Journal of Power Sources 247 (2014) 643e654

Contents lists available at ScienceDirect

Journal of Power Sources

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

Analysis and optimization of hybrid electric vehicle thermal

management systems
H.S. Hamut a, *, I. Dincer a, G.F. Naterer b
a
Faculty of Engineering and Applied Science, University of Ontario Institute of Technology, 2000 Simcoe St. North, Oshawa, ON L1H 7K4, Canada
b
Faculty of Engineering and Applied Science, Memorial University of Newfoundland, 240 Prince Phillip Drive, St. John’s, NL A1B 3X5, Canada

h i g h l i g h t s

A multi-objective optimization of HEV thermal management systems is proposed.

Exergetic, exergoeconomic and exergoenvironmental objective functions are utilized.

A Pareto frontier is obtained and a single desirable optimal solution is selected.

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

Article history: In this study, the thermal management system of a hybrid electric vehicle is optimized using single and
Received 26 February 2013 multi-objective evolutionary algorithms in order to maximize the exergy efficiency and minimize the
Received in revised form cost and environmental impact of the system. The objective functions are defined and decision variables,
24 August 2013
along with their respective system constraints, are selected for the analysis. In the multi-objective
Accepted 29 August 2013
Available online 9 September 2013
optimization, a Pareto frontier is obtained and a single desirable optimal solution is selected based on
LINMAP decision-making process. The corresponding solutions are compared against the exergetic,
exergoeconomic and exergoenvironmental single objective optimization results. The results show that
Keywords:
Multi-objective optimization
the exergy efficiency, total cost rate and environmental impact rate for the baseline system are deter-
Exergy mined to be 0.29, ¢28 h1 and 77.3 mPts h1 respectively. Moreover, based on the exergoeconomic
Exergoeconomics optimization, 14% higher exergy efficiency and 5% lower cost can be achieved, compared to baseline
Exergoenvironmental analysis parameters at an expense of a 14% increase in the environmental impact. Based on the exergoenvir-
Hybrid electric vehicle onmental optimization, a 13% higher exergy efficiency and 5% lower environmental impact can be
achieved at the expense of a 27% increase in the total cost.
Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction of inefficiencies in systems. However, exergy analysis alone provides

no information on the economic and environmental aspects of po-
In a world with finite natural resources and increasing energy tential impacts and possible improvements. Thus, an integrated
demand and prices, developing systems that are more efficient, procedure that combined all of these concerns should be developed
cost-effective and environmentally benign is one of the most sig- to find a viable solution. For this reason, it is aimed to couple the
nificant challenges that many engineers are facing today. In the past second law of thermodynamics through exergy with economic and
decades, the energy prices have been increasing while the legisla- environmental impact in order to develop a powerful tool for the
tions that aim to mitigate environmental problems (such as air systematic study and optimization of energy systems [1,2].
pollution, water pollution and hazardous wastes) have become Exergoeconomic analysis essentially combines both exergy
more stringent. In this regard, exergy analysis is employed as a po- analysis and economic factors, through cost accounting about
tential tool to design, analyze, assess and improve system compo- initial costs, operating and maintenance costs, resource costs, and
nents and helps determine the locations, types and true magnitude incorporates these costs into thermodynamic parameters for sys-
tem evaluation. This method will the help improve and optimize
the system for better efficiency and better cost effectiveness [3].
On the other hand, exergoenvironmental analysis combines
* Corresponding author.
E-mail addresses: [email protected] (H.S. Hamut), [email protected]
exergy analysis and the environmental impact, associated with
(I. Dincer), [email protected] (G.F. Naterer). construction, operation and maintenance and disposal stages, and

0378-7753/$ e see front matter Ó 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jpowsour.2013.08.131
644 H.S. Hamut et al. / Journal of Power Sources 247 (2014) 643e654

