This document is created with trial version of Document2PDF Pilot 2.5.82.

Solving the Multi-objective Nonlinear Problems for Heat Exchanger Network

Synthesis using Particle Swarm Optimization Technique

Soad M. Mohamed1, Omar. S. Soliman2, Assem A. Tharwat2,

Eman M. Gabr 1and Tahani S. Gendy1
Abstract:
Process synthesis problems mathematically represented as Multi-objective Mixed-
Integer Nonlinear Programming (MO-MINLP) models, are often irregular, large, non-convex
and difficult to get the overall optimum by traditional method. The Particle Swarm
Optimization (PSO) has very deep intelligent background, so it is suitable for science
computation and engineering applications. In this paper, the MO-MINLP problem was
modified, so as to be solved by (PSO) method. A PSO program has been developed and used
to optimize and design Heat Exchanger Networks (HENs). The developed PSO program and
GAMS program have been applied on a HEN literature problem and their results have been
compared with the published results. The PSO method achieved better results, so it can be
used to design an efficient HEN with minimum utility, minimum number of heat exchangers,
minimum total area of heat exchangers and so minimum overall annual cost.

Keywords: Particle swarm optimization, Multi-objective Mixed Integer Nonlinear

Programming, heat exchanger network synthesis

1. Introduction:
The difficulties associated with using mathematical optimization on large-scale
engineering problems have contributed to the development of alternative solutions. Nonlinear
programming and dynamic programming techniques, for example, often fail in solving NP-
hard problems with large number of variables and non-linear objective functions (Lovbjerg
M, 2002). Moreover these methods assume that goal and constraints are differentiable. To
overcome these problems, researchers have proposed evolutionary-based algorithms such as
the genetic algorithm (GA), analog neural networks, chaos optimization algorithm, ant colony
algorithm and particle swarm optimization(PSO) for searching near-optimum solutions to
problems. Evolutionary algorithms (EAs) are stochastic search methods that mimic the
natural biological evolution and/or the social behavior of species. EA works with a population
of potential solutions, where each individual within the population represents a particular
solution. The fitness value of the individuals expresses how good the solution is at solving the
problem. Such algorithms have been developed to arrive at near-optimum solutions to large-
scale optimization problems, for which traditional mathematical techniques may fail (Emad et
al, 2005).
In this paper, the particle swarm optimization (PSO) is reviewed in Section 2. In Section 3,
modification of the Multi-objective Mixed Integer Nonlinear Problem to Fit the PSO Method
is tackled. The Heat exchanger network synthesis (HENS) problem is defined and problem
formulation is depicted in Section 4. Applications of the PSO and GAMS to MO-MINLP
example are presented in Section 5. Finally we conclude the paper in Section 6.

2. Particle Swarm Optimization (PSO)—an overview

Particle swarm optimization (PSO) is an evolutionary computation technique developed
by Kennedy and Eberhart in 1995, which is similar to other population-based evolutionary

1
Egyptian Petroleum Research Institute (EPRI) Cairo.
2
Faculty of Computers and Information -Cairo University.

1
This document is created with trial version of Document2PDF Pilot 2.5.82.

algorithms in that the algorithm is initialized with a population of random solutions.

