Energy Conversion and Management 48 (2007) 355–365

www.elsevier.com/locate/enconman

A simplified modeling of mechanical cooling tower for control

and optimization of HVAC systems
Guang-Yu Jin a, Wen-Jian Cai a,*
, Lu Lu b, Eng Lock Lee b, Andrew Chiang b

a
School of Electrical and Electronic Engineering, Nanyang Technological University, Nanyang Avenue, Singapore 639798, Singapore
b
E-Cube Pte Ltd., 26 Ayer Rajah Crescent #05-01, Singapore 139944, Singapore

Received 2 January 2006; accepted 15 July 2006

Available online 25 September 2006

Abstract

This paper proposes a new, simple, yet accurate mechanical cooling tower model for the purpose of energy conservation and man-
agement. On the basis of Merkel’s theory and effectiveness-NTU method, the model is developed by energy balance and heat, mass trans-
fer analysis. Commissioning information is then used to identified, only three model parameters by Levenberg–Marquardt method.
Compared with the existing models, the proposed model has simple characteristic parameters to be determined and without requiring
iterative computation when the operating point changes. The model is validated by real operating data from the cooling towers of a
heating, ventilating and air conditioning (HVAC) system of a commercial hotel. The testing results show that the performance of the
cooling tower varies from time to time due to different operating conditions and the proposed model is able to reflect these changes
by tuning its parameters. With this feature, the proposed model can be simply used and accurately predict the performance of the
real-time operating cooling tower.
 2006 Elsevier Ltd. All rights reserved.

Keywords: Cooling tower; Effectiveness-NTU method; Energy balance; Heat and mass transfer; Modeling; Parameter identification

1. Introduction in order to simplify the analysis. The model has since

Cooling towers are widely used to remove heat from
industrial processes and from heating, ventilating and air
conditioning (HVAC) systems. Heat rejection in cooling
towers is accomplished by heat and mass transfer between
hot water droplets and ambient air. Much attention is
being currently paid to cooling tower designs as they are
important factors in energy conservation [1–3]. As such,
there have been substantial research interests in modeling
cooling towers [4–11].
The first milestone work may trace back to 1925, when
Merkel [4] developed a practical model for cooling tower
operation, where the water loss of evaporation is
neglected and the Lewis number is assumed to be one
*
Corresponding author. Tel.: +65 6790 6862; fax: +65 6793 3318.
E-mail addresses: jing0002@ntu.edu.sg (G.-Y. Jin), ewjcai@ntu.edu.sg
(W.-J. Cai).
been the basis for most modern cooling tower analysis
and a revised form was adopted by a software package
for energy calculation [12]. However, the Merkel method
does not accurately represent the physics of heat and mass
transfer process in the cooling tower fill [13].
HVAC1Toolkit [12] developed the equations necessary
to apply the effectiveness-NTU method directly to coun-
terflow cooling towers. This approach is particular accu-
rate and simplifies the method of solution when
compare to a more conventional numerical procedure.
The drawback of effectiveness-NTU method is that the
heat transfer coefficient-area product is difficult to deter-
mine and requires detailed geometric information of cool-
ing towers, besides, an iterative computational procedure
is a must when the operating point changes. Bernier [7]
proposed a one-dimensional cooling tower model for ide-
alized spray-type towers based on the analysis of heat and
mass transfer process in cooling towers at water droplet

0196-8904/$ - see front matter  2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.enconman.2006.07.010
356 G.-Y. Jin et al. / Energy Conversion and Management 48 (2007) 355–365

Nomenclature

A heat transfer area (m2) m_ a mass flow rate of air (kg/s)