allocates the corresponding impacts to the exergy streams, in order that the exergetic efficiency in the multi-objective optimum design
to point out the components causing the highest environmental is 11.11% higher than that of the exergoeconomic optimized system,
impact. It also suggests possibilities and trends for improvement, while the total product cost of the multi-objective optimal design is
based on the calculated exergoenvironmental variables. 16.7 times higher than that of the exergoeconomic optimal system.
Subsequently, multi-objective optimization with respect to the Even though many single and multi-objective optimizations
aforementioned analyses has been utilized in order to compensate were conducted using a wide variety of optimization tools and
shortcomings of traditional single objective approaches (namely objective functions in the literature, to our knowledge, there have
single objective exergy, exergoeconomic and exergoenvironmental been no multi-objective optimization models developed in the
optimizations) by allowing a larger perspective and determining a literature with respect to exergy efficiency, exergoeconomic and
more complete spectrum of solutions that optimize the design exergoenvironmental objective functions for hybrid electric vehicle
according to more than one objective at a time. In most practical thermal management systems with respect to evolutionary
decision making problems, the objectives are conflicting in nature algorithm.
and a unique optimal solution cannot be identified. Thus, Pareto
optimality is introduced to determine whether a solution is really
one of the best possible trade-offs [4,5]. In this regard, numerous 2. System description
single and multi-objective optimizations have been carried out in
the past with respect to various criteria and for a wide range of Hybrid electric vehicle thermal management systems (HEV
applications by many researchers. TMSs) are significantly different systems with unique requirements
Lazaretto and Toffolo [4] compared a single-objective thermo- with respect to their commercial and industrial counterparts such
economic optimization with two-objective energetic and economic as conventional vehicle and residential building air conditioning
optimization for thermal system designs using energy, economy systems [8]. The TMS needs to handle significant thermal load
and environment as separate objectives. They analyzed a test case variations and provide comfort under highly fluctuating conditions.
plant of the CGAM problem with respect to three-objective They must also be compact and efficient, and last several years
approach. The environmental impact objective function was without any significant maintenance. Thus, special attention needs
defined with respect to the amount of carbon dioxide and nitrogen to be given to hybrid electric vehicle TMSs [9].
oxide emissions. An evolutionary algorithm was used to find the A simplified thermal management system of an electric vehicle
surface of the optimal solutions based on the three objective with liquid battery cooling is considered in Fig. 1. The system is
functions. They determined the Pareto optimal curve for the multi- composed of two loops, namely a refrigerant and battery coolant
objective optimization and discussed possible points on the curve loop. The refrigerant loop enables air conditioning of the vehicle
based on the trade-off between the total cost and environmental cabin, while the coolant loop keeps the electric battery operating
impact. within its ideal temperature range. These two loops are connected
Gebreslassie et al. [6] proposed a systematic method based on via a chiller, which enables heat exchange among the loops to
mathematical programming for the design of environmentally provide supercooling to the battery cooling as it passes through the
conscious absorption cooling systems with respect to a multi- chiller unit. This increases the efficiency of the system significantly
objective formulation that simultaneously accounts for the mini- since cooling via the refrigeration circuit would consume more
mization of cost and environmental impact at the design stage. The energy than operating the battery coolant circuit due to the need of
environmental impact criterion was measured by the Eco-indicator the air compressor in the first case [10]. More detailed information
99 methodology, which follows the principles of life cycle assess- on the thermal management system along with its components can
ment (LCA). They picked three points on the Pareto optimal curve be found in Ref. [11].
that represents the minimum Eco-indicator solution (A), minimum
total cost solution (B) and a possible trade-off solution between the
two points (C). By switching from solution B to solution C, the total 3. Thermodynamic model
Eco-indicator 99 value was reduced by 3.8% at the expense of 4.8%
increase in the total cost. 3.1. Exergy analyses
Ahmadi et al. [7] conducted a comprehensive exergy, exer-
goeconomic and environmental impact analysis and a multi- In this paper, Engineering Equation Solver (EES) is used to solve
objective optimization for combined cycle power plants (CCPPs) mass, energy, entropy and exergy balance equations for each
with respect to the exergy efficiency, total cost rate and CO2 system component, where the work input, rate of entropy gener-
emissions of the overall plant. They determined that the largest ation and exergy destruction as well as the energy and exergy
exergy destructions occurred in the CCPP combustion chamber. efficiencies are calculated accordingly. These balance equations,
Increasing the gas turbine inlet air temperatures decreases the under steady-state assumptions, can be written in the general
CCPP cost of exergy destruction. They derived the expression for the form as follows:
Pareto optimal point curves for the determined exergy efficiency
range and concluded that the increase in total cost per unit exergy m _ out
_ in ¼ m (1a)
efficiency is considerably high after exergy efficiencies over 57%
and therefore a point below this should be chosen on the Pareto E_ in ¼ E_ out (1b)
optimal curve.
Sayyaadi and Babaelahi [5] analyzed a liquefied natural gas re-
S_ in þ S_ gen ¼ S_ out (1c)
liquefaction plant with respect to a multi-objective approach
which simultaneously considers exergy and exergoeconomic ob-
_ _ _
jectives. They used a MATLAB multi-objective optimization algo- Ex in ¼ Exout þ ExD (1d)
rithm of NSGA-II, which is based on the Genetic Algorithm, and
obtained a Pareto optimal frontier to find the Pareto optimal so- where
lutions. They compared the final optimal system with the base case
and exergoeconomic single-objective optimized system and found S_ gen ¼ m
_ Ds (1e)
 H.S. Hamut et al. / Journal of Power Sources 247 (2014) 643e654 645

Fig. 1. Simplified representation of the hybrid electric vehicle thermal management system.

_ D ¼ T S_ gen
Ex (1f) _ _
Ex _ _
þ Ex
0 Q Q
COPex;system ¼ evap ch
(4)
_ comp þ W
W _ pump
Here, m _ and E_ are associated with the mass flow rate and energy
transfer rate. The respective total rates in/out across the boundary
are conserved (neglecting reactions). In the third equation, S_ is the
3.2. Exergoeconomic analysis
entropy flow or generation rate. The amount transferred out of the
boundary must exceed the rate at which entropy enters, the dif-
In order to conduct an exergoeconomic analysis, the cost flow
ference being the rate of entropy generation within the boundary _ h1), is defined for each flow in a system. A cost balance is
_ rate, C($
due to associated irreversibilities. Similarly, in the equation (1d), Ex
written for each component to provide exergy costing as follows:
is the exergy flow rate and it shows that exergy transferred out of
the boundary must be less than the rate in which exergy enters, the X X
C_ q;k þ C_ i;k þ Z_ k ¼ C_ e;k þ C_ w;k (5)
difference being the rate of exergy destruction (or lost work) within e
i
the boundary due to associated irreversibilities which can be
calculated by the dead-state temperature (T0) multiplied by the where
entropy generation rate as given in equation (1f). The minimum
exergy destruction, or minimum entropy generation, characterizes C_ j ¼ cj Ex
_
j (6)
a system with minimum destruction of available work, which in the
By combining exergy and exergoeconomic balance equations,
case of refrigeration plants, is equivalent to the design with a
the following equation can be obtained:
maximum refrigeration load, or minimum mechanical power input
[12]. In cooling systems, T0 usually equals to the temperature of the _ _ _
Ex F;k ¼ ExP;k þ ExD;k (7)
high-temperature medium TH.
In addition, the specific flow exergy associated with the coolant The cost rate of exergy destruction is defined as follows:
medium is given below
C_ D;k ¼ cF;k Ex
_
D;k (8)
1
excoolant ¼ ðh  h0 Þ þ V 2 þ gZ  T0 ðs  s0 Þ (2) Here, the component exergy destruction costs are determined
2 by evaluating the exergy destruction rates associated with each
component ðEx_
D;k Þ with respect to the previous exergy balance
where the kinetic and potential terms are negligible. equations. Moreover, from Equation (5), the steady state form of the
For the entire cooling system, the energetic coefficient of per- control volume cost balance can be written as given in Equation (9)
formance (COP) becomes below. The cost balances are generally written so that all terms are
positive.
Q_ evap þ Q_ ch X  X 
COPen; system ¼ (3) _ e _ ¼ c Ex _ _
_
W comp þ W _ pump ce Ex þ cw;k W k q;k q;k þ ci Exi þ Z_ k (9)
k k
e i
Finally, for the thermodynamic analysis, using the aforemen-
tioned exergy equations, the exergetic COP of the system can be The above equation states that the total cost of the exiting
calculated as exergy streams equals the total expenditure to obtain them, namely
646 H.S. Hamut et al. / Journal of Power Sources 247 (2014) 643e654

