Accepted Manuscript

Title: Modeling and multi-objective optimization of an M-cycle cross-flow

indirect evaporative cooler using the GMDH type neural network

Author: Ali Sohani, Hoseyn Sayyaadi, Sina Hoseinpoori

PII: S0140-7007(16)30116-5
DOI: http://dx.doi.org/doi: 10.1016/j.ijrefrig.2016.05.011
Reference: JIJR 3337

To appear in: International Journal of Refrigeration

Received date: 14-1-2016

Revised date: 7-5-2016
Accepted date: 16-5-2016

Please cite this article as: Ali Sohani, Hoseyn Sayyaadi, Sina Hoseinpoori, Modeling and multi-
objective optimization of an M-cycle cross-flow indirect evaporative cooler using the GMDH
type neural network, International Journal of Refrigeration (2016), http://dx.doi.org/doi:
10.1016/j.ijrefrig.2016.05.011.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will
undergo copyediting, typesetting, and review of the resulting proof before it is published in its
final form. Please note that during the production process errors may be discovered which could
affect the content, and all legal disclaimers that apply to the journal pertain.
 Modeling and Multi-Objective Optimization of an M-Cycle Cross-flow

Indirect Evaporative Cooler Using the GMDH Type Neural Network

Ali Sohani, Hoseyn Sayyaadi*1, Sina Hoseinpoori

Faculty of Mechanical Engineering-Energy Division, K.N. Toosi University of Technology, P.O. Box: 19395-1999,

No, 15-19, Pardis St., Mollasadra Ave., Vanak Sq., Tehran 1999 143344, Iran

Highlights

 CrFIEC was modeled using the GMDH neural network.

 The GMDH model was used to generate predicting correlations for Tout.
 The model was validated using test data.
 The GMDH model was employed for optimization of a prototype CrFIEC.
 Feature of the proposed system were optimized for various worldwide weathers.

Abstract

A model was presented to determine product air properties of dew-point indirect evaporative coolers with cross flow

heat exchanger, M-cycle CrFIEC. In this regard, a most powerful statistical method known as the group method of

data handling-type neural network (GMDH) was employed. Then the developed GMDH model was implemented

for multi-objective optimization of a prototype CrFIEC and the average annual values of coefficient of performance

(COP) and cooling capacity (CC) were maximized, simultaneously, while working to air ratio (WAR) and inlet air

velocity was decision variables of optimization. Accordingly, features of the proposed system were optimized at

twelve diverse climates of the world based on Koppen–Geiger’s classification. Results implied that the optimized

inlet air velocity for all climates varied between 1.796 and 1.957 m.s-1, while the optimum WAR was 0.318 for ‘A’

class cities. Moreover, the mean values of the COP and CC were improved 8.1 % and 6.9%, respectively.

*Corresponding author
Email: [email protected] , [email protected], Tel: +98-21-8406-3212, Fax:+98-21-8867-4748

1 of 60
Page 1 of 59
 Keywords: Cross-flow heat and mass exchanger; Dew-point evaporative cooling; Group method of data handling;

Koppen–Geiger climatic classification; M-cycle cross-flow indirect evaporative cooler; Multi-objective optimization

Nomenclature T Temperature (ᵒC)

Maximum allowable neuron numbers in
Anon V Velocity (m.s-1)
each GMDH network layer
cp
Specific heat of air at constant pressure
W AR Working to inlet air ratio
(J.kg-1.K-1)
GMDH internal layers neurons
C1 Pareto optimal front equation coefficient x
outputs
Pareto optimal front equation coefficient
C2 X GMDH neuron input
(kW)-C1
CC Cooling capacity (kW) Y Actual value
Decision making parameter in TOPSIS
C li method for the ith answer in Pareto optimal Yˆ GMDH prediction value
front
COP Coefficient of performance Y Average of Actual values
 The ith answer in Pareto optimal front
di Scripts
distance from the ideal point
 The ith answer in Pareto optimal front
di a ir Air
distance from the non-ideal point
EPC Electrical power consumption (W) ave Average
F Normalized objective functions base Base condition
g GMDH neuron polynomial coefficient board Electrical board
GP GMDH neuron output EF Effective factors
h Dry channel height (m) fa n Fan
L Channel length (m) in Inlet air
* Dimensionless length of the channel la te n t Latent
L


m
Mass flow rate (kg.s-1) out Outlet (product) air
M SE Mean square error s e n s ib le Sensible
N um Number th e r m o s ta t Thermostat
O bj Objective function wa Working air
O pt ans Optimized answer Greek symbols
O th e r a n s Other answers  Density (kg.m-3)
p o ly d e g GMDH polynomials degree P Pressure drop (Pa)
P Number of inputs for each GMDH neuron  Relative humidity (%)
Absolute humidity (
Q Volumetric flow rate (m3.hr-1)  1
k g m o is tu r e . k g d r y a ir )

R
2
Coefficient of determination Superscripts
SH Fan static head (Pa) Id e a l Ideal
SSE Sum square error N o n  Id e a l Non ideal

2 of 60
Page 2 of 59
 1. Introduction

Achieving comfort has always been one of the most important human’s goals. Thermal comfort

is one of the most remarkable kinds of desired human’s comforts. It has management benefit

besides individual satisfaction since the thermal comfort level significantly affects staff

performance efficiency. HVAC systems are grouped of equipment that provides thermal

comfort. Indirect evaporative coolers, IECs, are one of the available HVAC systems for cooling.

In the same cooling capacity, IECs produce less CO2 and consume less electric power in

comparison to direct expansion, DX, systems, so IECs take account into one of the best

alternatives for DX ones, especially in low latent load ratio applications.

In recent years, the performance of IECs in comparison to early prototypes has been improved,

significantly. The best and the most effective improvements is the Maisotsenko cycle (M-cycle)

technology. As described in section 2, the main idea in the M-cycle IECs is using a part of the

main flow as the working air. It allows the main stream to be cooled to temperatures below the

wet-bulb temperature. Based on the report of the national renewable energy lab (NREL), the M-

cycle IECs are 80% more efficient than the conventional DX systems [1]. Among different types

of M-cycle IECs, the M-cycle cross flow indirect evaporative cooling type (M-cycle CrFIECs)

was studied in this paper.

Researchers presented numerical or experimental models and sensitivity analysis of M-cycle

CrFIECs in the past several years. Jradi and Riffat [2] conducted a two-dimensional numerical

modeling of M-cycle cross flow indirect evaporative cooling systems (M-cycle CrFIECs). They

implemented their numerical model to optimize working to inlet air ratio (WAR) of an M-cycle

IEC system with 500 mm channel length and 5 mm channel height at a specific inlet air property.

The corresponding wet-bulb and dew-point efficiencies were 112% and 78%, respectively. In

3 of 60
Page 3 of 59
addition, their experimental studies proved that the investigated IEC was capable of providing

comfort condition and cooling capacity for buildings, especially in hot climates.

Zhan et al. [3] implemented numerical model to compare the M-cycle counter-flow regenerative

and M-cycle cross-flow IECs. They evaluated the individual effect of operating parameters on

performance factors, including the coefficient of performance (COP) and cooling capacity (CC).

The results implied that under the evaluated conditions, counter regenerative cooler have higher

cooling efficiencies (15% and 23% higher percentage for dew-point and wet-bulb efficiencies,

respectively), but COP of the cross type was 10% higher. They also declared that among

operational parameters, the flow rate and working to inlet air ratio, WAR, are appropriate

parameters for optimization.

Pandelidis et al. [4] studied the performance of M-cycles in the cross air streams at two different

configurations using a two-dimensional heat and mass transfer numerical model. In the first

configuration, the main air and working air flowed in the same directions (the conventional type

called HMX1) and in the second one, the main and working air flowed in opposite directions

(HMX2). They studied the effects of inlet parameters, including temperature and inlet velocity

and concluded that the HMX1 had a higher efficiency due to more homogeneous temperature

distribution.

Weerts [5] examined the M-cycles as an option for common systems in the National Snow and

Ice Data Center (NSIDC). He reported that this substitution reduced the cooling energy

consumption about 90%. In another research, Zhao et al. [6] performed a comparison between

different materials for IECs heat, and mass transfers sheets. They declared that the water

retaining capacity, thermal properties of very thin sheets material including porosity and thermal

conductivity have insignificant effects on heat and mass transfer process. Zhan et al. [7] have

4 of 60
Page 4 of 59
developed a numerical model for M-cycle CrFIEC. They found that the cooling efficiency can be

increased up to 16.7% in comparison to conventional cross-flow heat and mass exchangers that

flows are supplied separately.

As it can be seen in almost all studies, numerical models were used. This model involves

complicated solution of numbers of differential equations. This process is not only very

complicated but also its computation time is too high.

Employing soft computing and statistical tools (SCST) is another option to be used instead of

numerical models to predict performance of M-cycles. SCST requires only basic knowledge

regarding parameters that affect the system performance as well as enough extracted data from

the system operation obtained from experimental or numerical studies. Artificial neural network

(ANN) genetic programming (GP) multiple linear regression (MLR) adaptive neuro-fuzzy

inference system (ANFIS) fuzzy inference system (FIS) and response surface methodology

(RSM) are different types of SCST approaches. There are two studies in the field of modeling

IECs using SCST. In one of them, Kiran and Rajput [8] implemented ANN, FIS, and ANFIS for

modeling the cross-flow systems that their main and working air was separated. According to

their results, ANN was easier and had higher precision compared to other two approaches. In the

other one, RSM was employed by Pandelidis and Anisimov [9] in order to present simplified

model for predicting the performance of the M-cycle CrFIEC performance. Both experimental

and validated numerical data were used. They analyzed the individual impact of independent

input parameters like inlet air temperature and relative humidity on COP and CC. The results

demonstrated that a single-objective optimization with either COP or CC objective, leads

unsatisfactory results on the other criterion. In other words, if the COP is maximized, the CC will

5 of 60
Page 5 of 59
be set to much undesired value; hence, a multi-objective optimization with both CC and COP

objectives was suggested.

