efciency by running as many cooling towers as
possible within the ow limit [26,38]. At the same time, the costs
linearly depend upon the number of used equipment components
and a break-even point can be determined based on the economic
analyses evaluating the cost-benet of various design congurations based on yearly operating prole. However, such analyses
fall outside of scope of the current study.
5.3. Overall chilled water system optimization: 3rd strategy
Optimal chilled water systems can be selected at the design
stage with respect to system design and conguration parameters. For the design optimization, it is signicant to vary condenser
water ow rate, temperature difference across the condenser,
and cooling tower fan speed for specic load demand. Optimization of the system conguration in terms of the number of
chillers and cooling towers also extensively affects the energy
performance of the whole system. For the overall system optimization, simulation-based optimization strategy was implemented by
coupling Dymola/Modelica with GenOpt and using a hybrid PSOGPSCCHJ optimization algorithm.

Fig. 6. Total power consumption and costs of the modied system congurations at various demand loads (A) Qload = 1055 kW, (B) Qload = 1582 kW, (C) Qload = 2637 kW, (D)
Qload = 3516 kW, (E) Qload = 5274 kW, and (F) Qload = 7032 kW.

M. Ali et al. / Energy and Buildings 59 (2013) 111122

119

Fig. 7. Iteration runs for the minimization of Ptotal at Qload of 3516 kW [1000 tons].

Fig. 7 shows a sample optimization search made for one specic

load demand. Difference between the two algorithms, PSO stochastic population-based and HJ direct search, forming the hybrid
PSOGPSCCHJ algorithm can be identied from the iterations shown
in Fig. 7. The iterations involve computation of objective function
with respect to varying parameters related to system design and
conguration. The execution time for an optimization using a single generation as PSO input parameter at a specic load demand
was about 1.5 h on a 3.2 GHz PC with 1 GB RAM, as compared to
about 10 min on a 3.16 GHz Core 2 PC with 4 GB RAM. Increasing
the number of PSO generations from 1 to 5 increased the execution
time threefold.
Peaks of Ptotal in both, PSO and GPS HookeJeeves regions of the
hybrid algorithm depict the penalty due to the unfeasible number
of cooling towers during the variation of T across the condenser.
Thus, for the optimal chilled water system design, the key concern is selection of the design temperature difference across the
condenser.
One way to ensure optimal design could be through the investigation of chiller lift, i.e. the difference between chilled water supply
and condenser water return temperatures, as a proxy for the difference between condenser and evaporator refrigerant pressures.
Lower lift would imply lower chiller power consumption but
increased consumption of other components. However, the current
strategy nds the optimal condenser supply-return temperature
difference which can be considered as a proxy for the chiller lift in
the case of xed chilled water supply temperature.
Feasible range of the possible system optimization parameters
against various load demands is shown in Table 6. Fig. 8 shows a
subset of the possible alternatives evaluated by the optimization
algorithm deciding the selection of parameters that cause minimization of the objective function. The presented ranges of Ptotal
are based on the variation of T across the condenser and variation of cooling tower fan speeds. The estimated initial costs of each
evaluated conguration are also presented. As the overall optimization resulted in the selection of maximum ve cooling towers, the
results indicate the benet of varying the water temperature difference across the condenser and cooling tower fan speeds instead
of using constant design values. In addition, such variation resulted