the cost of the entering exergy streams plus the investment costs the component with its output streams. In addition, there is also the
[13]. The investment cost correlations associated with each component-related environmental impact that is associated with
component is determined in Ref. [14]. the life cycle of each component. In order to conduct an exer-
Cost balances for each component are needed to be solved in goenvironmental analysis, an environmental impact balance is
order to estimate the cost rate of exergy destruction in each written for each component to provide environmental impact for-
component. In the cost balance equations with more than one inlet mation as follows:
or outlet flow, the number of unknown cost parameters exceeds the
X X
number of cost balances for that component. Thus, auxiliary exer- B_ q;k þ B_ i;k þ Y_ k ¼ B_ e;k þ B_ w;k (11)
goeconomic equations created by F (fuel) and P (product) rules are i e
used to equate the number of unknowns with the number of
equations [15]. Implementing Equation (9) for each component where Y_ k is the component related environmental impact associ-
together with the auxiliary equations form a system of linear ated with the life cycle of the component, which is an indicator of
equations as follows: the reduction potential of environmental impact of the component.
The environmental balance equation states that the sum of all
h i h i environmental impacts associated with all input streams plus the
_ _
Ex k  ½ck  ¼ Z k (10) component-related environmental impact is equal to the sum of
environmental impacts associated with all output streams.
where the equation entails matrices of exergy rate (from exergy In order to be able to solve the environmental balance equa-
analysis), exergetic cost vector (to be evaluated) and the vector of tions, the environmental impacts associated with each component
Z_ k factors (from economic analysis) respectively [7]. The matrix are determined with respect to Eco-indicator 99 points, which
form of the equation for each equation is given below: enable a fair comparison among different components. These

2 3 2 3 2 3
1 1 0 0 0 0 0 0 0 0 _ comp
W 0 0 0 C_ 1 Z_ comp
6 0 7 6 7
6 1 1 0 0 0 0 0 0 0 0 0 1 6 _
1 7 6 C 2 7 7 6 Z_ cond 7
6 0 _ _
Ex 0 7 6 _ 7 6 7
6 Ex 3 2 0 0 0 0 0 0 0 0 0 0 7 6 C3 7 6 0 7
6 7 6 _ 7 6 7
6 0 0 1 1 0 0 0 0 0 0 0 0 0 0 7 6 C4 7 6 Z_ etxv 7
6 _ _ 7 6 7
6 Ex4 0 0 Ex 0 0 0 0 1 0 0 0 0 0 7 6 C_ 5 7 6 Z_ evap 7
6 1
7 6
6
7
7 6 7
6 1 0 0 1 0 0 0 0 0 0 0 0 0 0 7 6 C_ 6 7 6 0 7
6 7 6 _ 7 6 7
6 0 0 1 0 1 0 0 0 0 0 0 0 0 0 7 6 C7 7 6 Z_ ctxv 7
6 7 _ 7 ¼ 6 7
6 1 0 0 0 1 1 1 0 0 0 0 0 0 0 7 6 C8 7 6 Z_ chil 7
6 7 6
6 _ 7 6 7
6 0 0 0 0 0 1 1 0 0 0 0 0 0 0 7 6 C9 7 6 0 7
6 _ pump 7 6_ 7 6 _ 7
6 0 0 0 0 0 0 1 1 0 0 0 W 0 0 7 6 C 10 7 6 Z pump 7
6 _ 7 6_ 7 6 7
6 0 W 0 7 C 11 7 6 Z_ bat 7
6 0 0 0 0 0 0 1 0 bat 0 0 0 7 6
6 _ 7 6 7
6 0 0 0 0 0 0 0 0 0 0 1 0 0 7 C
0 7 6 12 7 6 Z_ 7
6 4 _ 5 6 elect 7
4 0 0 0 0 0 0 0 0 0 0 0 _ pump
W _ comp
W 0 5 C 13 4 0 5
0 0 0 0 0 0 0 0 0 0 0 0 1 0 C_ 14 0

Here, celect is the unit cost of electricity, which is taken as impact points are approximated with respect to a combination of
0.075 $ kWh1 [16]. By solving these equations, the cost rate of each correlations created from numerous studies conducted in the
flow can be calculated, which can be used to determine the cost rate literature, available data as well as the LCA created for this study.
of exergy destruction in each system component. Even though EVs and HEVs can be a part of the solution related to
environmental issues such as urban air pollution and global warming
3.3. Exergoenvironmental analysis compared to the conventional vehicles with ICEs, when the EVs and
HEVs are evaluated, there are still environmental concerns associated
Exergoenvironmental analysis is considered to be one of the most with the electric battery itself [19]. Thus, determining the battery
promising tools to evaluate energy conversion processes from an environmental impact plays a significant role in accurately assessing
environmental point of view [17,18]. In order to be able to perform the the overall environmental impact of the system.
analysis, the allocation of environmental analysis results to exergy In the environmental analysis, LCA is carried out in order to
streams is performed analogously to the allocation of exergy stream obtain the environmental impact of the battery assembly. It is a
costs in exergoeconomics. Initially, an environmental impact rate B_ j is cradle to grave approach to study the environmental aspects
expressed in terms of Eco-indicator 99 points, which are determined throughout a product’s life from raw material acquisition through
through a combination of Sima Pro 7 analysis and available infor- production, use and disposal and provides quantitative data to
mation in the literature. Subsequently, these points are converted into identify the potential environmental impacts of the material and/or
hourly rates (mPts h1), based on 4 h of driving for 15 years. production on the environment [20]. For the LCA analysis,
The values for internal and output streams can only be obtained numerous damage categories are covered and the results are
by considering the functional relations among system components, weighted and expressed in terms of Eco-indicator points with a
which are obtained by formulating environmental impact balances European electric generation mix and hierarchist perspective (H/H)
and auxiliary equations. The basis for formulating impact balances is weighting set [21] by using software package SimaPro 7.1. The
that all environmental impacts entering a component have to exit analysis is conducted with respect to 1 kg of battery and later scaled
 H.S. Hamut et al. / Journal of Power Sources 247 (2014) 643e654 647

up to the full size of the considered battery assembly (197 kg). To be 4.1. Objective functions
consistent in the analysis, the eco-indicator points are converted to
hourly rates (mPts h1), based on 4 h of driving for 15 years. A multi-objective optimization problem requires the simul-
Environmental impact balances for each component are needed taneous satisfaction of a number of different and usually con-
to be solved in order to estimate the environmental impact rate of flicting objectives characterized by distinct measure of
exergy destruction in each component. For the balance equations performance. It should be noted that multi-objective optimiza-
with multiple inlet and outlet flows, auxiliary exergoenvironmental tion problems generally show a possible uncountable set of so-
equations (analogous to exergoeconomic equations) are created to lutions which represents the best possible trade-offs in the
match the unknown impact parameters with the number of envi- objective function space. No combination of decision variable
ronmental impact balance equations. Implementing Equation (11) values can minimize/maximize all the components of functions
for each component together with the auxiliary equations form a simultaneously [5]. In this study, the objective functions
system of linear equations as follows: considered for multi-objective optimization are the combinations
h i h i of exergy efficiency (to be maximized), the total cost rate of
_ _
k  ½bk  ¼ Y k
Ex (12) product (to be minimized) and environmental impact (to be
minimized). They are compared against single-objective optimi-
where the equation entails matrixes of exergy rate, environmental zations of these objectives. Consequently, the objective functions
impact vector and the vector of Y_ k factors respectively. The matrix in the hybrid electric vehicle thermal management system
form of the equation for each equation can be seen below: analysis can be expressed in Equations (13)e(15). Even though