In this paper, first, all affective parameters of the M-cycle CrFIEC were identified and be used to

develop SCST model based on data sets reported by others [2-4] to obtain conditions of outlet air

of a M-cycle CrFIEC. Among SCST approaches, the group method of data handling-type neural

network (GMDH) was employed as it has both advantage of ANNs and GPs. Its high precision is

comparable and even more than ANNs and it present expressions for predicted variables like as

what performs by GP. Therefore, its results are repeatable by others through the use of

expressions without any needs for generating the neural network structure. The GMDH was

employed to predict outlet temperature of M-cycle CrFIECs based on five input parameters,

including inlet air temperature, inlet air velocity, working air ration, and a non-dimensional

parameter of channel dimensions. Then, the GMDH model was used to model and optimizing a

prototype M-cycle CrFIEC system called Coolerado M-50 unit. In this regard, inlet air velocity

and working air ration (WAR) were optimized in a multi-objective optimization process while

the coefficient of performance (COP) and average annual cooling capacity, (CC) were two

objective functions. To avoid changes on the geometrical structure of the proposed model,

dimensional parameters of channel were not considered among decision variables of

optimization. Several cities were selected to be represented for every world based on Koppen-

Geiger climate classification, and the multi-objective optimization was performed on the

prototype system in each location to find an optimal point of the Coolerado M-50 unit for that

specific climatic conditions. Hence, optimal operating features of the Coolerado M-50-unit

system in all categories of Koppen-Geiger climate classification that are eligible for evaporative

cooling were presented. It was conducted using non-dominated sorting genetic algorithm

6 of 60
Page 6 of 59
(NSGA-II) and in each area a Pareto optimal frontier (POF) was obtained and the final optimal

solution was selected among optimal solutions located on the POF using a systematic decision

making method called the technique for order preference by similarity to ideal situation

(TOPSIS). In summary, contributions of this paper includes presenting the correlation to predicts

outlet temperature of M-cycle CrFIEC systems using the GMDH tools that trained by empirical

and numerical data on the one hand, and presenting optimal operation features of a prototype

system at diverse climatic condition on the other hand.

2. Working principle of M-Cycle CrFIEC

Consider 3 thin plates stacked to form 2 channels as illustrated in Fig. (1a)-(1c). The bottom side

of the middle plate located on channel ‘B’ is covered with a thin layer of water like the other side

of this channel. Air enters and passes through the channel ‘A’ and heat transfer between air and

water layer produces the vapor in channel ‘B’ and cooling in the channel ‘A’. Channel ‘A'’

surfaces are dry and there is no contact between air and water so, this air (product air) is cooled

without increasing in its absolute humidity. The system requires two air flows to operate. The air

which is cooled in the channel ‘A’ without increasing in its absolute humidity is called the main

air stream. The second stream that passes through the channel ‘B’ to evaporate the water is called

the working air stream. According to the properties of air streams, three states are possible in the

dry channels: 1- Non-condensation state (Fig. (1a)), 2- Partially condensation state ((Fig. (1b))

and 3-Totally condensation state (Fig. (1c)) [10, 11]. The main air supply source is usually 100%

ambient fresh air or a mixture of the ambient and room return airs. Working air is provided in

different manners, including 100% return air, a part of the main stream and etc. Since the main

air is colder and drier than the other steams, utilizing it as the working air improves the system

performance, significantly.

7 of 60
Page 7 of 59
 {Insert Fig. (1) here}

Based on the main and working flow's arrangement, IECs are divided into 3 groups of parallel-

flow, counter-flow and cross-flow as illustrated Fig. (2a)-(2c), respectively (These figures were

drawn in non-condensation state). Cross-flow and counter-flow types are more common than

parallel-flow. A special type of cross-flow IEC called M-cycle cross-flow IEC (M-cycle

CrFIEC) is the subject of this research. Schematic view of this system was depicted in Fig. (3).

{Insert Fig. (2) here}

{Insert Fig. (3) here}

In this type of cooler, the working air flows in the wet and dry channels cross each other. The

main air is cooled due to evaporation of water in a wet channel. Some dry channels have dead

end and contain several holes to transfer a part of pre-cooled air to the wet channel gradually.

The mentioned process to transfer a part of main air from dry to wet channel is called

Maisotsenko-cycle (M-cycle) that is a special type of indirect evaporative cooling system that

suggested and patented by Valery Maisotsenko [12-14]. This special type of IECs is capable of

decrease the product air temperature to the equivalent dew-point temperature of the inlet air in

theory. Despite conventional types of IECs in which the inlet air can be cooled to its wet-bulb

temperature [10], in the M-cycle IECs, sub-wet bulb temperatures can be achieved and in theory

the main air can be cooled to the dew-point temperature of its entrance condition; hence, M-

cycle IECs are also called as dew-point coolers. It should be mentioned that, in M-cycle IECs,

there is no condensation. More detail regarding, the performance of this system and capability to

reach the outlet temperature to the dew-point temperature of the inlet air was addressed in [15-

17].

8 of 60
Page 8 of 59
 3. Group method of data handling-type neural network, GMDH

Group method of data handling-type neural network (GMDH) is a predictive approach belongs

to artificial neural network, ANN family. Unlike other common types of ANN, in this type, the

network is a self-organizer which means, there is no need to determine either the number of

neurons and layers or the net and transfer functions; so that these parameters are determined

automatically. Furthermore, the GMDH model provides a number of mathematical equations

instead of ANN conventional matrix structure; so the calculation speed is increased, time is

saved and the output is more user-friend. Hence, anyone can use correlations without any need

for regeneration of the ANN structure. Neuron structure is special in this method. In this

structure, each neuron output is calculated according to its inputs from Eq. (1). Inputs of the first

layer are input variables of the model and following layers can accept only outputs of the

previous layer, or a combination of both outputs of the previous layer and the first layer’s inputs

[18].

P P P P P P

Yˆ  g 0   gkX k
  g km X k X m
  g km l X k X m
X l
 (1)
k 1 k 1 m 1 k 1 m 1 l 1

Eq. (1) is the discrete form of Voltera functional series known as Kolomogorov-Gabor

polynomial. In Eq. (1), X represents inputs of each neuron, and P is the index for the number of

input data to a neuron. Moreover, g stands for coefficients of Eq. (1) that will be determined in

a way that minimizes mean square error for the neuron (Eq. (2)) [19].

N u m d a ta 2
 
^

  Yq  Yq  (2)
SSE q 1  
M SE  
N u m d a ta N u m d a ta

9 of 60
Page 9 of 59
 ^

In Eq. (2), Y q and Y q , stand for actual values of the trained data and predicted output,

respectively. It should be mentioned that in GMDH method, same power polynomials have to

be chosen for all layers. The polynomial’s power is the number of variable that are multiplied

each other in Eq. (1). For example, in cases that the powers of polynomial are 1, 2, and 3, in the

P P P P

right-hand side of Eq. (1), we have g 0   g k X k , g 0   g k X k

  g km X k X m
, and
k 1 k 1 k 1 m 1

P P P P P P

g0   gkX k
  g km X k X m
  g km l X k X m
X l , respectively. For modeling using
k 1 k 1 m 1 k 1 m 1 l 1

GMDH, first, the number of inputs for each neuron, polynomial power (2 or 3 for our case) and

input sources of layers after the first layer are decided. Then, the design process begins from the

first layer and goes on. All possible combinations of allowable inputs (all possible neurons) are

considered. Then polynomial coefficients are determined using one of the available minimizing

methods such as singular value decomposition (with training data). Then, neurons that have

better external criterions (for testing data) are kept, and others are removed (The input data for

development of the model was divided into training and testing groups). If the external criterion

for layer’s best neuron surpasses the stopping criterion, network design is completed and the

polynomial expression of the best neuron of the last layer is introduced as the mathematical

prediction function; if not, the next layer will be generated, and this process goes on. The typical

flow chart of a GMDH-type neural network was illustrated in Fig. (4). More details regarding the

GMDH are found in [18-20].

{Insert Fig. (4) here}

10 of 60
Page 10 of 59
 4. Determination of the condition of the product air

If the air pressure is specified, two other properties of the moist air are required to specify the

state of the air in psychometrics. Dew-point systems cool the air without increasing its absolute

humidity; therefore, the outlet air absolute humidity is equal to inlet air absolute humidity;

therefore, we have:

 o u t   in
(3)

Since there is not another parameter that can be determined directly from the inlet air properties,

another parameter should be evaluated using an appropriate model. As discussed, developing the

SCSTs (such as GMDH) require extracting enough experimental or numerical data. For this

reason, the second property should be selected among parameters, which sufficient data have

been reported. With this regard, temperature was selected as the second property to specify the

state of air at the outlet of CrFIECs; hence, an appropriate GMDH model for determination of

the outlet temperature of CrFIECs was developed in the next section.

4.1. Product air temperature predicting using the GMDH method

4.1.1. Assumptions

Following assumptions were considered in this analysis:

1. The heat and mass transfer processes were steady state.

2. The heat loss to the environment was neglected.

3. Due to small thickness of plates, the effect of material and its porosity were neglected [6].

11 of 60
Page 11 of 59
 4. Thickness of the water layer was assumed to wet plates’ surface properly, and its

thickness can be ignored [6, 21].

5. Changes in the efficiency of the cooling system due to the variation of the water

temperature in the wet channel were ignored [22].