Aa heat transfer area of air side convection (m2) m_ w mass flow rate of condenser water (kg/s)
Aw heat transfer area of water side convection (m2) m_ w;e equivalent mass flow rate of condenser water
a interfacial contact area of the surface of water (kg/s)
droplets per unit volume of the cooling tower n number of experimental data (dimensionless)
(m1) N number of operating points to determine the
b1–b2 constants (dimensionless) proposed model (dimensionless)
C constant (dimensionless) NTU number of transfer units (dimensionless)
Cs derivative of saturation air enthalpy with respect Q_ rej heat rejection rate of the cooling tower (W)
to temperature R overall heat transfer resistance (C/W)
Cpa specific heat of moist air [J/(kg C)] Ra heat transfer resistance of air convection (C/W)
Cpa,e equivalent ideal gas specific heat [J/(kg C)] Rw heat transfer resistance of water convection (C/
Cpw specific heat of water under constant pressure [J/ W)
(kg C)] T temperature (C)
c1–c9 parameter (dimensionless) TCWS condenser water supply temperature (C)
dk search direction (dimensionless) TCWR condenser water return temperature (C)
D diameter of the round passage or equivalent Twb,i wet-bulb temperature of ambient air (C)
diameter for other shapes (m) Twb,o wet-bulb temperature of outlet air from the
Dcalc calculated data (dimensionless) cooling tower (C)
Dreal real experimental data (dimensionless) uk value of c1–c3 of the kth iteration (dimension-
e1–e2 constants (dimensionless) less)
F(uk) error between the real values and model predict- U heat transfer coefficient [W/(m2 C)]
ing values (dimensionless) UA actual heat transfer coefficient-area product (W/
h enthalpy of unsaturated air in the vicinity of the C)
wetted surface (J/kg) m average fluid velocity (m/s)
ha,i enthalpy of inlet air (J/kg) V total volume of the cooling tower (m3)
ha,o enthalpy of outlet air (J/kg) z elevation (m)
hs enthalpy of saturated air at the temperature of e effectiveness of heat transfer exchanger (dimen-
the wetted surface (J/kg) sionless)
hs,w,i saturation air enthalpy at the temperature of in- ea heat transfer effectiveness in Braun’s model
let water (J/kg) (dimensionless)
hs,w,o saturation air enthalpy at the temperature of u fluid viscosity (m/s)
outlet water (J/kg) q fluid density (kg/m3)
J(uk) Jacobian matrix of uk (dimensionless) x ratio of the minimum heat capacity flow rate to
k thermal conductivity [W/(m2 C)] the maximum heat capacity flow rate (dimen-
kx mass transfer coefficient [kg/(m2 s)] sionless)
M number of operating points to determine Stoec- kk control coefficient in Levenberg–Marquardt
ker’s model (dimensionless) method (dimensionless)
MeM Merkel number (dimensionless)

level. This particular model is useful for cooling tower
designers, but little information is provided to plant oper-
ators for cooling towers already in operation. Soylemez
[10] presented a method for estimating the size and per-
formance of forced draft countercurrent cooling towers
backed with experimental results. Then again, this model
also need iterative computation and is not suitable for
online optimization.
In this paper, a new engineering model, which can be
used to optimize the performance of operating cooling tow-
ers, is proposed. The methodology is based on energy bal-
ance and heat and mass transfer analysis, and introduce
only three characteristic parameters to predict the perfor-
mance of heat transfer in cooling towers without iterative
computation to calculate the heat rejection rate. Leven-
berg–Marquardt method is employed to determine the
three parameters by curve fitting manufacturers’ catalog
data or real time experimental data. Compared with the
existing models, the proposed model is simple and accu-
rate; therefore, it is expected to be used for real time pre-
dicting the performance of cooling tower. The testing
data of cooling tower of a commercial HVAC plant is
employed to validate the engineering model and the results
are satisfactory.
 G.-Y. Jin et al. / Energy Conversion and Management 48 (2007) 355–365
2. Review of cooling tower models
2.1. Mechanism analysis of mechanical cooling towers
Mechanical draft cooling towers are much more widely
used than natural draft cooling tower. Mechanical towers
utilize large fans to force air through circulated water.
The water falls downward over fill surfaces which help
increase the contact time between the water and the air.
This helps maximize heat transfer between the two. Fig. 1
shows the scheme schematic representation of two types
of mechanical draft cooling towers: counterflow and cross-
flow. The principle of counterflow means that the water
flows down while the air is forced upwards by a fan. As
the water flows further through the system, the air it
encounters is fresher, in other words cooler and less satu-
rated with moisture. And, the principle of crossflow means
that warm water flowing down through a cooling unit is
cooled by air drawn upwards by a fan. Evaporation and
direct heat exchange cause a rapid drop of temperature.
Fills or packing are employed in cooling towers to
increase the contact area and contact time between the
water that needs to be cooled and the cooling air. There
are basically three different types of fill designs: film, splash,
and trickle type fills. A thin film of water runs down the film
fill surface while the splash fill breaks the water stream into
smaller droplets. The trickle film is basically a combination
of film and splash type fill. Fig. 3 shows an idealized model
of the interface between the water and the air for all types of
counterflow cooling tower fill materials. The results of
Kloppers and Kröger’s study [13] reveal that the differences
between the Merkel, and effectiveness-NTU approaches,
are independent of the type of fill considered.
2.2. Merkel’s model
The Merkel model relies on several critical assump-
tions:
Fig. 1. Mechanical draft counterflow and crossflow tower.
357
1. the Lewis factor, Lef, relating heat and mass transfer is
equal to 1;
2. the air exiting the tower is saturated with water vapor
and it is characterized only by its enthalpy;
3. the reduction of water flow rate by evaporation is
neglected in the energy balance.
Eqs. (1) and (2) are obtained from mass and energy bal-
ances of the control volumes where air is in counterflow
with a downwards flowing water stream. For the Merkel
theory it is assumed that the change in water mass flow rate
due to evaporation is negligible, i.e. dm_ w ¼ 0.
dh k x aV
¼ ðhs  hÞ ð1Þ
dz m_ a
dT w m_ a 1 dh
¼ ð2Þ
dz m_ w C pw dz
Eqs. (1) and (2) describe, respectively, the change in the en-
thalpy of the air–water vapor mixture and the change in
water temperature as the air travel distance changes. Eqs.
(1) and (2) can be combined to yield upon integration the
Merkel expression:
Z T CWS
k x aV cpw dT w
MeM ¼ ¼ ð3Þ
m_ w T CWR ðhs  hÞ
where MeM is the tower characteristic, i.e. Merkel number,
according to the Merkel approach. It is often difficult to
evaluate the surface area per unit volume of fill due to
the complex nature of the two-phase flow in fills. It is, how-
ever, not necessary to explicitly specify the surface area per
unit volume or the mass transfer coefficient as these are
contained in the Merkel number which can be obtained
from the right-hand side of Eq. (3).
It is not possible to calculate the true state of the air
leaving the fill according to Eq. (3). Merkel assumed that
the air leaving the fill is saturated with water vapor. This
assumption enables the air temperature leaving the fill to
be calculated.
358 G.-Y. Jin et al. / Energy Conversion and Management 48 (2007) 355–365