2 3 2 3 2 3
1 1 0 0 0 0 0 0 0 0 _ comp
W 0 0 0 B_ 1 Y_ comp
6 0 7 _B 7 6 7
6 1 1 0 0 0 0 0 0 0 0 0 1 1 7 6 6 2 7 6 Y_ cond 7
6 0 _ _ 0 7 _ 7 6 7
6 Ex Ex 0 0 0 0 0 0 0 0 0 0 7 6 B 6 0 7
6 3 2
7 6 _
3 7
6 7
6 0 0 1 1 0 0 0 0 0 0 0 0 0 6
0 7 6 B4 7 7 6 Y_ etxv 7
6 _ _ 7 6 7
6 Ex4 0 0 Ex 0 0 0 0 1 0 0 0 0 0 7 6 B_ 5 7 6 Y_ evap 7
6 1
7 6 6 _
7
7 6 7
6 1 0 0 1 0 0 0 0 0 0 0 0 0 0 7 6 B6 7 6 0 7
6 7 6 7
6 0 0 1 0 1 0 0 0 0 0 0 0 0 0 7 6 6 B_7 7
7 6 Y_ ctxv 7
6 7  ¼ 6 7:
6 1 0 0 0 1 1 1 0 0 0 0 0 0 0 7 6 B_ 7 6 Y_ chil 7
6 7 6 _8 7 6 7
6 0 0 0 0 0 1 1 0 0 0 0 0 0 0 7 6 B 7 6 0 7
6 _ pump 7 6 6 _
9 7
7 6 _ 7
6 0 0 0 0 0 0 1 1 0 0 0 W 0 0 7 6 B10 7 6 Y pump 7
6 _ 7 6_ 7 6 7
6 0 W 0 7 B11 7 6 Y_ 7
6 0 0 0 0 0 0 1 0 bat 0 0 0 7 6 6 _ 7 6 bat 7
6 0 0 0 0 0 0 0 0 0 0 1 0 0 0 7 6 B 7 6 b 7
6 7 4_ 5 12 6 elect 7
4 0 0 0 0 0 0 0 0 0 0 0 _ pump
W _ comp
W 0 5 B13 4 0 5
0 0 0 0 0 0 0 0 0 0 0 0 1 0 B_ 14 0

Here, belect is the unit environmental impact associated with the each objective function varies in terms of the objective it is
electricity generation mix used (from Eco-indicator 99), which is optimizing, they all have the same underlying parameters which
taken as 22 mPts kWh1. By solving these equations, the environ- are affected by the changes in the selected decision variables. All
mental impact rate of each flow can be calculated, which can be objectives in the multi-objective optimization are assumed to be
used to determine the environmental impact rate of exergy equally important, and therefore no additional weighting criteria
destruction in each system component. are assigned to the objectives in order to minimize subjectivity in
the analysis. Instead, the LINMAP method is used where the
point on the Pareto optimal frontier closest to an ideal un-
4. Multi-objective optimization
reachable point (where all selected objectives are optimized) is
selected as the single best optimization point.
In the previous sections, a hybrid electric vehicle thermal
Exergy efficiency:
management system is analyzed with respect to energy and exergy
efficiencies. Moreover, investment and exergy destruction costs
associated with the TMS are calculated and a cost formation of the _ _
Ex _ _
þ Ex
Q Q
system is provided. Furthermore, an environmental analysis is also jsystem ¼ _ evap _ ch (13)
W comp þ W pump
conducted using a cradle to grave life cycle assessment (with Eco-
indicator 99) using SimaPro 7.1 as well as by creating environ-
where the inputs are the work of the compressor and the pump and
mental impact correlations from the literature in order to point out
the outputs are the exergy of heat with respect to the evaporator
the components causing the highest environmental impact.
and the chiller.
In this section, the TMS will be optimized using a multi-
Total cost rate:
objective evolutionary algorithm which considers exergetic, exer-
goeconomic and exergoenvironmental objectives with respect to
the decision variables and constraints. A Pareto frontier will be C_ system ¼ Z_ k þ C_ D;k (14)
obtained for the system and a single desirable optimal solution will
be selected based on a linear programming technique for multidi- where the total cost rate of the system consists of the total in-
mensional analysis of preference (LINMAP) decision making vestment cost and cost of exergy destruction respectively.
processes. Environmental impact:
648 H.S. Hamut et al. / Journal of Power Sources 247 (2014) 643e654

B_ system ¼ B_ k þ B_ D;k
Table 1
(15)
Constraints associated with the decision variables selected for the TMS.

where the total environmental impact of the system consists of the Constraints

component-related environmental impact and the impact associ- 0 C  Tevap  Tcabin  DTevap.min
ated with exergy destruction respectively. The environmental T0 þ DTcond.min  Tcond  65  C
Tevap < Tevap,air,in  DTevap.min  Tsh
impact points are determined from LCA conducted using SimaPro
Tcond > Tcond,air,in þ DTair.min þ DTcond.min þ DTsc
7.1 along with various correlations developed from the data avail- 0  C  DTsh  10  C
able in the literature. 0  C  DTsc  10  C
hcomp  0.95
hpump  0.95
4.2. Decision variables and constraints m_ air  0:35 kg s1