4.1.2. Effective factors

According to the previous researches [2-4], the inlet air temperature, relative humidity and

velocity as well as the dimensionless length of the channel and working air ratio (WAR) were

considered as effective factors on the system product air temperature. The dimensionless length

of the channel and working air ratio were defined by Eqs. (4) and (5), respectively.

* L
L  (4)
h

m wa
m out
W AR  1 (5)
m in
m in

In Eq. (4), L and h are channel length and dry channel height, respectively. The correlation

between the dimensionless length of the channel and the outlet air temperature was also obtained

and reported in some published literatures such as [7].

4.1.3. Extracting training and validating data

Determining the product air temperature using the GMDH method needs training the GMDH

neural network as well as evaluating the constructed network. Hence, training and validating data

are required. In this regard, 1343 series of data were gathered from previous experimental and

12 of 60
Page 12 of 59
numerical researches as addressed in Table (1). In addition, the variation range of those training

and validating data were indicated in Table (2).

{Insert Table (1) here}

{Insert Table (2) here}

4.1.4. Developing the best GMDH network for predicting the product temperature

The GMDH polynomial neural network toolbox created by Jacobsons [23] was used in

MATLAB R2013a software to model the M-cycle CrFIEC. The corrected Akaike’s information

criterion [24, 25] was used as the external criterion. Choosing different values for the number of

inputs for each neuron, input source for the neurons after the first layer, maximum allowable

number of neurons in each layer and polynomial’s power, various configurations of the GMDH

networks were constructed and tested. Thus, number of inputs for each neuron was set as 2 or 3.

In addition, inputs of the layer after the first layer were outputs of previous layer or the

combination of the outputs of previous layer and the effective input parameters of the network.

The allowable number of neurons were minimum 1 and maximum A n o n m a x , where A n o n m a x

is:

N u m EF
A n o n m ax  ( ) (6)
P o ly P o w

Polynomial’s power of neurons was set to 2 or 3. Using various combinations of setting

parameters, the number of 80 distinct GMDH network were constructed and tested. A sample for

these 80 GMDH networks was indicated in Table (3).

{Insert Table (3) here}

13 of 60
Page 13 of 59
Three error estimation parameters, including MSE (Eq. (2)), SSE and coefficient of

determination, R2 (Eq. (7)) were used for evaluation of the 80 GMDH networks, and the best

structure with least error was selected as the final GMDH model for M-cycle CrFIECs.

N u m d a ta 2
 
^

  Yi  Yi  (7)
i 1  

2
R 2
N u m d a ta
 
_

  Yi  Y 
i 1  

The specifications of the final selected GMDH network were indicated in Table (3) with bold

data.

The mathematical expressions for determination of the outlet temperature of the CrFIECs based

on the five input data (inlet temperature and humidity, WAR, L*) were presented by Eqs. (8a)-

(8g) in. It should be mentioned that in order to achieve the most accurate prediction, all digits of

coefficients of the given polynomials must be applied.

*
x  1 0 3 .3 0 2 0 8 2 4 6 2 2 9 5  0 .1 8 3 4 5 5 8 8 6 3 1 4 1 8 9 L  2 .9 5 1 6 8 4 3 1 0 0 7 6   0 .4 3 7 5 2 4 8 7 8 6 5 8 7 5 9 T
1 in in
* *
 0 .0 1 2 2 1 8 1 1 3 6 6 9 6 1 8 2  L  0 .0 3 4 5 9 7 4 6 7 4 1 1 8 7 1 1 T L  0 .0 0 6 8 9 4 7 8 0 8 3 2 7 2 8 9 7 T 
in in in in
2 2 2
 0 .0 0 1 3 4 8 1 8 7 3 8 6 7 2 0 4 4 L h  0 .0 2 7 1 5 4 7 7 6 9 6 7 4 4 0 8   0 .0 6 6 9 6 1 6 9 9 6 1 6 5 3 3 9 T
in in
* *2
 0 .0 0 0 2 7 8 7 2 1 8 2 9 1 2 7 6 5 5 T  L  ( 2 .1 8 3 1 5 3 9 7 1 2 5 2 7 8 e  0 5 )  L
in in in
2 * *2
 ( 9 .9 0 0 1 8 6 9 0 1 9 1 6 8 8 e  0 5 )  L  (1 .4 7 8 6 4 6 8 7 7 8 0 6 2 9 e  0 5 ) T L
in in
2 2 *
 0 .0 0 0 2 6 8 3 7 7 7 2 4 6 0 4 4 6 1 T   0 .0 0 0 5 4 1 3 6 8 2 1 2 8 8 8 2 4 4 T L
in in in
2 *3
 0 .0 0 0 1 7 3 2 6 4 1 8 8 6 3 4 0 9 6 T   ( 4 .3 5 8 3 5 2 3 7 1 6 7 8 8 2 e  0 8 ) L
in in
3 2
 (1 .4 2 7 8 9 5 5 8 7 2 3 0 3 9 e  0 6 )   0 .0 0 1 2 6 1 8 3 0 5 7 6 0 4 7 4 4 T
in in

(8a)

14 of 60
Page 14 of 59
x  6 8 5 .9 5 6 8 6 4 1 8 1 3 5 7  3 3 .0 2 5 8 2 0 0 4 2 0 8 5 7 x  1 2 0 3 .4 0 9 0 5 9 0 0 4 5 6 W A R  3 1 0 .0 3 7 2 9 7 8 3 2 7 5 6 V
2 1 in
 5 0 .3 5 6 8 0 4 4 0 3 6 9 0 2 W A R x  8 .9 6 7 9 9 3 3 9 7 7 4 5 3 9 V x  3 8 3 .5 3 8 5 5 9 6 7 8 7 0 3 V W AR
1 in 1 in
2 2 2
 0 .4 7 1 7 9 8 8 5 4 4 3 9 3 8 2 x  4 2 7 .6 9 0 1 5 7 9 1 2 5 1 4 W A R  5 1 .2 3 4 4 7 6 1 4 3 8 4 6 9 V
1 in
2 2
 2 .3 7 0 6 9 8 9 2 7 6 9 1 0 1V W AR x  0 .8 4 9 5 2 6 9 4 0 0 9 6 2 2 9 W A R x  7 .8 8 4 2 5 5 9 3 1 6 9 0 0 4 W A R x
in 1 1 1
2 2 2
 0 .0 0 3 3 4 3 4 6 5 0 2 2 3 6 0 8 V x  1 0 3 .5 5 5 6 6 4 1 3 7 1 V W AR  1 .4 7 5 9 7 3 7 2 2 2 9 4 2 5 V x
in 1 in in 1
2 3 3
 4 7 .0 2 9 6 2 2 9 5 5 8 1 1 8 V W A R  0 .0 0 0 6 0 1 4 5 0 5 5 3 1 8 0 2 7 2 x  2 1 .9 2 3 5 8 8 0 4 9 1 2 7 6 W A R
in 1
3
 0 .0 0 0 6 6 8 5 8 2 4 5 1 2 6 5 8 3 5 V
in

(8b)
*
x  8 3 .7 6 0 0 6 3 7 0 7 9 7 8  4 .7 7 5 3 5 5 1 0 7 5 7 9 2 6 x  1 .3 7 5 7 6 6 3 0 8 6 5 9 6 1 L  1 .4 3 2 2 0 8 4 6 1 6 8 2 3 5 
3 2 in
* *
 0 .0 4 5 4 4 2 7 5 0 8 3 1 5 9 2 8 L x  0 .0 9 6 0 6 7 7 5 9 8 2 8 5 6 3 7  x  0 .0 0 0 9 1 0 1 6 2 0 6 5 1 0 0 9 8 9  L
2 in 2 in
2 2 2
 0 .2 1 8 5 4 3 5 1 9 7 9 9 2 7 5 x  0 .0 0 5 9 1 5 6 9 6 0 7 3 0 4 9 8 3 L h  0 .0 0 4 1 7 4 5 8 6 3 9 8 7 2 5 4 
2 in
* * 2
 0 .0 0 0 3 0 1 2 1 2 4 3 1 3 0 1 9 2 4  L x  0 .0 0 0 4 6 2 8 6 0 7 1 5 5 6 4 9 2 6 L x
in 2 2
*2 2
 0 .0 0 0 1 2 4 4 8 5 8 1 6 4 2 7 8 2 2 L x  0 .0 0 1 0 9 3 8 9 0 2 2 1 1 6 0 1 5  x
2 in 2
*2 2
 ( 2 .4 0 1 6 8 1 8 2 9 3 1 7 6 1 e  0 5 )  L  0 .0 0 0 1 0 4 1 0 2 5 2 2 3 7 3 5 1 8  x
in in 2
2 * 3
 ( 2 .8 2 4 7 1 6 2 7 1 3 4 7 e  0 5 )  L – 0 .0 0 2 9 3 0 6 8 3 7 6 7 5 4 3 0 2 x
in 2
*3 3
 ( 4 .8 2 3 4 5 1 2 4 4 2 4 0 6 8 e  0 6 ) L  ( 5 .9 1 5 8 0 3 7 0 6 2 7 8 9 8 e  0 6 ) 
in

(8c)