However, it is unlike most of other population-based evolutionary algorithms, in that PSO is
motivated by the simulation of social behavior instead of survival of the fittest and each
candidate solution is associated with a velocity. Because PSO has very deep intelligent
background, it is suitable for science computation and engineering applications. It has been
successfully applied to a wide range of applications, such as optimization problem, traveling
salesman problem, job scheduling, etc. One of the important factors that particle swarm
optimization is attractive is its simplicity that there are very few parameters to adjust. It can
achieve the optimal or near-optimal solutions in a rather short time without enormous
iterative computations in digital implementation.
In PSO each single solution of the problem is an agent (particle). These agents (particles)
are candidate solutions in the search space similar to the individuals in a population of the
genetic algorithms/evolutionary programming. Each agent (particle) carries the information
about the solution of the problem as its position in the solution space. The candidate
solutions, then “fly” through the search space. The velocity is constantly adjusted according
to the corresponding particle’s experience and the particle’s companions’ experience. It is
expected that the particles will move towards better solution areas. Therefore particle’s
position is responsible for its fitness in the solution space (Aruldoss and Ebenezer, 2005).
The steps of applying PSO method as follows:
Suppose that the search space is n-dimensional, then the i-th particle of the swarm can be
T
represented by a n-dimensional vector, X i  ( xi1 , xi 2 ,..., xin ) . The velocity (position change)
T
of this particle, can be represented by another n-dimensional vector Vi  ( vi1 , vi 2 ,..., vin ) The
T
best previously visited position of the i-th particle is denoted as Pi  ( pi1 , pi 2 ,..., pin ) . The
swarm is manipulated according to the following two equations (Eberhart et al., 1996):
 t t 
v t  1    wv  c r  p  x   c r  G  x  
t t t (1)
id  id 1 i1  i , best id  2 i 2  best id 
t 1  x t  v t
xid (2)
id id
where d = 1, 2, . . .,n; i = 1, 2, . . . , T pop , and T pop is the size of the swarm (Population

size); w is called inertia weight; c1 , c 2 are two positive constants, called cognitive and social
parameter respectively; and χ is a constriction factor, which is used, alternatively to w to limit
velocity. p is the best position of the i-th individual, G is the best position among
i, best best
the individuals (group best).; ri1 , ri 2 are random numbers, uniformly distributed in [0, 1];

and t = 1, 2, . . ., N iter , N iter is the maximum iteration number (Parsopoulos, K., E. and
Vrahatis, M. N., 2005).
In this research, the constriction factor χ is setting as a function of the iteration number and
the magnitude of the particle’s velocity, instead of using a constant value like Vmax , this
function is defined by ( Omer,2005) as:
  f (t , m)  m.r (1  t N iter ) (3)

2
This document is created with trial version of Document2PDF Pilot 2.5.82.

where r is a random numbers, uniformly distributed within the range [0, 1]; m is the
magnitude of the particle’s velocity. This function returns a value in range [0, m], which lead
to the probability of χ value being close to 0 as t increase.
The weighting factor ‘w’ is modified using Eq. (4) to enable quick convergence (Kannan.
S. et al, 2004):
wmax  wmin
w  wmax  t (4)
N iter
where w max is the initial weight, w min the final weight.

3. Modifying the Multi-objective Mixed Integer Nonlinear Problem to Fit the PSO
Method
Multi-objective Mixed Integer Nonlinear problems with k conflicting objective functions,
n variables and m constraints may be written as the following:
min  f1 ( X , Y ), f 2 ( X , Y ),..., f k ( X , Y ) 
h j ( X ,Y )  0 j  1,2,..., m1
g j ( X ,Y )  0 j  m1  1, m1  2,..., m
s.t. (5)

X  X / X  R n1 , X L  X  X U 

Y  Y / Y  0,1 2
n

Where f i ( X , Y ) is the i-th objective function, h j ( X , Y )  0 are the m1 equations that

describe the performance of the system, and g j ( X , Y )  0 are the m 2 inequalities constraints.

The variables X are n1 continuous variables, for which generally X L are lower bounds and

X U are upper bounds, while Y are the n2 discrete variables, which generally are restricted to
take 0-1 values to define binary variables (Ying, (2005), Bi Rongshan &Yang (2003)).
The following three methods have been used to adapt the MO-MINLP problem to PSO:
1) Weighting method
k k
F ( X , Y )   wi f i ( X , Y ), 0  wi  1, w i 1 (6)
i 1 i 1
2) Penalty function has been used to eliminate the implicit constraints (Bi Rongshan, Yang,
2003).
3) A sigmoid ([]) function is introduced to deal with the discrete variables. In process
synthesis problem, the discrete variables are always 0-1 variables, so the sigmoid function is
used to constrain the variables within [0, 1], then compare the sigmoid and a random value 
within [0, 1] to determine the discrete value (Bi Rongshan & Yang, 2003)
t 1 1
sig ( v )
id t 1
1  exp(  v )
id
t 1 (7)
t 1
0 sig ( v
id
)
y
id
  t 1
1 sig ( v
id
)