in lower Ptotal and smaller number of cooling towers than the other
two previously considered strategies. Therefore, such design is also
more appropriate from economic aspects as it would decrease the
initial cost of the overall system.
Table 7 shows the optimal combination of system parameters
to achieve the minimum total power consumption. For example, at
the low load demand of 1054.8 kW (300 tons), the chilled water system optimally operates with one chiller and three cooling towers
maintaining the optimal design T of 4 C (7.2 F) across the condenser, while the optimal cooling tower fan speed is 47.94%. The
system has a minimum total power consumption of 171.56 kW. All
other feasible combinations in terms of the number of chillers, cooling towers, temperature differences across the condenser, and CT
fan speeds will result in higher total system power consumptions.
Similarly, at the peak design load of 7032 kW (2000 tons), the optimal system design conguration yields a total power consumption
of 1557.3 kW comprising of three chillers and ve cooling towers
operating with optimal design T of 16.94 C (30.5 F) at fan speed
of 85.3%. The optimal congurations comprise of maximum ve
cooling towers despite of the fact that the number of cooling towers within the ow limit can be up to 25. Consequently, in the case of
varying condenser water temperature difference, water mass ow
rate and cooling tower fan speed, may not be benecial to run as
many cooling towers as possible within the ow constraints. The
results shown in Table 7 are in agreement with the real system congurations. The optimal fan speeds below 90% also conrm the best
design practice, as suggested by Mark Hydeman, P.E., Taylor Engineering: running the cooling tower fans below 90% speed would
achieve higher energy efciency. Due to the cubic relation between
the fan power and airow rate, the cooling tower energy consumption greatly increases with the top 10% fan speed, achieving a very
small drop in condenser water supply temperature.
5.4. Summary of the chilled water system analyses
A methodology is proposed for design optimization of the chilled
water system by applying three strategies: (1) baseline system
comprising of a maximum of ve cooling towers with xed design
temperature difference across the condenser at full fan speed,

Table 6
Range of system parameters used for optimization at various loads.
Load, Qload (kW [tons])

No. of chillers, CH

No. of cooling towers, CT

Temperature difference across condenser, T ( C [ F])

Fan speed, F (%)

7032 [2000]
5274 [1500]
3516 [1000]
2461 [700]
1582 [450]
1055 [300]

3
23
23
13
13
12

325
318
312
38
36
24

325 [5.445]
315 [5.427]
312 [5.421.6]
38 [5.414.4]
35 [5.49]
34 [5.47.2]

30100
30100
30100
30100
30100
30100

120

M. Ali et al. / Energy and Buildings 59 (2013) 111122

Fig. 8. Total power consumption and costs of the overall optimized system congurations at various demand loads (bars represent power range due to varying optimization
input parameters) (A) Qload = 1055 kW, (B) Qload = 1582 kW, (C) Qload = 2637 kW, (D) Qload = 3516 kW, (E1) Qload = 5274 kW with 2 chillers, (E2) Qload = 5274 kW with 3 chillers
and (F) Qload = 7032 kW.

(2) modied system with increased number of cooling towers

according to the ow turndown limit and xed design temperature
difference across the condenser at full fan speed, and (3) modied
system varying both, the system design and conguration parameters. While the rst two strategies conrm the established best
design practice and verify the simulation models, the third strategy
denes a systematic approach for the overall design optimization

of chilled water systems. In addition, the strategies consider initial costs of equipment, including chillers, cooling towers and
piping/ttings/valves, having in mind that additional chillers and
cooling towers typically decrease the annual energy costs and payback period [33,41].
Optimal total system power consumptions and system energy
use in kW/ton from the considered approaches are shown in

Table 7
Optimal values of system parameters and objective function at various loads.
Load, Qload
(kW [tons])

No. of chillers,
CH

No. of cooling
towers, CT

Temperature difference across

condenser, T ( C [ F])

Fan speed,
F (%)

Optimal
Ptotal (kW)

7032 [2000]
5274 [1500]
3516 [1000]
2461 [700]
1582 [450]
1055 [300]

3
3
2
2
1
1

5
5
5
4
4
3

16.94 [30.5]
12.68 [22.8]
8.45 [15.2]
7.63 [13.7]
4.88 [8.8]
4 [7.2]

85.31
75.44
62.8
59.75
49.3
47.9

1557.3
993.1
582.7
394.9
240.7
171.6

M. Ali et al. / Energy and Buildings 59 (2013) 111122

121

Table 8
Optimal values of Ptotal and system energy use of all strategies (minimum values highlighted).
Qload (kW [tons])

1st strategy Ptotal (kW) kW/ton

2nd strategy Ptotal (kW) kW/ton

3rd strategy Ptotal (kW) kW/ton

Percentage of power saving (%)

7032 [2000]
5274 [1500]
3516 [1000]
2461 [700]
1582 [450]
1055 [300]