15 of 60
Page 15 of 59
 *
x  2 0 0 .5 0 8 8 4 8 3 6 0 3 8  7 .8 0 3 3 8 1 6 7 1 1 7 7 1 3 x  2 .1 9 3 5 7 4 7 0 0 3 8 4 9 6 L  5 4 .8 7 0 7 5 5 5 9 4 4 4 2 1V
4 2 in
* *
 0 .0 7 4 2 8 2 2 2 3 4 0 4 8 6 6 5 L x  2 .4 8 8 7 7 7 7 6 9 3 9 7 1 V x  0 .4 7 1 9 3 7 6 1 2 3 8 0 4 4 2 V L
2 in 2 in
2 *2 2
 0 .0 4 6 5 2 9 3 5 3 8 3 3 1 6 0 9 x  0 .0 0 6 3 5 5 8 6 6 5 1 9 8 7 6 7 L  2 . 2 3 7 9 0 5 6 1 6 3 6 0 2 4V
2 in
* * 2
 0 .0 1 5 0 8 3 0 6 7 1 3 9 2 3 4 V L x  0 .0 0 0 1 4 9 6 8 9 6 6 4 0 8 7 4 9 4 L x
in 2 2
*2 2
 0 .0 0 0 1 2 0 5 6 5 4 1 7 4 4 8 4 9 1 L x  0 .0 0 3 8 1 9 0 1 2 2 3 1 1 7 5 8 8 V x
2 in 2
*2 2
 0 .0 0 0 7 2 9 5 6 2 8 9 6 7 4 7 2 4 2 V L  0 .0 4 0 7 0 8 1 6 6 1 5 5 4 2 3 9 V x
in in 2
2* 3
 0 .0 1 2 9 8 7 1 1 4 6 5 6 3 9 3 8 V L  0 .0 0 0 1 5 3 4 7 7 4 4 3 5 0 4 7 4 2 x
in 2
*2 3
 ( 5 .1 8 6 9 8 0 2 9 3 1 7 0 0 7 e  0 6 ) L  0 .1 5 0 9 7 8 1 0 3 9 9 5 1 7 1 V
in

(8d)

x   1 7 .1 0 4 1 9 3 3 2 4 9 1 7 6  4 5 .2 9 3 0 8 5 3 0 0 4 8 1 2 x  4 2 .3 4 9 4 5 5 0 2 3 2 9 6 9 x
5 4 3
* *
 0 .0 5 2 4 4 3 7 7 0 1 3 7 4 0 8 4 L  4 3 .4 4 7 7 4 2 2 5 4 5 0 7 6 x x  0 .3 5 4 1 2 6 5 3 1 0 9 3 3 5 5 L x
3 4 4
* 2 2
 0 .3 4 8 4 6 7 4 1 4 6 0 2 4 5 6 L x  2 0 .7 9 3 3 9 3 2 5 4 5 4 9 6 x  2 2 .6 0 5 5 9 0 5 0 6 9 9 1 9 x
3 4 3
*2 * 2
 0 .0 0 0 9 6 2 6 8 3 5 5 1 0 3 9 2 0 5 L  0 .0 7 2 2 8 4 7 8 6 4 8 3 6 2 4 4 L x x  9 .0 1 1 4 8 5 3 8 3 9 6 4 2 2 x x
3 4 3 4
2 * 2
 9 .6 7 2 2 5 7 5 0 2 4 1 6 6 8 x x  0 .0 3 1 9 8 2 5 4 5 9 4 5 8 4 7 6 L x
3 4 4
* 2 *2
 0 .0 4 0 0 4 9 9 6 0 0 0 2 6 9 1 5 L x  0 .0 0 0 5 6 3 8 3 5 2 4 5 1 6 4 8 8 8 L x
3 4
*2 3
 0 .0 0 0 5 8 5 0 8 7 1 6 1 8 1 1 2 7 6 L x  2 .7 8 8 6 6 8 0 4 6 2 8 6 2 x
3 4
3 *3
 3 .4 4 9 4 6 1 6 7 5 5 3 5 9 x  (1 .3 3 2 5 6 5 6 1 3 2 0 8 7 5 e  0 6 ) L
3

(8e)

16 of 60
Page 16 of 59
 *
x  7 6 .8 7 0 8 0 8 7 4 5 0 4 0 9  3 .1 1 0 7 4 5 0 2 3 4 3 6 3 3 x  0 .9 2 2 1 7 4 0 0 0 7 4 8 8 9 5 L  6 .8 3 5 8 4 7 6 3 1 4 2 6 4 3 V
6 5 in
* *
 0 .0 3 9 4 8 9 9 7 1 4 8 5 8 4 8 L x  0 .1 9 9 6 9 0 8 6 7 6 5 4 9 7 2 V
x  0 .1 5 1 5 6 8 3 3 1 4 9 4 0 5 8 V L
5 in 5 in
2 *2 2
 0 .0 6 1 4 6 6 2 5 8 6 1 2 8 1 5 5 x  0 .0 0 2 2 9 6 9 2 6 1 6 0 5 8 8 3 9 L  0 .4 1 4 0 0 0 1 2 4 2 9 9 7 3 V
5 in
* * 2
 0 .0 0 3 7 1 3 7 4 7 0 0 3 1 2 1 6 1V L x  0 .0 0 0 3 8 3 1 0 2 5 7 4 4 7 3 6 7 1 L x
in 5 5
*2 2
 ( 4 .6 7 0 7 4 5 5 4 6 2 9 4 5 6 e  0 5 ) L x  0 .0 1 5 5 1 2 7 4 4 9 0 9 9 9 8 7 V x
5 in 5
*2 2
 0 .0 0 0 3 1 7 7 4 1 1 8 7 9 1 0 6 1 4 V L  0 .0 3 3 6 0 1 9 5 2 6 4 6 2 9 6 5 V x
in in 5
2 * 3
 0 .0 0 2 1 4 3 6 0 2 2 5 6 0 3 2 7 5 V L  0 .0 0 0 3 4 3 9 4 8 4 1 8 5 6 8 8 3 9 x
in 5
*3 3
 (1 .3 4 5 2 6 1 2 2 6 8 0 7 1 3 e  0 6 ) L  0 .0 2 1 4 4 5 9 9 8 2 5 2 1 9 3 4 V
in

(8f)

T   9 .2 5 8 2 7 8 0 6 3 4 9 5 4 5  1 .8 7 9 8 7 6 4 5 3 4 3 4 5 2 x  3 4 .1 0 0 4 5 3 3 5 0 4 5 1 7 W A R
out 6

 0 .2 0 5 2 8 7 6 1 1 4 6 0 6 3 5 T – 2 .2 6 1 3 9 6 5 0 7 2 1 6 7 6 W A R x
in 6
 0 .0 9 2 1 0 8 3 8 4 2 2 2 9 1 2 7 T  1 .7 3 6 5 6 2 9 3 7 8 3 2 5 9 T W A R
x
in 6 in
2 2 2
 0 .0 4 9 1 5 4 7 3 3 4 8 0 8 6 9 8 x  2 9 .6 4 9 0 9 6 2 9 3 3 1 0 4 W A R  0 .0 5 2 6 2 3 3 6 6 7 0 8 2 1 7 5 T
6 in
2
 0 .3 1 1 1 5 9 3 8 5 5 3 7 1 7 2 T W A R x  0 .0 5 9 3 3 3 5 9 1 9 6 6 3 6 7 1W A R x
in 6 6
2
 4 .8 0 6 2 1 0 3 0 7 3 1 0 4 7 W A R x – 0 .0 0 0 6 0 4 6 4 2 0 2 8 8 6 8 4 T x
6 in 6
2 2
 3 .4 1 2 8 4 1 5 3 4 2 6 7 0 8 T W A R  0 .0 0 0 4 8 0 4 9 6 0 1 9 4 4 2 9 5 5 T x
in in 6
2 3
 0 .1 3 3 4 2 2 0 0 3 2 3 9 5 8 7 T W A R  ( 3 . 8 1 2 9 6 4 0 6 9 0 2 6 2 3 e  0 5 ) x
in 6
3 3
 2 0 .8 6 0 1 4 6 6 8 5 2 3 4 4 W A R  0 .0 0 0 2 2 4 6 5 5 0 9 7 8 7 7 8 4 7 T
in

(8g)

5. Case study

The M-cycle CrFIEC which is intended to be modeled and optimized is the Coolerado M-50

unit. A schematic for this cooler and the flow direction of the unit were depicted in Fig. (5a) and

Fig. (5b), respectively. The general specifications of the Coolerado M-50 unit were indicated in

Table (4).

17 of 60
Page 17 of 59
 {Insert Fig. (5) here}

{Insert Table (4) here}

Since the M-cycle CrFIEC cools the ambient air, and the inlet air properties are effective in the

system performance, the operation of the system is highly affected by the weather condition. In

order to evaluate performance of the system in various categories of weather conditions,

Koppen–Geiger's climate classification as one of the most common and well known climate

classifications was considered. Then, potential usage of the system in various weathers was

assessed.

In the Koppen–Geiger's climate classification, every class is introduced by a label that consisting

of two or three letters. Every letter has a special meaning. Table (5), indicates the Koppen–

Geiger's climate symbols and their meanings.

{Insert Table (5) here}

For each class, a representing city was selected. It was tried that the selected cities cover all parts

of the world. Table (6) presents the specification of selected cities, and their outdoor summer

designs condition.

{Insert Table (6) here}

As seen, in some cities like Moscow (with Dfb label), the weather condition is so that less than

15% of people feels dissatisfaction from being space warm; therefore, these climates do not

require any cooling system. For the remained cities, the difference between dry-bulb and dew-

point temperatures is a good criterion for judgment about the IEC application potential [26].

Hence, based on Table (6), the M-cycle CrFIEC is applicable in 12 climate classes of Koppen–

18 of 60
Page 18 of 59
Geiger's classification system. Thus, the Coolerado M-50 unit was optimized in these 12 climate

regions.