In this study, the following six decision variables are chosen for
the analysis:
Holland in the 1970s in order to simulate growth and decay of living

the condenser saturation temperature (Tcond), organisms in a natural environment and various improvements

the evaporator saturation temperature (Tevap), were conducted ever since. GAs today apply an iterative and sto-

the magnitude of superheating in the evaporator (DTsh), chastic search strategy to drive its search towards an optimal so-

the magnitude of subcooling in the condenser(DTsc), lution through mimicking nature’s evolutionary principles and

the _ e Þ,
evaporator air mass flow rate ðm have received increasing attention by the research community as

the compressor efficiency (hcomp). well as the industry to be used in optimization procedures.
Based on the inspired evolutionary process, the weak and unfit
In engineering applications of the optimization problems, there species are faced with extinction while the strong ones have greater
are usually constraints on the trade-off decision variables that arise opportunity to pass their genes to future generations via repro-
from appropriate feasibility, commercial availability and engineer- duction. Throughout this process, given a long enough time line,
ing constraints [2]. The limitations on the minimum and maximum the species carrying the suitable combination in their genes
ranges of decision variables are given in Table 1. become the dominant population.
As can be seen from Table 1, the lower bound for the evaporator In the analysis, the GA terminology adopted by Konak et al. [24]
temperature is taken to be higher than 0  C since lower tempera- is used. Based on this terminology, a solution vector is called an
tures would cause icing on the surface of the evaporator due to the individual or a chromosome, which consists of discrete units called
formation of the water droplets. This reduces the volume of air genes. Each gene controls one or more features of the chromosome,
flowing through the evaporator and in turn reduces the efficiency which corresponds to a unique solution in the solution space.
of the system [22]. On the other hand, the upper bound of the Moreover, the collections of these chromosomes are called a pop-
evaporator is limited by the cabin cooling temperatures. For ulation, which are initialized randomly at first and includes solu-
condenser, the lower temperature bound is based on the ambient tions with increasing fitness as the search evolves until converging
temperature, whereas the upper bound is constrained with respect to a single solution. Furthermore, operators called crossover and
to the compression ratio of the compressor, since very high mutation are used to generate new solutions from existing ones.
compression ratios increase the probability of the high pressure Crossover is one of the key operators where two chromosomes,
vapor to leaking back to the low pressure side and even causing called parents, are combined together to form new chromosomes
compressor failure. Moreover, constraints are provided between called offspring. Due to the having preference towards fitness, these
the evaporator and condenser temperatures and the incoming air offsprings will inherit good genes from the parents and through the
temperatures in order to have feasible and adequate heat transfer iterative process, the good genes are expected to appear more
in the heat exchangers. Furthermore, the compressor and pump frequently in the population, where they eventually converge to an
efficiencies are limited to 0.95 due to previously mentioned tech- overall good solution.
nological limitations, whereas the air mass flow rates are limited The mutation operator on the other hand introduces random
with respect to the vehicle speed and fan power. changes into the characteristics of the chromosomes at the gene
level. Usually the mutation rate (probability of changing properties
4.3. Genetic algorithm of a gene) is very small and therefore the new chromosome pro-
duced will not be very different than the original one. While the
Currently, there are many search techniques that are used to crossover leads the population to converge (by making the chro-
solve multi-objective optimization problems. These include, but are mosome in the population alike), the mutation reintroduces ge-
not limited to, generic algorithm, simulated annealing, tabu and netic diversity and assists to the escape from local optima [24].
scatter search, ant system, particle swarm and fuzzy programing. Reproduction involves selection of chromosomes for the next
Among these, there is no technique that provides the optimum generation, where the fitness of an individual usually determined
results for all problems and thus the best method should be the probability of its survival. The selection procedures can vary
selected with respect to the current system. In this research, a depending on how the fitness values are used (such as proportional
generic algorithm is used since it requires no initial conditions, selection, ranking and tournament). The basic schematic for the
works with multiple design variables, finds global optima (as evolutionary algorithm for the case used in the study is given in
opposed to local optima), utilizes populations (as opposed to in- Fig. 2.
dividuals) and uses objective function formation (as opposed to The GA has major advantages since constraints of any type can
derivatives). be easily implemented and that they can find more than one near-
In the past few decades, genetic algorithms (GAs) have been optimal point in the optimization space, which enables users to
extensively used as search and optimization tools in various pick the most applicable solution for the specific optimization
problem domains due to their broad applicability, ease of use and problem and therefore are widely used for various multi-objective
global perspective [23]. The concept of GAs was first conceived by optimization approaches. [1,7,25].
 H.S. Hamut et al. / Journal of Power Sources 247 (2014) 643e654 649

Even though maximizing/minimizing a criterion would be

beneficial, many real-world problems involve multiple measures of
performance, or objectives, which should be optimized simulta-
neously. Objectives that are optimized individually can provide
optimal results with respect to their own criteria while providing
very low performance in other objective functions. Thus, a trade-off
is needed among the different dimensions in order to obtain a
family of optimal “acceptable” solutions for the problem [26]. This
ability along with not requiring the user to prioritize, scale or weigh
objectives makes them unique in solving multi-objective optimi-
zation problems.
The first real application of EAs for finding multiple trade-off
solutions in one single simulation run was suggested and used by
Schaffer in 1984 [27]. Schaffer used a vector-evaluated genetic al-
gorithm (VEGA) to capture multiple trade-off solutions for a small
number of iterations. This was followed by Goldberg in 1989 [23] Fig. 3. A general Pareto optimal curve.