pwm_ a C
2.3. Effectiveness-NTU method As can be seen, when , the NTU value according
<
ðdTdhws Þ
m_ w

to Merkel theory can be given by

The effectiveness-NTU method is based the same simpli-
fying assumptions as the Merkel method. HVAC1Toolkit
[12] presents a counterflow cooling tower model by typical
effectiveness-NTU method. Based on this method, the total
heat rejection rate due to the direct contact between air and
water under steady-state conditions:
U dA
dQ_ rej ¼ ðhs  hÞ
C pa
The first step in the modeling procedure is to assume
that the moist air enthalpy can be defined by the wet-bulb
temperature only and that the moist air can be treated as
an equivalent ideal gas characterized by the following mean
specific heat:
Dh ha;o  ha;i
C pa;e ¼ ¼ ð5Þ
DT wb T wb;o  T wb;i
The water side conductivity is much greater than the air
side conductivity. Therefore, the wetted surface tempera-
ture is also assumed to be equal to the water temperature.
On the basis of these assumptions, the expression of the
total heat rejection rate becomes:
UC pa;e
dQ_ rej ¼ dAðT w  T wb Þ
C pa
An energy balance on the water and air sides gives the
following:
dQ_ rej ¼ m_ w C pw dT w ¼ m_ a C pa;e dT wb
The water-side effectiveness of the cooling tower can be de-
fined by analogy to the effectiveness of a simple heat
exchanger:
T CWS  T CWR
e¼
T CWS  T wb;in
The combination of Eqs. (6) and (7) is integrated. The re-
sult of this integration is combined with Eq. (8) to provide
the following expression of the cooling tower effectiveness:
m_ w
NTU ¼ MeM : ð12Þ
m_ a
From the derivation of Eq. (12), it is self evident that the
value of NTU has direct relation with tower characteristic
MeM. Consequently, Eq. (12) establish the relationship be-
tween NTU and physical variables in order to predict the
ð4Þ performance of cooling towers. It was pointed out that
the main influential factors are m_ a and m_ w , and Twb and
TCWS theoretically have no effect on the value of NTU
and tower characteristic MeM [14]. Consequently, the cool-
ing tower can be modeled, in the steady-state regime, by an
equivalent classical counterflow heat exchanger (Fig. 2.).
The first fluid is water and the second fluid is an equiv-
alent fluid entering the heat exchanger at temperature Twb,i
and characterized by the specific heat Cpe (Eq. (5)). The
heat exchanger is characterized by a single parameter, its
global equivalent coefficient-area product UAe, which is
related to the actual cooling tower heat transfer coeffi-
cient-area product by the following expression:
C pe
UAe ¼  UA ð13Þ
Cp
ð6Þ Therefore, in the steady-state regime, the combination of
the mass and heat transfer of water-vapour towards the
air can be equivalent into a behavior of heat transfer by
convection between water and moist air.
ð7Þ 2.4. Stoecker’s model
Apart from using NTU based on heat transfer theory,
an entirely empirical model of cooling tower was proposed
by Stoecker [5]. In this model, the cooling tower was sup-
posed with constant airflow rate and water flow rate based
ð8Þ
on polynomial approximation.
Q_ rej ¼ m_ w C pw ðT CWS  T CWR Þ ð14Þ
T CWR ¼ c1 þ c2 T wb þ c3 T 2wb þ c4 T CWS þ c5 T 2CWS