6. Multi-objective performance optimization

6.1. Objective functions

The cooling capacity (CC) of the proposed M-cycle CrFIEC (Coolerado M-50 unit) was

considered as the first objective function to be maximized. In general, the cooling capacity of a

cooling system defined as follows:

C C  C C s e n s ib le  C C la te n t
(9)

Considering the fact that M-cycles provide only sensible cooling load, Eq. (9) was transformed

to Eq. (10):

C C  C C s e n s ib le  m in c p , a ir ( T in  T o u t )
(10)

On the other hand, the providing a more cooling load using less electric power consumption,

means a more efficient system; therefore, C O P was selected as the second objective function

for maximization. The C O P of an M-cycle CrFIEC is:

CC
COP  (11)
EPC

where C C and E P C are cooling capacity and electric power consumption of the system. The

electric power consumption of the M50 unit can be estimated by summation of its individual

sub-components’ electricity usage; therefore, we have:

19 of 60
Page 19 of 59
E P C  E P C b o a r d  E P C th e r m o s ta t  E P C fa n
(12)

Since the soap pump and solenoid operates periodically in only short time, they were not taken

into account in electric power consumption of the system [5]. The magnitudes of electricity

consumption of various components of the base M50 unit were indicated in Table (7).

{Insert Table (7) here}

It was assumed that the value of two first terms in the right side of Eq. (12) is not dependent on

decision variables; therefore, they remain unchanged during optimization; however electric

consumption of the fan depends on system operating variables as well as decision variables. In

this regard, the fan law [27] was employed as follows:

 Q SH
EPC fa n
 EPC fa n ,b a s e
( )( )( ) (13)
 base Q base SH base

where  , Q , and S H are air density, volumetric flow rate of the fan, and fan’s static head,

respectively. Because of small changes in the density ratio in Eq. (13), it is very close to 1.

Moreover, it is assumed that the fan static head ( S H ) is a multiplication of the pressure drop in

a safety factor. We suppose that safety factors of the base and optimized condition are the same.

Therefore, Eq. (13) were converted to the following expression:

Q P
EPC fa n
 EPC fa n ,b a s e
( )( ) (14)
Q base  Pb a s e

The pressure drop was obtained using the empirical correlations obtained by experimental data

of [5]. Hence pressure drops of dry and wet channels were given by Eq. (15) and Eq. (16),

respectively.

20 of 60
Page 20 of 59
 Pd r y c h a n n e l  0 .0 9 7 0 4 1 Q d r y c h a n n e l  8 0 .2 0
(15)

 Pw e t c h a n n e l  0 .1 3 1 1 8 5 Q w e t c h a n n e l  5 1 .0 4
(16)

Consequently, the total pressure drop through the unit is the summation of these two pressure

losses as indicated in Eq. (17):

 P   Pd r y c h a n n e l   Pw e t c h a n n e l
(17)

It is required to be mentioned that high COP doesn’t mean high CC [3, 9]. Since a desired cooler

is the one that has both high COP and CC, a comprehensive optimization should be dealt with

both objective functions, simultaneously. In addition, since the investigation should cover these

objectives for the whole cooling season, maximizing the average values of these objective

functions, was the goal of the multi-objective optimization. Hence, the objective functions were

considered as the average annual CC and COP.

O b j1  C C a v e , a n n u a l
(18)

O b j 2  C O Pa v e , a n n u a l
(19)

In evaluating annual values of objective functions, it was assumed that when more than 15% of

people in the space feel unsatisfactory indoor condition, the cooler is turned on. In this regard,

the predicted percent dissatisfaction (PPD) and predicted mean vote (PMV) indexes [28] were

used. It was supposed that once the PPD exceeds 15% PPD while P M V  0 , the M-cycle CrFIEC

is started to cool the space. In positive value of PMV, people feel warm. Negative PMV means

21 of 60
Page 21 of 59
people feel cold. In PMV calculations, insulations of clothes, average air velocity and metabolic

rate were considered 0.8 m2.K.W-1, 0.01 m.s-1 and 58.2 W.m-2, respectively.

6.2. Decision variables and constraints

Among the five effective parameters on the system performance (inlet air temperature, relative

humidity and velocity, dimensionless length and WAR), the inlet air temperature and relative

humidity cannot be considered as decision variables since their magnitudes are imposed by

outdoor condition (weather condition). In addition, it is not intended to change channel

dimensions of the current M50 unit, and it is desired to optimize the proposed system only based

on controlling its operation parameter. Consequently, the inlet air velocity and WAR were

selected as decision variables. The variation ranges of these variables were indicated in Table

(8).

{Insert Table (8) here}

6.3. Fundamentals of the multi-objective optimization

In order to explain the process, it was assumed that all objective functions are intended to be

minimized. Since maximizing +Obj is equivalent to minimizing –Obj, there is no difference

between these two processes. The second assumption was that none of the objective functions

have priority to the other one. If minimizing only one objective function is desired (single-

objective optimization), the answer is the result that has the least value of objective function

compared to others (Eq. (20)).

O b j  O p t a n s   O b j  O th e r s a n s 
(20)

22 of 60
Page 22 of 59
This inequality is written in single dimensional space according to well ordering principle, which

means two scalars can be compared; therefore, for a group of scalars, there is a minimum. Non-

commensurability of vectors is the challenging problem in multi-objective optimizations. For a

better comprehension, in Fig. (6), an example of vector comparison in minimizing two objective

functions was depicted. Suppose that ‘S’ is a point or vector (with two components), which is

intended to be compared to other vectors. Clearly, all vectors in the feasible area that have at

least a higher corresponding component while its other components are equal or greater than the

corresponding components of ‘S’, are worse than the vector ‘S’. To the contrary, vectors that

have at least one smaller corresponding component while other components are smaller or equal

to the corresponding components of the ‘S’, are better than the ‘S’. The main challenge is in the

case that we intend to compare a vector that has a higher and a lower corresponding component

(two components vector). Since in actual multi-objective optimizations (with conflict between

objectives), answers move in the opposite direction when decision variables change, finding a

way to compare these vectors is essential. In order to solve this, a concept that is called

“dominancy” has been employed. This implies that in the context of minimizing, vector ‘D’

dominants vector ‘S’, if all components of ‘D’ are less or equal to corresponding components of

‘S’ and at least one component of ‘D’ is strictly less than the corresponding component of ‘S’.

There is no preference between all solutions that do not dominant each other. In multi-objective

optimizations, such as solutions (vectors) make a set called Pareto optional Frontier, POF.

Multi-objective optimization problems generally show a possibly uncountable set of solutions;

whose evaluated vectors represent the best possible trade-offs in the objective function space.

Pareto optimality is the key concept to establish a hierarchy among the solutions of a multi-

23 of 60
Page 23 of 59
objective optimization problem, in order to determine whether a solution is really one of the best

possible trades-off. A schematic of a POF in the two-objective space was depicted in Fig. (6).

{Insert Fig. (6) here}

In this paper, an evolutionary method called non-dominated sorting algorithm II (NSGA-II) has

been used. This method was introduced by Deb et al. [29]. A brief description of the NSGA-II

was given in [30].

All solutions of the POF have same worthiness based on optimization criteria, because as stated

before, they do not dominate each other. However, in practical case, a final optimal solution is

required to be chosen among optimal points of the POF. In this regard, decision making methods

can be used. There are several decision methods to extract a final optimal solution from the POF.

Technique for order preference by similarity to ideal situation called as TOPSIS is one of the

simplest and efficient decision making methods for selecting the final optimal solution from

POF. In this method, non-dimensional vectors of solution were determined by Euclidian non-

dimensioning method as expressed in Eq. (21).

O b jij
F ij  (21)
N umi


2
( O b jij )
i 1

Then, an ideal and non-ideal solutions in objectives’ space as illustrated in Fig. (6) were defined.

It is clear that these solutions are not located on the POF; hence they are not achievable. In the


TOPSIS method, for each solution of the POF, special distances from the ideal ( d i ) and non-


ideal point ( d i ) are calculated by Eq. (22) and Eq. (23), respectively.

24 of 60
Page 24 of 59
 N u m O bj
 (22)

id e a l 2
di  ( F ij  F j )
j 1

N u m O bj
 n o n  id e a l (23)

2
di  ( F ij  F j )
j 1

At the end, a parameter denoted as C li was defined for each optimal point of the POF as per Eq.

(24):


di
C li   
(24)
di  di

Finally, a solution that had a maximum C li parameter was selected as the final desired optimal

solution [31]. Maximum C li means a maximum special distance from non-ideal point and/or a

minimum special deviation from the ideal one.

7. Results and discussion

7.1. Model validation

In order to validate obtained GMDH model, the predicted product air temperatures of the M-

cycle CrFIEC were compared with experimental data reported by Pandelidis et al. [4] as well as

the previous RSM prediction model reported by Pandelidis and Anisimov [9] as depicted in Fig.

(7). It should be mentioned that validation data were only selected among test data not numerical

one. Moreover, those experimental data that were used for validation had not been used for

training the GMDH network. As it is clear from Fig. (7), an appropriate accommodation exists

between the developed GMDH model prediction and experimental results, and the average error

25 of 60
Page 25 of 59
was only 1.76%. Moreover, it can be seen that the developed GMDH model is more accurate

than the previous RSM model [9].

{Insert Fig. (7) here}

7.2. Features of optimized systems

The accuracy of the developed GMDH model was validated in section 7.1. Then, the validated

model was employed to optimize the proposed Coolerado M50 unit with objective functions and

decision variables described in sections 6.1 and 6.2. The multi-objective optimization was

performed the developed code of NSGA-II in MATLAB R2013a. 2500, 0.7 and 0.3 were

selected as tuning parameters of the NSGA-II for the population size, crossover and mutation

rate, respectively. The value of decision variables at the base case condition were 1.96 m.s-1 and

0.425 for the inlet air velocity and WAR. These are the current values the M50 unit [32]. The

optimized values of the inlet air velocity and WAR were found for all 12 cities. Fig. (8a), Fig.

(8b), and Fig. (8c) show sample POFs obtained for Ahmedabad, Madrid and Windsor as

representing cities for ‘B’, ‘C’ and ‘D’ climate classes of Koppen–Geiger classification. were

considered in this paper, similar POFs were obtained for other nine areas. The best functions to

fit POFs were in in the form of power functions indicated in Eq. (25):

{Insert Fig. (8) here}

C2
C O Pa v e , a n n u a l  C 1 ( C C a v e , a n n u a l )
(25)

The corresponding coefficients of Eq. (25) for twelve cities were presented in Table (9).