2757.3
1498.8
752.7
479.7
297.8
207.4

1808.4
1178.8
706.9
479.7
297.8
207.4

1557.3
993.1
582.7
394.9
240.7
171.6

43.5
33.8
22.6
17.6
19.2
17.3

1.37
0.99
0.75
0.68
0.66
0.69

Table 8. The values of energy use (kW/ton) are in agreement with

the typical values for the chilled water systems [41]. It can be
observed that the lowest total power consumption, Ptotal , and system energy use is achieved for the modied system with varying
temperature difference across the condenser and varying fan speed.
Also, the performance of chilled water systems strongly depends on
the appropriate selection of mass ow rates and temperature differences across the condenser. The simulation-based optimization
approach enhances the decisioning about optimal variation in these
parameters resulting in the minimum total system power consumption. Signicant power savings between 17% and 43.5% could
be achieved operating chilled water systems at the optimal design
and conguration parameters compared to the baseline case.

5.5. Potential benets and limitations

The current study developed a model of the chilled water system
based on EOO approach and using open source component libraries.
Optimization of HVAC systems involved variation of component
design parameters as well as congurations. Dymola/Modelica
enabled control of all component parameter values from external
text les for the purpose of design optimization. Serial and/or parallel arrangements, as well as the number of component models
were also decided from external les. Multiple instances of single component models were used within the component arrays to
vary the number of cooling towers. The aforementioned features
helped to develop efcient and effective multi-domain system
models for simulation and optimization. Automated optimization
of the systems involving many design and conguration parameters offered the possibility to identify the most energy efcient
design. The applied optimization algorithm used fewer generations as compared to the literature recommendations [25], as no
variation in system optimization parameters was observed after
the initial two generations. Such a selection of optimization input
parameters saved simulation time, while possibly hindering the
achievement of globally optimal results. The methodology of coupling Dymola/Modelica and GenOpt proved as efcient in terms of
efforts, time, and ease to vary the design and conguration parameters for the selection of optimal system congurations. The main
weakness of Dymola/Modelica application to building simulations
is the lack of sufcient HVAC component models within the existing modeling libraries. However, research activities are underway
to enhance the available libraries by developing additional component models [45,46].
The current study assumes that the chilled water system component sizes are predened and optimizes the number of components
and design operating conditions using xed parameters: chilled
water supply and return temperatures, dry bulb and wet bulb
temperatures. Measurement uncertainties associated with the
equipment selection are neglected as well as their potential impacts
on the optimization results of the total power consumption.
As such the proposed approach could be considered as part of
the iterative design process, rather than an all-inclusive procedure.

0.9
0.78
0.71
0.68
0.66
0.69

0.78
0.66
0.58
0.56
0.53
0.57

6. Conclusions and future work

In the current study, an incremental development of the
methodology for chilled water system optimization is proposed.
The equation-based object oriented modeling approach was used
to model a real chilled water system located at a Symantec
Corporation building in South California, USA. The chilled water
system model developed in Dymola/Modelica was capable of
varying system design and conguration parameters at different
load demands. Dymola/Modelica model was coupled with GenOpt
optimization software. The simulation-based optimization used a
hybrid PSOGPSCCHJ optimization algorithm to nd the optimal
system conguration. The current study analyzes several design
parameters having signicant impact on the system performance.
In addition, the estimated initial cost of each conguration is provided for economic considerations. The chilled water system was
analyzed considering three strategies, two of which involved constant temperature difference across the condenser at full fan speeds
applied to the baseline and modied system congurations. The
third strategy varied the temperature difference across the condenser as well as the cooling tower fan speeds in addition to
the system conguration parameters and proved to be the most
energy efcient. Consequently, the optimal values of the considered
parameters were provided together with the calculated power consumption. Operating chilled water systems at the optimal conditions could result in signicant total system power savings amounting up to 43.5% for the considered cases. The implemented automated simulation-based optimization approach proved efcient in
terms of model development and computational time needed to
nd the optimal congurations. The methodology represents a step
toward the design of software systems able to synthesize new and
optimal HVAC system congurations. Such development should
help the design practitioners to select the optimal system conguration parameters at the initial stage. In the future, the developed
simulation-based methodology will be applied in a more comprehensive analysis using hourly load proles and climatic data.