who suggested a 10-line sketch of a plausible multi-objective

evolutionary algorithm optimization using the concept of domi- 5. Results and discussion
nation. Consequently, many different implementations of MOEAs
have been developed such as weight-based GA, non-dominating A software code in Engineering Equation Solver (EES) was
sorting GA, Pareto-GA (NPGA), fast non-dominating sorting constructed to analyze a baseline model, with respect to afore-
generic algorithm (NSGA-II) and multi-objective evolutionary al- mentioned balance equations and system parameters. Based on the
gorithm along with different ways of using EAs to solve multi- baseline analysis, the exergy efficiencies and exergy destruction
objective optimization problems such as diploidy, weight-based rates associated with each component are provided in Fig. 4.
and distance based approaches. Among these components, the heat exchangers have the lowest
One of the most prominent differences of classical search exergy efficiencies with respect to the high temperature differences
and optimization algorithms is that EAs use population of so- between the refrigerant, coolant and the cabin air, along with phase
lutions in each iteration (instead of single solutions), which change in the refrigerant which results in more entropy generation
produces a final outcome of a population of multiple non- between the cooling media.
dominated solutions (that are in parallel) by taking advantage When evaluated from an exergoeconomic point of view, the
of similarities in the family of possible solutions. Usually EAs do components that have the highest priority are the ones that have
not converge in a single solution (due to conflicting criteria). EA the highest sum of total capital investment and exergy destruction
captures multiple optimum solutions in its final population. cost rate ðZ_ þ C_ D Þ. These values for each component are provided in
These solutions are called “Pareto optimal”, where no other Table 2. Based on these results, the electric battery has the highest
feasible solution can reduce some objective function without total cost rate compared to the rest of the components, mainly due
causing a simultaneous increase in at least no other objection to having significantly larger investment costs. After the battery,
function. The objective function values corresponding to these the highest sum of total capital investment rate and cost rate of
feasible non-dominating solutions are called “Pareto optimal exergy destruction are determined to be the compressor followed
frontier” [2,24,26,28]. The general concept of Pareto optimal by the evaporator and condenser. These components are followed
frontier is illustrated in Fig. 3. by the pump and thermal expansion valves which have relatively
insignificant cost rates compared to the rest of the system
components.
When evaluated from an exergoenvironmental point of view,
the most important component would be the one with the highest
sum of component-related environmental impact and environ-
mental impact due to exergy destruction rate ðY_ þ B_ D Þ. The
component-related environmental impact associated with the
electric battery is calculated with respect to the conducted LCA. The
major battery components along with their calculated Eco-
indicator 99 impact points are provided in Table 3. The environ-
mental impacts associated with the rest of the components are
determined with respect to impact correlations determined from
the literature. The component and exergy destruction related
environmental impact rates and for all components are provided in
Table 4. When the components are analyzed with respect to Y_ þ B_ D ,
the electric battery by far has the highest environmental impact
mainly due to the high copper mass used in the lithium-ion battery
anodes. The evaporator, compressor and condenser have the next
highest environmental impacts respectively, followed by the
evaporator TXV and the chiller where the environmental impact is
significantly lower. Finally, the chiller TXV and the pump impacts
are found to be exergoenvironmentally insignificant compared to
the aforementioned components.
Subsequently, multi-objective optimizations with aforemen-
Fig. 2. Sample schematic for the evolutionary algorithm. tioned objective functions (Equations (13)e(15)), constraints
650 H.S. Hamut et al. / Journal of Power Sources 247 (2014) 643e654

Table 3
Component-related and exergy-related environmental impacts, and total environ-
mental impact associated with the TMS components.

Component Y_ k B_ D;k Total env. impact

(mPts h1) (mPts h1) (mPts h1)

Compressor 0.90 9.47 10.37

Condenser 0.28 9.24 9.514
Evaporator 0.22 12.65 12.87
Chiller 0.13 1.18 1.31
Evaporator TXV 0.02 4.82 4.83
Chiller TXV <0.01 0.56 0.56
Pump 0.13 <0.01 0.13
Battery 33.78 3.95 37.72

traditional method called LINMAP decision-making [29] is used to

select a desirable final solution as shown in Figs. 8 and 9. This
method creates a hypothetical ideal point in which all objectives
have their corresponding optimum values independent of each
other and would stay below the Pareto optimum frontier. Even
though this point would be impossible in reality, it would serve a
useful purpose by assisting the decision makers to select the point
on the Pareto optimum frontier that has the closest distance to this
ideal point as the desirable final solution [5]. In Figs. 8 and 9, the
solutions along the Pareto optimum frontier where one objective is
maximized/minimized at the full expense of the other are marked
by “A” and “B”, whereas the points closest to the hypothetical ideal
point is marked by “C”.
Table 5 shows the values for the decision variables in the base
case design along with the four different optimization criteria. In
addition, the results of exergy, economic and environmental ana-
lyses for each optimization criteria are shown in Tables 5e8. It
should be noted that the values for the decision variables are
considered to be continuous over the determined constraints for
Fig. 4. Baseline model (a) exergy efficiency and (b) exergy destruction rate of each
the multi-objective optimization problem. However, usually pa-
component in the refrigerant and coolant cycles. rameters associated with some of these variables (especially size
and efficiency) are only available in discrete units. Therefore, in a
case where the determined parameter values are not available, the
(Table 1) and six decision variables are performed with the help of closest available values should be utilized in the system for most
genetic algorithms. In the analysis, five optimization scenarios with optimal results.
the objective functions of exergy efficiency (single-objective), total In these tables, it can be seen that each single objective opti-
cost rate (single-objective), environmental impact rate (single- mization approach pays attention only to its own criterion without
objective), along with exergoeconomic (multi-objective) and taken others into consideration. Exergetic single-objective optimi-
exergoenvironmental (multi-objective) optimizations are per- zation scenario maximizes the exergy efficiencies for each
formed. The corresponding optimization scenarios are illustrated in component; however no attention is paid to economic or envi-
Figs. 5e9. ronmental objectives. Similarly, an exergoeconomic single-
As previously mentioned, all the points on the Pareto optimum objective optimization scenario has the lowest unit costs for each
frontier are potentially an optimum solution for the analysis and component at the expense of exergy efficiency and environmental
therefore a weighting factor is needed to be assigned for each impact. Also, an exergoenvironmental single-objective optimiza-
objective and/or decision is needed to be made (often based on tion has the lowest Eco-indicator 99 points for each component at
experience or importance of each objective) in order to select a the expense of exergy efficiencies and cost. In the multi-objective
single final solution among them. In this selection process, a optimization scenario however, these objectives are considered

Table 2
Investment cost rate, cost rate of exergy destruction and total cost rate associated Table 4
with the TMS components. Major components in the LCA analysis and their corresponding weights per kg of Li-
ion battery.
Component Z_ k (¢ h1) C_ D;k (¢ h1) Total cost (¢ h1)
Component Weight (kg) Eco-indicator 99 pts
Compressor 2.7 3.2 5.9
Condenser 1.4 3.9 5.2 Electrode paste 0.25 0.199
Evaporator 0.9 5.4 5.7 Cathode 0.034 0.033
Chiller 0.6 0.5 1.2 Anode 0.083 2.001
Evaporator TXV 0.1 2.0 0.8 Electrolyte 0.120 0.043
Chiller TXV 0.0 0.2 0.2 Casing 0.201 0.187
Pump 0.4 <0.1 0.4 Module packaging 0.170 0.102
Battery 3.5 3.2 6.4 BMS 0.029 1.650
 H.S. Hamut et al. / Journal of Power Sources 247 (2014) 643e654 651

Fig. 5. Single objective optimization of TMS over generations with respect to exergy efficiency with best and mean fitness values of 0.321 and 0.318, respectively.