{Insert Table (9) here}

26 of 60
Page 26 of 59
For each POF, a final optimal solution was selected using the TOPSIS decision making method

described in section 6.3. Table (10a) shows specifications of the optimized systems for

representing cities of class ‘B’. Results imply that this class had a relatively high potential for

improving the average annual COP. As seen, the system performance can improve much more in

hot and dry weathers. Furthermore, the optimized velocity values are close to the base case

condition for ‘B’ class climates. In this class as the weather gets warmer and drier the optimized

velocity value decreases. Furthermore, the optimized WAR values are very close together, so

that 0.318, which is the average WAR value for class ‘B’ cities, can be selected as optimized

WAR value for all class ‘B’ climates.

{Insert Table (10a) here}

Despite class ‘B’, this class has a wide range of optimum WAR from 0.198 to 0.296, so a certain

value cannot be selected for all cities of this class. Furthermore, according to Table (10b),

decreasing the humidity and temperature, reduce the optimized value of WAR. On the other

hand, the ambient temperature is a more effective factor on the optimized WAR than the relative

humidity.

{Insert Table (10b) here}

Similarly, Table (10c) presents optimized feature of the M50 CrFIEC for sample cities of class

‘D’. Similar fashions to other two classes were observed for the optimum velocity and WAR.

The value of velocity in optimum conditions is just a little less than base condition. Moreover,

the variation range of the optimum WAR was wider than class ‘B’ but smaller than class ‘C’.

{Insert Table (10c) here}

27 of 60
Page 27 of 59
It should be pointed out that for all optimized systems at ‘B’, ‘C’ and ‘D’ classes, the average

outlet temperature (product air temperature) was a bit higher than the corresponding outlet

temperature of the base case system (unit M50); nevertheless, this does not affect comfort

conditions of the space.

As it can be seen, in average, for all 12 cities, the average annual values of COP and CC can be

improved up to 8.1% and 6.9%, respectively. Since the optimum velocity range is 1.796 to 1.951

m.s-1 and this range is very close to the inlet velocity of the air of the base case system, there is

no need to change the fan and a little adjustment in fan operating parameter (for example,

example rotation speed) can provide desired velocity for the optimized system. With this regard

and according to this point that the system dimensions keep constant, optimization needed a low

cost and minimum changes. This takes account as the most remarkable and important advantage

of this optimization.

8. Conclusion

M-cycle indirect evaporative cooling with a cross-flow heat and mass exchanger (M-cycle

CrFIEC) was studied. A GMDH model for predicting the product air temperature as a function of

inlet temperature, inlet humidity, inlet velocity (WAR) and non-dimensional channel length was

developed. Based on the GMDH network and predicting polynomials, expressions of outlet

temperature (product air temperature) based on aforementioned five input parameters were

presented. The model was validated by experimental data, and its accuracy was compared with

similar previous model. It was found that the mathematical expression was simple in use and

accurate in predicting the system performance. On the other hand, it was found that its accuracy

was much superior to similar previous models. Hence, the coefficient of determination, R2, in

28 of 60
Page 28 of 59
prediction of input data was 0.991167. Moreover, the average error between predicted data and

experimental one was only 1.76%. Such simple expressions for product air temperature, enables

designers of CrFIECs to avoid complex numerical model.

Then, the GMDH model was employed in multi-objective optimization of the Coolerado M50

unit as a successful commercialized example of M-cycle CrFIEC cooler. The purpose of

optimization was to maximize annual COP and CC, while the inlet air velocity and WAR were

decision variables. It was found that, the M-cycle CrFIEC is applicable in cooling season of 12

classes of Koppen–Geiger's climate classification system. A sample city was selected for each

class and optimized features of the M50 unit was found in each city. It was found that the

optimized velocity decreases when inlet air temperature decreases or inlet relative humidity

increases. The optimized WAR decreases when temperature or relative humidity of the inlet air

(ambient condition) decrease. The optimized velocities were in the range of 1.796-1.951 m.s-1

which are very close to the corresponding value of the base case system (1.96 m.s -1 for the M50

unit prior to optimization). So there is no need to change the system’s fan; only with small

changes in operation of the fan (i.e. fan speed), the optimized condition can be obtained for each

area. The optimum WARs were considerably smaller than the corresponding value of the base

case system (0.425). The optimum WAR values for cities of class ‘B’ were very close to each

other, so that the average optimized value of 0.318 can be applicable to all cities of this class.

Among all classes, this class had the highest potential for improving the average annual COP and

lowest potential for improving the average annual cooling capacity (CC). Cities of class ‘C’ had

a wide range of the optimum WAR, i.e. 0.196- 0.296. This class of weather condition had a

higher potential for improving the average annual cooling capacity (CC) compared to the

weather classes ‘C’ and ‘D’. In class ‘D’, the optimum range of WAR and the average annual

29 of 60
Page 29 of 59
CC were somehow between class ‘B’ and ‘C’. The potential for improving the COP in this class

was less than other classes (‘B’ and ‘C’). The range of optimized was 0.251-0.280, in class ‘C’.

In average for all 12 cities at a wide range of world’s weather condition, the average annual COP

and CC were increased 8.1% and 6.9%, respectively. The most remarkable point about this

improvement was that it can be put into practice with minimum changes and costs.

30 of 60
Page 30 of 59
 References

1. Anisimov, S., D. Pandelidis, and V. Maisotsenko, Numerical study of heat

and mass transfer process in the Maisotsenko cycle for indirect evaporative
air cooling. Heat Transfer Engineering, 2016: p. 1-40.
2. Jradi, M. and S. Riffat, Experimental and numerical investigation of a dew-
point cooling system for thermal comfort in buildings. Applied Energy,
2014. 132: p. 524-535.
3. Zhan, C., et al., Comparative study of the performance of the M-cycle
counter-flow and cross-flow heat exchangers for indirect evaporative
cooling – Paving the path toward sustainable cooling of buildings. Energy,
2011. 36(12): p. 6790-6805.
4. Pandelidis, D., S. Anisimov, and W.M. Worek, Performance study of the
Maisotsenko Cycle heat exchangers in different air-conditioning
applications. International Journal of Heat and Mass Transfer, 2015. 81: p.
207-221.
5. Weerts, B.A., NSIDC green data center project: Coolerado and modeling an
application of the Maisotsenko cycle, 2012, UNIVERSITY OF
COLORADO AT BOULDER.
6. Zhao, X., S. Liu, and S.B. Riffat, Comparative study of heat and mass
exchanging materials for indirect evaporative cooling systems. Building and
Environment, 2008. 43(11): p. 1902-1911.
7. Zhan, C., et al., Numerical study of a M-cycle cross-flow heat exchanger for
indirect evaporative cooling. Building and Environment, 2011. 46(3): p.
657-668.
8. Kiran, T.R. and S.P.S. Rajput, An effectiveness model for an indirect
evaporative cooling (IEC) system: Comparison of artificial neural networks
(ANN), adaptive neuro-fuzzy inference system (ANFIS) and fuzzy inference
system (FIS) approach. Applied Soft Computing, 2011. 11(4): p. 3525-3533.
9. Pandelidis, D. and S. Anisimov, Application of a statistical design for
analyzing basic performance characteristics of the cross-flow Maisotsenko
cycle heat exchanger. International Journal of Heat and Mass Transfer,
2016. 95: p. 45-61.
10. Chen, Y., H. Yang, and Y. Luo, Indirect evaporative cooler considering
condensation from primary air: Model development and parameter analysis.
Building and Environment, 2016. 95: p. 330-345.
11. Chen, Y., Y. Luo, and H. Yang, A simplified analytical model for indirect
evaporative cooling considering condensation from fresh air: Development
and application. Energy and Buildings, 2015. 108: p. 387-400.

31 of 60
Page 31 of 59
12. Maisotsenko, V.S., Method of air cooling and heat exchange apparatus,
1995, Google Patents.
13. Maisotsenko, V., et al., Indirect evaporative cooling mechanism, 2002,
Google Patents.
14. Maisotsenko, V., et al., Method and apparatus of indirect-evaporation
cooling, 2002, Google Patents.
15. Gillan, L., Maisotsenko cycle for cooling processes. International Journal of
Energy for a Clean Environment, 2008. 9(1-3).
16. Wicker, K., Life below the wet bulb: The Maisotsenko cycle. Power, 2003.
147(9): p. 29.
17. Wani, C., S. Ghodke, and C. Shrivastava, A review on potential of
Maisotsenko cycle in energy saving applications using evaporative cooling.
International journal of advance research in science, engineering and
technology, 2012. 1(1): p. 15-20.
18. Ivakhnenko, A. and G. Ivakhnenko, The review of problems solvable by
algorithms of the group method of data handling (GMDH). Pattern
Recognition And Image Analysis C/C Of Raspoznavaniye Obrazov I Analiz
Izobrazhenii, 1995. 5: p. 527-535.
19. Madala, H.R. and A.G. Ivakhnenko, Inductive learning algorithms for
complex systems modeling. 1994: cRc press Boca Raton.
20. Farlow, S.J., Self-organizing methods in modeling: GMDH type algorithms.
Vol. 54. 1984: CrC Press.
21. Duan, Z., Investigation of a novel dew point indirect evaporative air
conditioning system for buildings, 2011, University of Nottingham.
22. Zhao, X., J.M. Li, and S.B. Riffat, Numerical study of a novel counter-flow
heat and mass exchanger for dew point evaporative cooling. Applied
Thermal Engineering, 2008. 28(14–15): p. 1942-1951.
23. Jacobsons, G. GMDH type polynomial neural network toolbox [cited 2015
May, 6]; Available from: http://www.cs.rtu.lv/jekabsons/regression.html.
24. Sugiura, N., Further analysts of the data by akaike's information criterion
and the finite corrections: Further analysts of the data by akaike's.
Communications in Statistics-Theory and Methods, 1978. 7(1): p. 13-26.
25. Hurvich, C.M. and C.-L. Tsai, Regression and time series model selection in
small samples. Biometrika, 1989. 76(2): p. 297-307.
26. Hasani Balyani, H., et al., Acquiring the best cooling strategy based on
thermal comfort and 3E analyses for small scale residential buildings at
diverse climatic conditions. International Journal of Refrigeration, 2015. 57:
p. 112-137.
27. Fox, R.W., A.T. McDonald, and P.J. Pritchard, Introduction to fluid
mechanics. 2006: John Wiley & Sons.