1  eNTUð1xÞ þ c6 T wb T CWS þ c7 T 2wb T CWS þ c8 T wb T 2CWS

e¼ ð9Þ
1  xeNTUð1xÞ þ c9 T 2wb T 2CWS ð15Þ
where If M operating points (M > 9) are collected for model iden-
C_ min tification, Eq. (15) falls into a system of linear equations:
x¼ ð10aÞ
C_ max
C_ min ¼ minðC pw m_ w ; C pa;e m_ a Þ ð10bÞ
C_ max ¼ maxðC pw m_ w ; C pa;e m_ a Þ ð10cÞ
C pa;e UA
NTU ¼  ð10dÞ
C pa C_ min
Then, the heat rejection rate is expressed by following
equation:
Q_ rej ¼ e  C_ min  ðT CWS  T wb;i Þ ð11Þ Fig. 2. Conceptual scheme of the counter-flow cooling tower.
 G.-Y. Jin et al. / Energy Conversion and Management 48 (2007) 355–365 359

WX ¼ C þ D ð16Þ As we mentioned in previous section, the cooling tower

2 3 can be modeled, in the steady-state regime, by an equiva-
c1 lent classical counterflow heat exchanger. Thus, during
6 .. 7
where D is error, X ¼ 4 . 5, W ¼ the process of heat transfer, heat is transferred from water
2 3 2 3 c drops to the surrounding moist air by convection. As
1 T wb;1    T wb;1 T 2CWS;1
2
Q_ rej;1 9
61 T 7 6 _ 7 shown in Fig. 3, the equivalent heat transfer process can
6 wb;2    T 2wb;2 T 2CWS;2
7 Qrej;2 7
6. .. .. ..7, C ¼ 6
6 . 7 and D ¼ be classified into two parts: water convection and air
6. 7 4 .. 5 convection.
4. . . .5
1 T wb;M    T 2wb;M T 2CWS;M Q_ rej;M The heat rejected rate can be calculated using the overall
2 3 heat resistance according to heat transfer theory and the
d1 energy balance [15].
6 d2 7
6 7 T CWS  T wb;i
6 .. 7. The best estimation X* of X can then be obtained Q_ rej ¼ ð18Þ
4. 5 R
dM
Compare Eqs. (9) and (18), we can have the another defini-
using standard least-squares method as tion of the overall hear resistance:
X  ¼ ðWT WÞ1 WT C ð17Þ 1
R¼ ð19Þ
e  C_ min
3. Proposed cooling tower model Theoretically, the overall heat resistance R consists of two
parts: heat resistance of the water convection, heat resis-
3.1. Construction of cooling tower model tance of the air convection.
R ¼ Rw þ R a ð20Þ
Before developing the cooling tower model, we use the
follow reasonable assumption: The term convection heat transfer refers to the heat ex-
changed between an interface and the fluid moving over
1. heat and mass transfer in a direction normal to the flow the interface. The amount of heat thus transferred depends
only; on the nature of the interface and the fluid, the geometry of
2. the Lewis factor is a constant; the cooling tower and the velocity of fluid over the interface
3. the reduction of water flow rate by evaporation is as well as the temperature differences.
neglected in the energy balance; As water and air in cooling towers are moved by pumps
4. the cross-sectional area of the tower and temperature and fans, respectively, the type of heat transfer between
distribution throughout the water stream at any cross- water and air can be considered as forced convection.
sections is uniform; Assuming the uniform cross-section area of water flow is
5. the saturation enthalpy is approximately linear with round, with the diameter D, by dimensional analysis [16],
respect to web-bulb temperature. the heat resistance of water convection Rw and mass flow
rate m_ w can be calculated by Eqs. (21) and (22).

Fig. 3. Heat exchange scheme using an electric analogy.

360 G.-Y. Jin et al. / Energy Conversion and Management 48 (2007) 355–365
 e  e
1 Dmq 1 C pw l 2 k coefficient is constant, thus, k can be lumped in the
¼C   Aw ð21Þ
Rw l k D unknown parameters c1 and c2 in Eq. (25).
The proposed model is deduced from the analysis of
D2
m_ w ¼ m  p  q ð22Þ effective-NTU method, thus, the utilization of the proposed
4 approaches is independent of the type of fill considered.
The values of the constant C and the exponents e1 and e2 in This feature make it can be wildly used to describe the
Eq. (21) are difficult to be determined exactly. characteristic of cooling tower with different kinds of fill
For steady stable flow, it is reasonable to assume that arrangement.
the value of Aw, q and m remain constant. Moreover, l, k
and Cpw are approximately constants if the temperature dif- 3.2. Parameter identification
ference is not too big (not greater than 6 C for water and
50 C for gas) [17]. Replace m in Eq. (21) by Eq. (22), we If variables TCWS, Twb, m_ w , m_ a and Q_ rej can be obtained,
obtain: manipulating the cooling tower into N states, it follows
" #
1 e1
e1 4 C
C epw
2
 k 1e2 c1 m_ cw;1
3