Fig. 6. Single objective optimization of TMS over generations with respect to product cost rate with best and mean fitness values of 0.248 and 0.249, respectively.

Fig. 7. Single objective optimization of TMS over generations with respect to product cost rate with best and mean fitness values of 64.9 and 65.0, respectively.
652 H.S. Hamut et al. / Journal of Power Sources 247 (2014) 643e654

Fig. 8. Multi-objective optimization of TMS with respect to exergy efficiency and total cost rate.

Fig. 9. Multi-objective optimization of TMS with respect to exergy efficiency and total environmental impact rate.

simultaneously. They provide optimized solutions with values in 6. Conclusions

between the extremes yielded by the single-objective approaches
as a result of the trade-offs made between the solutions of the two In this study, the TMS of a hybrid vehicle is optimized using a
conflicting objectives. Normalized values of the objectives with multi-objective evolutionary algorithm using exergoeconomic and
respect to each optimization criteria are provided in Fig. 10. exergoenvironmental objectives. The optimization is performed in
Moreover, when the exergoeconomic and exergoenvironmental order to maximize the exergy efficiency (based on exergetic effi-
optimizations are compared against the results using energy effi- ciency), minimize the unit exergy cost (based on cost of unit exergy
ciencies, the selected values for the decision variables in the LIN- destruction and investment costs) and unit environmental impact
MAP optimization points are determined to have a 4.8% lower cost (based on Eco-Indicator 99 impact points). Condensing and evap-
and a 3.9% lower environmental impact rates than the one calcu- orating, superheating and subcooling temperatures, evaporator air
lated by the energy approach, which yields total cost and envi- mass flow rate and compressor efficiency are selected as the deci-
ronmental impact rates of 1.41 $ h1 and 87.27 mPts h1 sion variables for the analyses and various constraints are applied
respectively. based on appropriate feasibility and engineering constraints. The

Table 5
Decision variables for the base case design under various optimization criteria.

Decision Base case Single-obj. Single-obj. Single-obj. Multi-obj. Multi-obj.

variable design exergetic economic environmental exergoeconomic exergoenvironmental

Tcond ( C) 55 55.23 54.20 55.18 56.01 55.25

Tevap ( C) 5 0.40 8.94 8.92 8.93 8.82
DTsh ( C) 5 4.75 3.86 2.40 9.69 0.96
DTsc ( C) 5 9.94 9.68 4.90 9.99 1.71
m_ e ðkg s1 Þ 0.17 0.35 0.19 0.18 0.21 0.25
hcomp 0.63 0.80 0.66 0.79 0.72 0.79
 H.S. Hamut et al. / Journal of Power Sources 247 (2014) 643e654 653

Table 6
Exergetic analysis results for the base case design under various optimization criteria.

Decision Base case Single-obj. Single-obj. Single-obj. Multi-obj. Multi-obj.

variable design Exergetic economic environmental exergoeconomic exergoenvironmental

hex,comp 0.67 0.82 0.69 0.81 0.75 0.81

hex,cond 0.22 0.24 0.26 0.21 0.23 0.20
hex,evap 0.24 0.21 0.27 0.27 0.26 0.28
hex,chil 0.37 0.32 0.43 0.42 0.43 0.42
hex,etxv 0.89 0.88 0.92 0.91 0.92 0.88
hex,ctxv 0.89 0.89 0.92 0.91 0.92 0.89
hex,pump 0.80 0.83 0.84 0.80 0.78 0.77
hex,bat 0.81 0.83 0.85 0.83 0.84 0.83

Table 7
Economic analysis results for the base case design under various optimization criteria.

Decision Base case Single-obj. Single-obj. Single-obj. Multi-obj. Multi-obj.

variable ($ h1) design exergetic economic environmental exergoeconomic exergoenvironmental

Z_ comp 5.90 20.38 5.10 6.87 5.57 9.80

Z_ cond 5.32 10.96 5.08 5.31 5.46 7.27
Z_ evap 6.32 18.39 5.75 6.44 6.39 9.23
Z_ chil 1.06 1.38 0.92 1.01 0.91 1.04
Z_ etxv 2.13 5.56 1.47 2.15 1.53 3.68
Z_ ctxv 0.25 0.33 0.16 0.25 0.15 0.30
Z_ pump 0.38 0.38 0.38 0.38 0.38 0.38
Z_
bat 6.67 7.45 6.38 6.66 6.37 6.74

Table 8
Environmental analysis results for the base case design under various optimization criteria.

Decision Base case Single-obj. Single-obj. Single-obj. Multi-obj. Multi-obj.

variable (mPts h1) design exergetic economic environmental exergoeconomic exergoenvironmental

B_ comp 9.34 9.13 8.58 4.93 9.11 6.61

B_ cond 8.48 12.62 8.77 7.10 9.61 8.64
B_ evap 12.83 24.05 10.49 9.09 14.06 12.54
B_ chil 1.31 1.29 0.98 0.92 0.90 0.95
B_ etxv 4.82 7.56 3.22 3.55 5.12 6.01
B_ ctxv 0.57 0.44 0.35 0.41 0.31 0.49
B_ pump 0.14 0.14 0.14 0.14 0.14 0.14
B_
bat 37.73 37.42 37.01 36.79 36.77 35.67

the single-objective approaches provided optimal solutions for

their objectives, they have provided poor solutions for the
remaining objectives. Thus, the multi-objection optimization
approach provided a solution set within the extremes of the
single-objective results by evaluating two objectives simulta-
neously and providing trade-off between them to obtain desirable
solution sets.
Based on the analyses, the following specific concluding re-
marks can be made