32 of 60
Page 32 of 59
28. Iso, E., 7730: 2005:“Ergonomics of the thermal environment–Analytical
determination and interpretation of thermal comfort using calculation of the
PMV and PPD indices and local thermal comfort criteria”. International
Organization for Standardisation, Geneva, 2005.
29. Deb, K., et al., A fast and elitist multiobjective genetic algorithm: NSGA-II.
Evolutionary Computation, IEEE Transactions on, 2002. 6(2): p. 182-197.
30. Sayyaadi, H. and H.R. Aminian, Design and optimization of a non-TEMA
type tubular recuperative heat exchanger used in a regenerative gas turbine
cycle. Energy, 2010. 35(4): p. 1647-1657.
31. Ahmadi, M.H., et al., Application of the multi-objective optimization method
for designing a powered Stirling heat engine: Design with maximized power,
thermal efficiency and minimized pressure loss. Renewable Energy, 2013.
60: p. 313-322.
32. M50 Product Brochure. [cited 2015 October, 11]; Available from:
http://www.coolerado.com/products-
files/2012/M50/M50_brochure_1_2012.pdf.
33. Coolerado M50 [cited 2015 October, 26]; Available from:
http://www.blueplanetgreenliving.com/wp-
content/uploads/2009/04/cooleradom50whtbckgrnd.jpg.
34. Coolerado M50 submittal data sheet. [cited 2015 October, 7]; Available
from: http://www.coolerado.com/wp-content/uploads/2014/11/Coolerado-
M50B-Submittal-11-2014.pdf.
35. Peel, M.C., B.L. Finlayson, and T.A. McMahon, Updated world map of the
Köppen-Geiger climate classification. Hydrology and Earth System
Sciences Discussions Discussions, 2007. 4(2): p. 439-473.
36. Handbook-Fundamentals, A., American society of heating, refrigerating and
air-conditioning engineers. Inc., NE Atlanta, GA, 2009. 30329.
37. World climate & temperature. [cited 2015 November, 21]; Available from:
http://www.climatemps.com/.

33 of 60
Page 33 of 59
 Thin plate
Cooled and
Hot unchanged
and dry Dry (A) channel absolute
air humidity air

Wet (B) channel

Wet surface by water film

(a)

34 of 60
Page 34 of 59
 Thin plate Condensate

Hot
and dry Dry (A) channel Cooled air
air

Wet (B) channel Evaporated water

Wet surface by water film

(b)

35 of 60
Page 35 of 59
 Thin plate Condensate

Hot
and dry Dry (A) channel Cooled air
air

Wet (B) channel Evaporated water

Wet surface by water film

(c)
Fig. (1): Description of the working principle of indirect evaporative cooling systems in three different
possible states in dry channels: (a) non-condensation state; (b) partially condensation state; (c) totally
condensation state

36 of 60
Page 36 of 59
 Cooled and
Hot unchanged
and dry absolute
air humidity air

Outlet
Inlet humid
working working
air air

(a)

37 of 60
Page 37 of 59
 Cooled and
Hot unchanged
and dry absolute
air humidity air

Outlet
humid Inlet
working working air
air

(b)

38 of 60
Page 38 of 59
 Cooled and
Hot unchanged
and dry absolute
air humidity air

Working Working
air air
streams streams

(c)
Fig. (2): Different types of indirect evaporative cooling based on dry to wet channel flow arrangement: (a)
parallel flow (PFIEC); (b) counter flow (CoFIEC); (c) cross flow (CrFIEC)

39 of 60
Page 39 of 59
 Cooled and unchanged absolute humidity (product) air

Dead end
Outlet humid working air

Hole
Hot and dry inlet air

Fig. (3): Schematic of investigated CrFIEC (M-cycle CrFIEC) from top view

40 of 60
Page 40 of 59
 Specify Specify Specify input Specify
number of maximum source for external
Start layers after 2nd k=0
inputs for polynomials criterion and
each neuron degree layer set its limits

m≤mk Create all possible (mk)

m= 1 k = k +1
neurons (kth layer)

Yes m = m +1

No
No
Eliminate Keep
neuron neuron
Comprise mth
neuron external
criterion to
Worst adjusted value Better

Has the best

Sort layer
neuron external
neurons
criterion better
based on
than stopping
external
limit?
criterion

Yes

Introduce best neuron polynomial as prediction function End

Fig. (4): The GMDH-type neural network design process flowchart

41 of 60
Page 41 of 59
 (a)

42 of 60
Page 42 of 59
 Working air exhaust
(out top, or both sides)

Outside air intake

Conditioned/
product air

Working air exhaust

(out top, or both sides)

(b)
Fig. (5): (a) Coolerado M50 model M-cycle CrFIEC [33]; (b) flow directions in M50 [34]

43 of 60
Page 43 of 59
 Obj2

F
E

S S
H
D
The
J
Pareto
optimal
front
Obj2,min Solution infeasible area
Ideal point
Solution feasible area
Obj1
Obj1,min

Fig. (6): Introducing non-dominated solutions and Pareto optimal front schematically in the solution area

44 of 60
Page 44 of 59
 Pandelidis et al. [4] GMDH Model RSM Model [9]
22

21
Product air temperature (⁰C)

20

19

18

17

16

15
2.5 2.6 2.7 2.8 2.9 3.0 3.1 3.2 3.3
Inlet air velocity (m.s-1)

Fig. (7): Evaluation of GMDH model ability in prediction of experimental data reported by
Pandelidis et al. [4] that were not used in order to develop this model. Its comparison with
previous RSM model [9]

45 of 60
Page 45 of 59
 90
80
70
Average annual COP

60
50 TOPSIS selection
40
30 COPave,annual = 4565.6 (CCave,annual)-2.159
20 R² = 0.9948

10
0
0 5 10 15 20 25
Average annual cooling capacity (kW)

(a)

46 of 60
Page 46 of 59
 100
90
80
Average annual COP

70
60
50 TOPSIS selection
40
COPave,annual = 3141.8 (CCave,annual)-2.1
30
R² = 0.9955
20
10
0
0 5 10 15 20 25
Average annual cooling capacity (kW)

(b)

47 of 60
Page 47 of 59
 90
80
70
Average annual COP

60
50
40 TOPSIS selection
30
20 COPave,annual = 1534.6 (CCave,annual)-2.285
R² = 0.993
10
0
0 2 4 6 8 10 12 14 16
Average annual cooling capacity (kW)

(c)
Fig. (8): Obtained Pareto optimal front for (a) Ahmedabad, India; (b) Madrid, Spain; (c)
Windsor, Canada

48 of 60
Page 48 of 59
 Table (1): The studies data for make M-cycle CrFIEC GMDH product air temperature model
extracted from them.
Study Data type

Jradi and Riffat [2] Experimental and numerical

Zhan et al. [3] Numerical

Pandelidis et al. [4] Numerical

49 of 60
Page 49 of 59
 Table (2): Variation ranges of the M-cycle CrFIEC system effective variables extracted for
making GMDH product air temperature model
Parameter Variation range

Inlet air temperature (⁰C) 25-45

Inlet air relative humidity (%) 10-90

Inlet air velocity (m.s-1) 1.2-6.0

Dimensionless length 40-400

Working to inlet air ratio 0.10-0.95

50 of 60
Page 50 of 59
 Table (3): The results of some iterations have been done to determine the best GMDH model for
the M-cycle CrFIEC product air temperature
Number of Maximum allowable SSE MSE
Input source for neurons Polynomials
inputs for number of neurons R2
after 2nd layer power (⁰C) 2
(⁰C)
each neuron in each layer

2 preceding layer 1 2 7229.91 5.383400 0.707974

2 preceding layer 8 3 990.18 0.737288 0.960005

2 preceding layer+inputs 2 3 460.67 0.343013 0.981393

2 preceding layer+inputs 3 2 530.71 0.395171 0.978564

2 preceding layer+inputs 4 3 393.53 0.293026 0.984105

2 preceding layer+inputs 5 3 389.50 0.290025 0.984267

2 preceding layer+inputs 9 3 351.16 0.261477 0.985816

2 preceding layer+inputs 10 2 383.11 0.285263 0.984526

2 preceding layer+inputs 10 3 343.25 0.255582 0.986136

3 preceding layer 1 2 2335.16 1.738764 0.905680

3 preceding layer 5 3 362.80 0.270142 0.985346

3 preceding layer 7 3 365.57 0.272202 0.985234

3 preceding layer 10 2 367.70 0.273787 0.985148

3 preceding layer 10 3 328.86 0.244871 0.986717

3 preceding layer+inputs 3 3 229.62 0.170972 0.990726

3 preceding layer+inputs 5 3 219.54 0.163466 0.991133

3 preceding layer+inputs 6 3 218.69 0.162833 0.991167

3 preceding layer+inputs 8 2 276.36 0.205779 0.988837

3 preceding layer+inputs 9 3 230.07 0.171311 0.990707

3 preceding layer+inputs 10 3 229.62 0.170972 0.990726

51 of 60
Page 51 of 59
 Table (4): The specifications of Coolerado M50 CrFIEC Cooler [5, 32, 34]
Parameter Value