¼ m_ w  Aw ¼ b1 m_ ew1 Q_ rej;1 ¼  c3 ðT CWS;1  T wb;1 Þ

Rw pe1 le1 e2  D1þe1 m_
1 þ c2 m_ w;1
ð23Þ a;1
e 1e
4e1 C C pw2  k 2 c1 m_ cw;2
3
b1 ¼ e  e e  Aw Q_ rej;2 ¼  c3 ðT CWS;2  T wb;2 Þ
p 1 l 1 2  D1þe1 m_
1 þ c2 m_ w;2 a;2 ð26Þ
Following the same argument, the heat resistance of air
..
convection can be expressed. .
1 c1 m_ cw;N
3
¼ b2 m_ ea1 Q_ rej;N ¼  c3 ðT CWS;N  T wb;N Þ
Ra m_
1e ð24Þ 1 þ c2 m_ w;N
e a;N
4e1 C C pe2  k 2
b2 ¼   Aa To determine the empirical parameter c1–c3, an objective
pe1 le1 e2  D1þe1
function is defined as
Substituting Eqs. (20), (23) and (24) into (18), a cooling
T
tower model for computing heat rejection rate is obtained MIN F ð½c1 ; c2 ; c3  Þ
as 0 12
XBN c 3
c1 m_ w;i
Q_ rej ¼
b1 m_ ew1  b2 m_ ea1
ðT CWS  T wb;i Þ ¼ MIN @  c3 ðT CWS;i  T wb;i Þ  Q_ rej;i C
A
m_
b1 m_ ew1 þ b2 m_ ea1 i¼1 1 þ c2 m_ w;i a;i
c1 m_ cw3
¼  c3 ðT CWS  T wb;i Þ ð25Þ ð27Þ
1 þ c2 mm__ wa
Since it is a non-linear unconstraint optimization problem,
where c1 = b1, c2 = b1/b2 and c3 = e1. These three parame- several existing algorithms can be used to find the solution.
ters can be determined either by manufacturers’ data or by Here, Levenberg–Marquardt method [20] is implemented,
real time experimental data. which uses a search direction between the Gauss–Newton
Forced convection heat transfer is very complicated and and the steepest descent direction by solving the linear set
influenced by many factors. The approach of combining of equations:
the property factors and geometric factors into the con- T
ðJ ðuk Þ J ðuk Þ þ kk IÞd k ¼ J ðuk ÞF ðuk Þ ð28Þ
stants (characteristic parameters) was used in Eq. (25).
T
Compare with the existing cooling tower model, Eq. (25) where uk is the kth iteration value of u = [c1, c2, c3] ;
is characterized by fewer characteristic parameters and F i ðuk Þ ¼ F i ð½c1 ; c2 ; c3  Þ;
T
ð29aÞ
simplicity.
T
Comparing the existing model mentioned above, the F ðuk Þ ¼ ½F 1 ðuk Þ; F 2 ðuk Þ    ; F N ðuk Þ ; ð29bÞ
proposed model release the control variable, mass flow rate 2 oF oF oF
3
1 1 1
oc oc2 oc3
of water and air, from the lumped parameter MeM and 6 oF 1 7
6 2 oF 2 oF 27
NTU, thus, the iterative computation can be avoided when 6 oc1 oc2 oc3 7
the m_ w and m_ a change. J ðuk Þ ¼ 6
6 . ..
7;
.. 7 ð29cÞ
6 .. . . 7
By dimensional analysis, the crossflow arrangement 4 5
oF N oF N oF N
cooling tower can be hypothetically considered as a coun- oc1 oc2 oc3
terflow arrangement with the same terminal temperatures
[18,19]. Therefore, this model can also be applied in cross- ukþ1 ¼ uk þ d k ð29dÞ
flow cooling tower, with the equivalent water mass flow In the equations, kk controls both magnitude and direction
rate m_ w;e ¼ km_ w . For a special cooling tower, the convert of dk. When kk is zero, the direction dk is identical to that of
 G.-Y. Jin et al. / Energy Conversion and Management 48 (2007) 355–365
the Gauss–Newton method. If kk tends to infinity, dk tends
towards a vector of zeros and a steepest descent direction.
This makes it less effective than that of Gauss–Newton
search direction but more robust.
The whole procedure of model identification is given as
following:
Step 1. Obtain cooling tower operating points from either
manufacturers’ data or real time experimental data.
Step 2. Initialize u0 and k0, calculate F(u0) and J(u0) by
Eqs. (29b) and (29c).
Step 3. Determine initial search direction d0 by Eq. (28).
Step 4. Compute the F(uk) by Eq. (27) and determine the
searching direction of the next step by Eqs. (28)
and (29).
Step 5. Stop computation when F(uk) or F(uk)  F(uk+1) is
less than a pre-specified small number (usually in
the range of 1 · 106–1 · 105). Otherwise, return
to Step 4.
3.3. Comparison with existing models
3.3.1. Merkel’s model
1. Merkel method, employing a numerical method to
determine tower characteristic MeM, does not accurately
represent the physics of heat and mass transfer process
in the cooling tower fill.
2. The iterative computation is involved.
3. The geometric specification of cooling tower is required.
3.3.2. Effectiveness-NTU model
1. The value of UA, which varies with the changing mass
flow rate of water and air, is difficult to determine and
requires geometric specification of cooling tower.
2. Initial guess of outlet air condition and iterative computa-
tion are needed in order to find true outlet air conditions.
3.3.3. Stoecker’s method
1. It is an entirely empirical model and incapable of consid-
ering the variations of airflow and water flow rates.
Extending this model to variable airflow rate or water
flow rate requires more coefficients, which increases
the complexity for model identification.
Table 1
Comparison of different cooling tower models
Cooling tower model Stoecker’s Merkel’s
Number of parameters 9 (c1–c9) 1 (MeM)
Variable mass flow rate No Yes
Geometric data No Yes
Iterative computation No Yes
Modeling technique Empirical Physical
Model application Simulation Simulation and design
361
2. The model only uses curve-fitting technique to approxi-
mate the real experimental data. It might work well
under the predefined conditions within the experimental
data region. When the assumption is no longer true and
the operating region is beyond the experimental data,