Acknowledgements
The authors express profound gratitude to Mark Hydeman, P.E.,
Fellow ASHRAE, Principal at Taylor Engineering LLC, Alameda, CA,
USA, and Anton Haumer, Senior Engineer, AIT Austrian Institute
of Technology, Vienna, Austria, for their insightful comments and
expertise.

References
[1] L. Perez-Lombard, J. Ortiz, C. Pout, A review on buildings energy consumption
information, Energy and Buildings 40 (2008) 394398.
[2] J.C. Lam, R.Y.C. Chan, C.L. Tang, D.H.W. Li, Electricity use characteristics of
purpose-built ofce buildings in subtropical climates, Energy Conversion and
Management 45 (2004) 829844.
[3] J.C. Lam, Energy analysis of commercial buildings in subtropical climates, Building and Environment 35 (2000) 1926.

122

M. Ali et al. / Energy and Buildings 59 (2013) 111122

[4] F. Casella, Object-oriented modeling of power plants: a structured approach,

in: Proceedings of the IFAC Symposium on Power Plants and Power Systems
Control, July 58, Tampere, Finland, 2009.
[5] Modelica and the Modelica Association. Available at: http://www.modelica.org
(accessed 18.04.11).
[6] M. Wetter, C. Haugstetter, Modelica versus TRNSYS a comparison between
an equation-based and a procedural modeling language for building energy
simulation, in: Proceedings of the 2nd SimBuild Conference, Cambridge, MA,
USA, 2006.
[7] M. Trcka, J.L.M. Hensen, Overview of HVAC system simulation, Automation in
Construction 19 (2010) 9399.
[8] M. Wetter, Modelica-based modeling and siumlation to support research and
development in building energy and control systems, Journal of Building Performance Simulation 2 (2) (2009) 143161.
[9] Modelica Libraries, Modelica Association. Available at: https://modelica.
org/libraries (accessed 21.03.11).
[10] XRG Simulation, Building Energy Libraries. Available at: http://xrgsimulation.de/en/products (accessed 21.03.11).
[11] University of Kaiserslautern, Chair for Automation. http://www.eit.
uni-kl.de/litz/forschung/BuildingSimulation/Modelica-html/help/ATplus.html
(accessed 03.12.03).
[12] M. Wetter, A Modelica-based model library for building energy and control
system, in: 11th International IBPSA Conference, July 2730, Glasgow, Scotland,
2009.
[13] M. Wetter, Modelica library for building heating, ventilation and airconditioning systems, in: Proceedings of 7th Modelica Conference, September
2022, Como, Italy, 2009.
[14] Modelica Tools, Modelica Association. Available at: https://modelica.org/tools
(accessed 21.03.11).
[15] Dymola, Dynamic Modeling Laboratory. Available at: http://www.3ds.
com/products/catia/portfolio/dymola (accessed 18.04.11).
[16] Dymola, User Manual, Dynasim AB 2004. Available at: http://www.
inf.ethz.ch/personal/cellier/Lect/MMPS/Refs/Dymola5Manual.pdf (accessed
21.05.10).
[17] TRNSYS, Transient System Simulation Tool. Available at: http://www.
trnsys.com/ (accessed 04.06.11).
[18] K.F. Fong, V.I. Hanby, T.T. Chow, HVAC system optimization for energy management by evolutionary programming, Energy and Buildings 38 (2006)
220231.
[19] IDA ICE, IDA Indoor Climate and Energy, Equas Simulation Technology. Available at: http://www.equa.se/eng.ice.html (accessed 15.06.11).
[20] A. Hasan, M. Vuolle, K. Siren, Minimization of life cycle cost of a detached house
using combined simulation and optimization, Building and Environment 43
(2008) 20222034.
[21] EnergyPlus Energy Siumlation Software, U.S. Department of Energy. Available
at: http://apps1.eere.energy.gov/buildings/energyplus/ (accessed 11.06.11).
[22] N. Djuric, V. Novakovic, J. Holst, Z. Mitrovic, Optimization of energy consumption in buildings with hydronic heating systems considering thermal
comfort by use of computer-based tools, Energy and Buildings 39 (2007)
471477.
[23] GenOpt Generic Optimization Program, Building Technologies Department,
Simulation Research Group, Lawrence Berkeley National Laboratory. Available
at: http://simulationresearch.lbl.gov/GO (accessed 21.06.11).
[24] M. Wetter, GenOpt a generic optimization program, in: 7th International
IBPSA Conference, August 1315, Rio de Janeiro, Brazil, 2001.