Length (m) 0.762

Width (m) 0.667

Height (m) 1.384

Dry channels gap (mm) 3.23

Wet channels gap (mm) 2.15

Rated inlet air flow rate (m3.hr-1) 2250

52 of 60
Page 52 of 59
 Table (5): Koppen-Geiger climate symbols and their meanings [35]
Symbol place in the name
meaning
1st 2nd 3rd

A Tropical

f Rainforest

m Monsoon

w Savannah

B Arid

W Desert

S Steppe

h Hot

k Cold

C Temperature

s Dry summer

w Dry winter

f Without dry season

a Hot summer

b Warm summer

c Cold summer

D Cold

s Dry summer

w Dry winter

f Without dry season

a Hot summer

b Warm summer

c Cold summer

d Very cold summer

E Polar

T Tundra

F Frost

53 of 60
Page 53 of 59
 Table (6): The specification of selected cities from each Koppen-Geiger climate region, their
outdoor summer design conditions and applicability of using system in them [35-37]
Group name
Altitude Lat Long Tdb RH Tdp Tdb -Tdp Application
in Koppen City Country
(m) (⁰N) (⁰E) (⁰C) (%) (⁰C) (⁰C) status
classification
Af Entebbe Uganda 1155.0 0.0 32.5 27.0 80.0 23.3 3.7 Not applicable
Am Jakarta Indonesia 7.9 -6.2 106.8 32.2 65.4 24.9 7.3 Not applicable
Aw Darwin Australia 31.0 -12.3 130.5 34.5 64.0 26.7 7.8 Not applicable
BWh Riyadh Saudi Arabia 611.7 24.6 46.7 43.9 5.1 -3.4 47.3 Applicable
BWk Tulufan China 34.5 42.9 89.2 40.7 24.2 16.3 24.4 Applicable
BSh Ahmedabad India 54.9 23.0 72.6 42.2 19.8 14.4 27.8 Applicable
Bsk Tehran Iran 1219.8 35.7 51.4 38.9 30.6 18.5 20.4 Applicable
Csa Madrid Spain 581.9 40.4 -3.7 36.1 24.9 12.9 23.2 Applicable
Csb Porto Portugal 72.8 41.2 -8.6 30 36.9 13.7 16.3 Applicable
Cfa Atlanta U.S.A. 314.9 33.8 -84.4 33.9 44.2 20.0 13.9 Applicable
Cwa Hong Kong China 23.8 22.3 114.2 33.3 57.0 23.6 9.7 Not applicable
Cfb London United Kingdom 5.0 51.5 -0.1 28 48.0 16.0 12.0 Applicable
Cwb Mexico City Mexico 2233.9 19.4 -99.1 28.9 21.5 4.7 24.2 Applicable
Cwc La Paz Bolivia 4071.0 -16.5 -68.2 16.8 27.3 -1.9 18.7 Not required
Cfc Reykjavík Iceland 61.0 64.1 -21.9 14.0 80.0 10.6 3.4 Not required
Csc Balmaceda Chile 516.0 -45.9 -71.7 17.5 62.0 10.1 7.4 Not required
Dfa Windsor Canada 189.9 42.3 -83.0 31.7 47.2 19.1 12.6 Applicable
Dwa Seoul South Korea 18.9 37.6 127.0 31.7 58.4 22.6 9.2 Not applicable
Dsa Arak Iran 1718.0 34.1 49.7 35.6 18.3 7.9 27.7 Applicable
Dfb Moscow Russia 155.8 55.8 37.6 27.8 45.9 15.1 12.7 Not required
Dsb Ankara Turkey 948.8 40.1 33.0 32.2 22.7 8.3 23.9 Applicable
Dwb Khabarovsk Russia 86.0 48.5 135.1 26.6 78.0 22.4 4.2 Not applicable
Dfc Nome U.S.A. 6.0 64.5 -165.4 14.3 88.0 12.3 2.0 Not required
Dwc Mohe county China 435.0 53.0 122.5 25.8 43.3 11.3 14.5 Not required
Dsc Mayo Canada 504.0 63.6 -135.9 22.4 52.0 12.1 10.3 Not required
Dfd Yakutsk Russia 101.0 62.0 129.7 25.0 59.0 16.4 8.6 Not required
Dwd Oymyakon Russia 740.0 63.5 142.8 22.7 65.0 15.8 6.9 Not required
Dsd Eastern Siberia Russia 162.0 55.0 82.9 20.2 59.0 11.9 8.3 Not required
ET Vardo Norway 15.0 70.4 31.1 11.8 87.0 9.7 2.1 Not required
EF McMurdo Antarctica 24.0 -77.8 166.7 -2.8 65.0 -7.8 5.0 Not required

54 of 60
Page 54 of 59
 Table (7): The details of M50 power consumption [5]
Component Power consumption (W)

Board 8.6

Thermostat 1.2

Fan 618.2

Soap pump 6.5

Solenoid 14

55 of 60
Page 55 of 59
 Table (8): Considered variation ranges for decision variables in optimization process
decision variable Minimum Maximum

Inlet air velocity (m.s-1) 0.0 5.0

Working to inlet air ratio 0.00 1.00

56 of 60
Page 56 of 59
Table (9): The value of Pareto optimal front (POF) curve coefficients for the selected cities
Group name in
Koppen-Geiger City C1 (kW) -C2 C2 R2
classification
BWh Riyadh 5991.4 -2.015 0.9983
BWk Tulufan 3582.4 -2.159 0.9979
BSh Ahmedabad 4565.6 -2.159 0.9948
Bsk Tehran 1870.0 -2.208 0.9976
Csa Madrid 3141.8 -2.100 0.9955
Csb Porto 2100.0 -2.044 0.9957
Cfa Atlanta 1470.6 -2.294 0.9948
Cfb London 1415.7 -2.070 0.9920
Cwb Mexico City 1465.4 -1.872 0.9985
Dfa Windsor 1534.6 -2.285 0.9930
Dsa Arak 2391.1 -2.010 0.9979
Dsb Ankara 2449.5 -1.959 0.9984

57 of 60
Page 57 of 59
 Table (10a): Results of cities that are representing various classification of the class ‘B’

Group Optimum condition Base case condition

Operation CC COP
name in
Tdb RH hours CC TAir,out CC TAir,out Improvement Improvement
Koppen- City V
(⁰C) (%) (hr.year-1) WAR COP COP (%) (%)
Geiger (m.s-1) (kW) (⁰C) (kW) (⁰C)
classification
BWh Riyadh 43.9 5.1 5505 1.836 0.313 14.752 26.260 15.2 14.580 23.281 13.1 1.179 12.795

BWk Tulufan 40.7 24.2 3528 1.901 0.322 11.396 18.788 19.4 11.060 17.660 17.9 3.037 6.384

BSh Ahmedabad 42.2 19.8 6752 1.853 0.319 12.062 21.125 19.0 11.972 19.117 17.3 0.753 10.502

Bsk Tehran 38.9 30.6 2159 1.944 0.316 9.141 14.198 20.4 8.715 13.915 18.9 4.893 2.034

Average 2.466 7.929

The values of CC, COP and TAir,Out in this table are the annual averages of them. Vbase=1.96 m.s-1, WARbase =0.425

Table (10b): Results of cities that are representing various classification of the class ‘C’

Group Optimum condition Base case condition

Operation CC COP
name in
Tdb RH hours CC T Air,out CC TAir,out Improvement Improvement
Koppen- City V
(⁰C) (%) (hr.year-1) WAR COP COP (%) (%)
Geiger (m.s-1) (kW) (⁰C) (kW) (⁰C)
classification
Csa Madrid 36.1 24.9 1794 1.923 0.296 11.632 18.122 17.8 10.874 17.363 16.0 6.974 4.372

Csb Porto 30.0 36.9 699 1.867 0.233 10.735 16.342 17.3 9.481 15.139 15.9 13.226 7.947

Cfa Atlanta 33.9 44.2 2098 1.951 0.290 8.131 12.146 21.0 7.501 11.977 19.8 8.407 1.414

Cfb London 28.0 48.0 182 1.848 0.198 9.384 13.702 18.2 7.953 12.670 17.1 17.993 7.893

Cwb Mexico City 28.9 21.5 182 1.796 0.204 10.734 16.991 13.5 9.452 15.093 11.7 13.563 12.576

Average 12.033 6.840

The values of CC, COP and TAir,Out in this table are the annual averages of them. Vbase=1.96 m.s-1, WARbase =0.425

58 of 60
Page 58 of 59
 Table (10c): Results of cities that are representing various classification of the class ‘D’

Group Optimum condition Base case condition

Operation CC COP
name in
Tdb RH hours CC TAir,out CC TAir,out Improvement Improvement
Koppen- City V
(⁰C) (%) (hr.year-1) WAR COP COP (%) (%)
Geiger (m.s-1) (kW) (⁰C) (kW) (⁰C)
classification
Dfa Windsor 31.7 47.2 882 1.914 0.251 8.326 12.300 20.5 7.507 11.987 19.1 10.910 2.609

Dsa Arak 35.6 18.3 1125 1.861 0.280 11.195 18.493 16.4 10.567 16.874 14.0 5.943 9.596

Dsb Ankara 32.2 22.7 669 1.904 0.254 12.150 18.275 15.6 10.840 17.309 13.5 12.085 5.5792

Average 9.643 5.928

The values of CC, COP and TAir,Out in this table are the annual averages of them. Vbase=1.96 m.s-1, WARbase =0.425

59 of 60
Page 59 of 59