the model cannot reflect the real performances of the
cooling tower.
3.3.4. Proposed model
1. The proposed model lumps the geometric information
as constants and only considers the inlet and outlet con-
ditions of cooling towers, therefore it is incapable of
providing useful information for cooling tower designs
and retrofits.
2. Due to the global uniqueness of the characteristic
parameters and the simple procedure to determine them,
the model is more desirable for real time optimization of
operating cooling towers.
In summary, the advantages of the proposed model are
simple, flexible, relative accurate, and easy for engineering
applications. It focuses on the characteristics of exiting
cooling tower and the relationship between heat rejection
rates and mass flow rates under different circumstances.
Table 1 summarizes the comparison of the proposed new
model and existing model mentioned above.
4. Model validation
In order to validate the proposed cooling tower model, a
cooling tower operating in Grand Hyatt Singapore hotel
HVAC system with following specifications is used:
1. Type: induced draft, counter-flow.
2. Tower dimension: 4.94 m L · 4.94 m W · 3.96 m H.
3. Nominal water circulation flow rate: 80.44 kg/s (1275
gallon/min).
4. Designed entering water temperature: 35.00 C
(95 F).
5. Designed leaving water temperature: 29.44 C
(85 F).
6. Designed wet-bulb temperature: 26.67 C (80 F).
7. Heat rejection at design condition: 2461.83 kW
(700 ton).
8. Nominal fan power: 15 kW.
E-NTU New model
1 (UA) 3 (c1  c3)
Yes Yes
No No
Yes No
Hybrid Hybrid
Design and control Control and optimization
362
9. Water distribution system: low pressure down spray.
10. Fan motor location: outside air discharge steam.
Both the cooling tower fan and the condenser water
pump are equipped with Variable Speed Drive to regulate
the air and water speed during part-load condition. The
air speed through the cooling tower can be estimated by
the frequency of the motor driven the fan. A flow meter
installed at the condenser water loop shows the water flow
through the cooling tower. The dry-bulb and wet-bulb tem-
perature of ambient environment are measured by sensors
in the weather station, which has been set up near the
HVAC plant. Two accurate temperature sensors are
installed to monitor the water temperatures entering and
leaving the cooling towers. With the temperature differ-
ences and water flow rates, heat rejection rates of the cool-
ing tower can be determined as the real operating data. All
these variables are recorded by Building Automation Sys-
tem (BAS) in the hotel.
According to the settings of BAS in Grand Hyatt Singa-
pore hotel, the sampling rate of each variable is one min-
utes. Therefore, there are totally 1440 values for each
variable in one day. In order to quantitatively show the
performances of the model prediction, an error index,
Root-Mean-Square of Relative Error (RMSRE), is
adopted:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uP  
u n Dreal Dcalc 2
t i¼1 Dreal
RMSRE ¼ ð30Þ
n ruary 03 2004 is used to predict the cooling tower perfor-
A whole day’s data (February 03 2004) is used for param-
eter identification of the proposed cooling tower model
2000
1800
1600
calculated heat rejection rate (kW)
1400
1200
1000
800
600
400
200
0
0 200
G.-Y. Jin et al. / Energy Conversion and Management 48 (2007) 355–365
and the results are showed in Fig. 4. The x-axis is the real
data of heat rejection rate measured on site, while the y-
axis is the calculated data predicted by the cooling tower
model.
There are three lines in the figure to indicate perfor-
mance of cooling tower model. The line in the middle
means the ideal case, where the cooling tower model
exactly predicts the heat rejection rate of operating compo-
nent. The other two lines give the boundary of ±10% devi-
ation from the ideal case. If the data points fall in margin
of 10% error, it means that the model under-predict the
heat rejection rate of the cooling tower. When the data
points fall in margin of +10% error, the model over-predict
the real heat rejection rate. Using Levenberg–Marquardt
method, the parameters are estimated as c1 = 4.43;
c2 = 1.12 and c3 = 1.11. The RMSRE for this model iden-
tification of 1440 data points is 0.060 and this is acceptable
for control and optimization purposes.
This model is then used to predict the cooling tower per-
formance of very next day (February 04 2004). Fig. 5
shows that the results are good enough for the model
validation.
The RMSRE for this model validation of 1440 data
points is 0.056 (c1 = 4.43; c2 = 1.12; c3 = 1.11). If all these
1440 data are used to identify the new cooling tower model,
the RMSRE for the model identification is 0.054
(c1 = 5.73; c2 = 1.61; c3 = 1.09). There is no big difference
on the prediction accuracies of these two models.
Secondly, the original model identified by data on Feb-
mance after seven days (February 10 2004). Fig. 6 shows
the results of the model validation.
+10% error
-10% error
400 600 800 1000 1200 1400 1600 1800 2000
real heat rejection rate (kW)
Fig. 4. Model identification on February 03 2004.
 G.-Y. Jin et al. / Energy Conversion and Management 48 (2007) 355–365 363