The exergy efficiency, total cost rate and environmental impact

for the baseline system is determined to be 0.29, ¢28 h1 and
77.3 mPts h1 respectively. The exergy efficiency could be
Fig. 10. Normalized values of different objectives with respect to various optimization
increased by up to 27% (by single objective exergy) and the cost
functions.
and environmental impact can be reduced by 10% (by single
objective cost) and 19% (by single objective environmental
impact) respectively, at the expense of the non-optimized
decision variables along with exergy efficiency, total cost and outputs.
environmental impact (for each component) are compared under
Based on the exergoeconomic optimization, it is concluded that
each optimization approach. a 14% higher exergy efficiency and 5% lower cost can be ach-
In the multi-objective optimization, a Pareto frontier is ob- ieved, compared to baseline parameters at an expense of 14%
tained and a single desirable optimal solution is selected based on increase in the environmental impact.
LINMAP decision-making process. The corresponding solutions are
Based on the exergoenvironmental optimization, a 13% higher
compared against each exergetic, exergoeconomic and exer- exergy efficiency and 5% lower environmental impact can be
goenvironmental single objective optimization result. Even though achieved at the expense of 27% increase in the total cost.
654 H.S. Hamut et al. / Journal of Power Sources 247 (2014) 643e654

Acknowledgments tot total

txv thermal expansion valve
Financial support from Automotive Partnerships Canada (APC) w work
and the Natural Sciences and Engineering Research Council of
Canada (NSERC) is gratefully acknowledged. Acronyms
A/C air conditioning
Nomenclature COP coefficients of performance
EV electric vehicle
b specific environmental impact (mPts kJ1) HEV hybrid electric vehicle
B_ environmental impact rate (MPts h1) ICE internal combustion engine
c cost per unit of exergy ($ kJ1) LCA life cycle assessment
C_ cost rate associated with exergy ($ h1) LINMAP linear programming technique for multidimensional
E_ energy rate (kW) analysis of preference
_
Ex exergy rate (kW) TMS thermal management system
h specific enthalpy (kJ kg1) TPIM traction power inverter module
m _ mass flow rate (kg s1 or L min1) TXV thermal expansion valve
P pressure (kg m1 s2)
Q_ heat transfer rate (kW)
s specific entropy (kJ kg1 K1) References
S_ entropy (kJ K1)
[1] I. Dincer, Int. J. Energy Res. 26 (2002) 717e727.
T temperature (K or  C) [2] I. Dincer, Int. J. Energy Res. 27 (2003) 687e702.
T0 ambient temperature (K or  C) [3] C. Zamfirescu, I. Dincer, M. Stern, W.R. Wagar, Int. J. Energy Res. 36 (2012)
V velocity (m s1) 397e408.
_ [4] A. Lazzaretto, A. Toffolo, Energy 29 (2004) 1139e1157.
W work rate or power (kW) [5] H. Sayyaadi, M. Babaelahi, Appl. Energy 88 (2011) 3012e3021.
Y_ component-related environmental impact rate associated [6] B.H. Gebreslassie, G. Guillén-Gosálbez, L. Jiménez, D. Boer, Appl. Energy 86
with LCA (mPts h1) (2009) 1712e1722.
[7] P. Ahmadi, I. Dincer, M.A. Rosen, Energy 36 (2011) 5886e5898.
Z_ cost rate associated with the sum of capital investment [8] J.M. Jabardo, W.G. Mamani, M.R. Ianekka, Int. J. Refrig. 25 (2002) 1157e1172.
($ h1) [9] S.W. Wang, J. Gu, T. Dickson, T. Dexter, I. McGregor, Exp. Therm. Fluid Sci. 30
Z height (m) (2005) 59e66.
[10] Behr GmbH & Co. KG, Press Official Website, Technical Press Day, www.
behrgroup.com/Internet/behrcms_eng.nsf, 2009, accessed October 01, 2012.
Greek symbols [11] H.S. Hamut, I. Dincer, G.F. Naterer, Energy 46 (2012) 17e125.
D change in variable [12] A. Bejan, Thermodynamic Optimization of Heat Transfer and Fluid Flow Pro-
cesses. Developments in the Design of Thermal Systems, Cambridge Univer-
j exergy sity Press, Cambridge U.K., 1997.
[13] A. Abusoglu, M. Kanoglu, Appl. Therm. Eng. 29 (2009) 234e241.
[14] H.S. Hamut, I. Dincer, G.F. Naterer, Int. J. Energy Res. 37 (2013) 1e12.
[15] A. Bejan, Entropy Generation Minimization: the Method of Thermodynamic
Subscripts
Optimization of Finite-size Systems and Finite-time Processes, CRC Press LLC,
0 ambient Florida U.S., 1996.
act actual [16] Toronto Hydro, Electricity Rates and Charges, www.Torontohydro.com, 2012,
bat battery accessed on October 2012.
[17] L. Meyer, G. Tsatsaronis, J. Buchgeister, L. Schebek, Energy 34 (2009) 75e89.
cool coolant [18] A. Boyano, A.M. Blanco-Marogorta, T. Morosuk, G. Tsatsaronis, Energy 26
c,cond condenser (2011) 2202e2214.
ch chiller [19] J. Matheys, J.M. Timmermans, J. Van Mierlo, S. Meyer, P. Van den Bossche, Int.
J. Sust. Manuf. 1 (2009) 318e329.
comp compressor [20] ISO 14040, Environmental Management e Life Cycle Assessment e Principals
D destruction and Framework, International Organization for Standardization, Geneva,
e exit Switzerland, 2006.
[21] F. Petrakopoulou, A. Boyano, M. Cabrera, G. Tsatsaronis, Int. J. Greenhouse Gas
elect electricity Control 5 (2011) 475e482.
en energy [22] S. Daly, Automotive Air-conditioning and Climate Control Systems, first ed.,
ex exergy Elsevier Ltd, Burlington, U.S., 2008.
[23] D.E. Goldberg, Genetic Algorithms in Search, Optimization, and Machine
evap evaporator
Learning, first ed., Addison Wesley, 1989.
F fuel [24] A. Konak, D.W. Coit, A.E. Smith, Reliab. Eng. Safe 91 (2006) 992e1007.
g, gen generation [25] A. Ghaffarizadeh, Investigation on Evolutionary Algorithms Emphasizing Mass
Extinction. B.Sc. Thesis, Shiraz University of Technology, 2006.
i in
[26] C.M. Fonseca, P.J. Fleming, Evol. Comput. 3 (1995) 1e15.
k component [27] J.D. Schaffer, Some Experiments in Machine Learning Using Vector Evaluated
P product Genetic Algorithms. Ph. D. Thesis, Vanderbilt University, 1984.
q heat [28] K. Deb, Multi-objective Optimization Using Evolutionary Algorithms, John
Wiley & Sons, Inc, New York NY, U.S., 2001.
ref refrigerant [29] P.L. Yu, Multiple Criteria Decision Making: Concepts, Techniques and Exten-
s isentopic sions, Plenum, 1985.