[25] M. Wetter, GenOpt Generic Optimization Program, User Manual. Available at:
http://gundog.lbl.gov/GO/download/manual-2-1-0.pdf (accessed 16.08.10).
[26] M. Hydeman, G. Zhou, Optimizing chilled water plant control, ASHRAE Journal
(June) (2007) 4454.
[27] W.L. Lee, S.H. Lee, Developing a simplied model for evaluating chiller-system
congurations, Applied Energy 84 (2007) 290306.
[28] R.T. Olson, J.S. Liebman, Optimization of a chilled water plant using
sequential quadratic programming, Engineering Optimization 15 (1990)
171191.
[29] F.W. Yu, K.T. Chan, Strategy for designing more energy efcient chiller
plants serving air-conditioned buildings, Building and Environment 42 (2007)
37373746.
[30] M. Ali, V. Vukovic, M.H. Sahir, Methodology for automated optimization of
HVAC system conguration, in: 41st KGH International Congress on HVAC&R,
Belgrade, December 13, Serbia, 2010.
[31] F.W. Yu, K.T. Chan, Optimization of water-cooled chiller system with load-based
speed control, Applied Energy 85 (2008) 931950.
[32] F.W. Yu, K.T. Chan, Economic benets of optimal control for water-cooled chiller
system serving hotels in a subtropic climate, Energy and Buildings 42 (2010)
203209.
[33] Taylor Engineering Company, Chilled Water Plant Design Guide. Availat:
http://www.taylor-engineering.com/downloads/cooltools/EDR
able
DesignGuidelines CoolToolsChilledWater.pdf (accessed 07.03.11).
[34] Air-Conditioning and Refrigeration Institute, Performance Rating of WaterChilling Packages Using the Vapor Compression Cycle, ARI Standard
550/590-2003, 2003.
[35] Cooling Technology Institute, CTI Standard STD-201 Certied Cooling Towers,
2009.
[36] ASHRAE, ASHRAE Standard 90.1-2004, Energy Standard for Building Except
Low-Rise Residential Buildings, SI edition, 2004.
[37] J.K. Cross, An evaluation of ice and chilled water supply system as thermal
storage media for combustion turbine inlet air cooling systems, MS Thesis at
University of Wisconsin-Madison, 1994.
[38] M. Hydeman, S. Taylor, J. Stein, Codes and Standards Enhancement Report, 2005
Title 24 Building Energy Efciency Standards Update, Code Change Proposal for
Cooling Towers, April 8, 2002.
[39] The TRANE Company, Chilled Water Plant Cost Estimated. Available at:
http://www.trane.com/commercial/uploads/pdf/865/ctv-trt-001-en.pdf
(accessed 11.06.12).
[40] The Carrier Company, Personal Communication with Business Development
Manager in Vienna.
[41] W.P. Bahneth, E.B. Peyer, Variable primary ow chilled water systems:
potential benets and application issues, in: Technical Report Under HVAC&R
Research for 21st Century (21CR) program, March 2004.
[42] J.W. Furlong, F.T. Morrison, Optimization of water-cooled chiller-cooling tower
combinations, Cooling Technology Institute Journal 26 (2004) 1219.
[43] S.T. Taylor, M. McGuire, Sizing pipe using life-cycle costs, ASHRAE Journal
(2008) 2432.
[44] M. Wetter, J. Wright, A comparison of deterministic and probabilistic optimization algorithms for non-smooth simulation-based optimization, Building and
Environment 39 (2004) 989999.
[45] Modelica Library for Building Energy and Control System. Available at:
https://gaia.lbl.gov/bir (accessed 05.08.11).
[46] Modelica Libraries, Catalog Version 2.1. Available at: http://www.bauschgall.de/Modelica Libraries Catalog.pdf (accessed 05.08.11).