2000

1800
+10% error

1600

calculated heat rejection rate (kW) 1400

-10% error
1200

1000

800

600

400

200

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
real heat rejection rate (kW)

Fig. 5. Model validation on February 04 2004.

2000

1800
+10% error

1600
calculated heat rejection rate (kW)

1400
-10% error
1200

1000

800

600

400

200

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
real heat rejection rate (kW)

Fig. 6. Model validation on February 10 2004.

The RMSRE for this model validation of 1440 data some avoidable errors are induced for the performance pre-
points is 0.082. (c1 = 4.43; c2 = 1.12; c3 = 1.11) If all these diction by original model.
1440 data are used to identify the new cooling tower model, Finally, the original model identified by data on Febru-
the RMSRE for the model identification is 0.063 ary 03 2004 is used to predict the cooling tower perfor-
(c1 = 4.71; c2 = 1.21; c3 = 1.09). Although the results of mance one month away (March 03 2004). Fig. 7 shows
the model validation are tolerant for engineering purpose, the results of the model validation.
364 G.-Y. Jin et al. / Energy Conversion and Management 48 (2007) 355–365

2000

1800
+10% error

1600
calculated heat rejection rate (kW)
1400
-10% error
1200

1000

800

600

400

200

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
real heat rejection rate (kW)

Fig. 7. Model validation on March 03 2004.

The RMSRE for this model validation of 1440 data
points is 0.143 (c1 = 4.43; c2 = 1.12; c3 = 1.11). The results
of the model may not be tolerated for accurate control and
optimization. For the majority of data points, the model
over-predicts the performance of the cooling tower.
For direct contact cooling tower used for testing, the
water quality and effective contact area is time varying
because the pollution from circumstance. The reason
why the predicted performance is not the same for differ-
ent testing time is due to the decreasing effective contact
area in cooling tower. And the change of water quality
either results in the fall of the heat exchange efficiency.
As the time passing by, the performance of cooling tower
becomes worse. Therefore, the predicted cooling tower
performance by the identified parameters in February 04
2004 is better than the real performance in February 10
2004 and much better than the real performance in March
03 2004.

2000

1800
+10% error

1600
calculated heat rejection rate (kW)

1400
-10% error
1200

1000

800

600

400

200

0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
real heat rejection rate (kW)

Fig. 8. New model identification on March 03 2004.

 G.-Y. Jin et al. / Energy Conversion and Management 48 (2007) 355–365
Since the characteristics of cooling tower are slow time-
varying, we need to identify the characteristic parameters
of the cooling tower operating in March 03 2004. With
the new parameters c1 = 1.53; c2 = 0.62 and c3 = 1.29 can
perfectly predict the cooling tower performance with
RMSRE 0.048, as shown in Fig. 8.
According to the validation of the proposed cooling
tower model, the model with new parameters is accurate
enough for the process control and optimization
(RMSRE < 0.1). For long period of time, the characteristic
parameter should be updated periodically to accurately
predict the cooling tower performance since the character-
istics of the cooling tower is time varying. Generally, the
characteristic parameters can be used for some time inter-
val after it is identified, and the time interval depends on
the circumstance condition of operating cooling tower.
After this time interval, the original cooling tower model
with previous characteristic parameters might not be accu-
rate enough and should be updated by new performance
data.
5. Conclusions
A simple cooling tower model has been presented in this
paper. The proposed model is based on heat resistance and
energy balance principles. It induced only three empirical
parameters whose values are determined by either manu-
facturers’ catalog data or real operating data. Unlike other
existing cooling tower models, this model captures the geo-
metric characteristics without requiring geometric specifi-
cations and no iterative computation and initial guess are
required to determine the parameters. Consequently,
online determination of the model parameters becomes
practical and simple. The real operating data from a
HVAC system of a commercial hotel is used to identify
and validate the proposed the model. The slow time-vary-
ing characteristics of the cooling tower under test can be
specified by altering the model parameters and accurately
predict the cooling tower performance. As time evolves,
the parameters of cooling tower should be constantly
updated to maintain the prediction accuracy.
The significance of the developed model is that it can be
easily applied to real-time HVAC system optimization
where set points of each energy consumption devices in
condenser water loop are related to the model. Therefore,
365
the cooling tower model is the constraint condition in the
course of energy minimization of the loop. Further work
in the area will concentrate on the development of optimi-
zation methods and testing them on the real industrial
plant based on the proposed model.
References
[1] ASHRAE. ASHRAE handbook—2000 HVAC systems and equip-
ment. American Society of Heating, Refrigerating, and Air-Condi-
tioning Engineers; 2000.
[2] Rocky Mountain Institute Research Associates, Commercial Space
Cooling and Air Handling, Technology Atlas, E Source, 1995.
[3] Burger R. Cooling tower technology—maintenance, updating and
rebuilding. Fairmont Press; 1995.
[4] Merkel F. Verdunstungskuhlung, 275. Berlin: VDI Forschungsarbe-
iten; 1925.
[5] Stoecker WF. Procedures for simulating the performance of compo-
nents and systems for energy calculations. Atlanta: ASHRAE; 1976.
[6] Braun JE, Klein SA, Mitchell JW. Effectiveness models for cooling
towers and cooling coils. ASHRAE Trans 1989;95(2):164–74.
[7] Bernier MA. Cooling tower performance: theory and experiments.
ASHRAE Trans 1994;100(2):114–21.
[8] Michel AB. Thermal performance of cooling towers. ASHRAE J
1995;37(4):56–61.
[9] Mick S. Take it to the limit . . . or just halfway? ASHRAE J
1998;40(7):32–9.
[10] Soylemez MS. Theoretical and experimental analyses of cooling
towers. ASHRAE Trans 1999;105(1):330–7.
[11] Hasan A, Sirén K. Theoretical and computational analysis of closed
wet cooling towers and its applications in cooling of buildings. Energy
Build 2002;34:477–86.
[12] ASHRAE. HVAC 1 toolkit: a toolkit for primary HVAC system
energy calculation. American Society of Heating, Refrigerating, and
Air-Conditioning Engineers; 1999.
[13] Kloppers C, Kröger G. Cooling tower performance evaluation:
Merkel, Poppe, and e-NTU methods of analysis. J Eng Gas Turb
Power 2005;127:1–7.
[14] Arora CP. Refrigeration and air conditioning. New Delhi: Tata
McGraw-Hill Publishing Company Limited; 2000.
[15] Burghardt DM. Engineering thermodynamics with applica-
tions. HarperCollins Publishers Inc.; 1986.
[16] Waldram JR. The theory of thermodynamics. New York: Cam-
bridge University Press; 1985.
[17] Perry RL. Thermodynamics for engineering technology. Califor-
nia: Wadsworth Inc.; 1984.
[18] Holman JP. Heat transfer. 7th ed. Singapore: McGraw-Hill Book
Company; 1992.
[19] White FM. Fluid mechanics. 2nd ed. Singapore: McGraw-Hill Book
Company.; 1986.
[20] Nocedal J, Wright SJ. Numerical optimization. New York: Springer
Verlag, Inc.; 1999.