3
This document is created with trial version of Document2PDF Pilot 2.5.82.

With these three methods, the original MO-MINLP problem is converted to the following
form
min F  F ( X ,Y )    h 2 ( X , Y )   g 2 ( X , Y )  ( X , Y )
0 g( X ,Y )  0
 ( X ,Y )  
1 g( X ,Y )  0 (8)
X  X / X  R n1 , X L  X  X U 

Y  Y / Y  0,1 2
n

Where  is a very big positive number.

So the above function F is the fitness function of PSO. The fitness function F measures
to what extent the particle solution ( X , Y ) satisfy the constrains of MO-MINLP problem.

The PSO algorithm

The procedure of PSO is summarized as follows.
Suppose that the search space is n-dimensional, which is the first n1 continuous variables
are and the other n2 binary variables, then the i-th particle of the swarm can be represented by
a n-dimensional vector, X i  ( xi,1, xi,2 ,..., xi,n , x ,... x )T .
1 i , n i, n
11
Step 1: Initialize a population of particles with random positions and velocities, where the
continuous variable ith particle is xi , d  xd  r1 ( xd  xd ) which
0 L U L
xdL , xdU , are lower and
upper bound at d dimension respectively and (d = 1,2,…,n1), binary variable
xi0,b  0,1,dimension b  n1  1, n1  2,..., n and vi0,d  r2 ,where r1 and r2 are uniform
random value of U(0,1).
Step 2: Evaluate the objective values of all particles, set p of each particle and its
i , best
objective value equal to its current position and objective value, and set G and its
best
objective value equal to the position and objective value of the best initial particle.
Step 3: Update the velocity and position of every particle according to Eqs. (1)-(2) and (7).
If xi , d  xd then xi , d  xd and if xi , d  xd then xit,d1  xdU ,  d  1,2,..., n1
t 1 L t 1 L t 1 U

Step 4: Evaluate the objective F values of all particles according to Eqs. (8).
Step 5: For each particle, compare its current objective value with the objective value of
its p . If current value is better, then update p and its objective value with the
i , best i , best
current position and objective value.
Step 6: Determine the best particle of current whole population with the best objective
value. If the objective value is better than the objective value of G , then update G and
best best
its objective value with the position and objective value of the current best particle.
Step 7: If a stopping criterion is met, then output G and its objective value; otherwise
best
go back to Step (3).

4
This document is created with trial version of Document2PDF Pilot 2.5.82.

4. Heat Exchanger Network synthesis (HENS)

Heat exchanger network synthesis (HENS) has been one of the most well studied issues
within process synthesis during the last three decades. Process synthesis, a part of process
design, has the objective of developing systematically a flow-sheet which describes the
overall process which meets certain specified performance criteria, to transform the raw
materials into the desired products (Floudas (1995)). The overall process system consists of
three main interactive components; Industrial processes, Heat recovery system and Utility
system (Aaltola (2002)).
In the heat recovery system, the process streams exchange heat so as to reduce the hot and
cold utility requirements. The only units in a heat recovery system are the heat exchangers.
A heat exchanger network synthesis (HENS) problem can be described as the one that
synthesize a HEN configuration to reach some assigned targets such as minimum utility
consumption, minimum total number of heat exchangers, minimum area etc., with given
heating/cooling utilities and hot/cold process streams be cooled / heated from nominal inlet
temperatures to specified target temperatures (Chen& Hung(2005)) .
Most of the synthesis methods for heat exchanger networks rely on sequential or step-
wise procedures (Gundersen and Grossmann, 1988). The design problem is decomposed into
three steps in order to progressively determine targets for synthesizing a network. The first
two steps involve the solution of the LP and MILP transshipment model of Papoulias and
Grossmann (1983). For a particular heat recovery approach temperature (HRAT) value, the
LP model determines the minimum utility requirement for the network. With the utility
consumption fixed at the LP solution, the MILP model is solved to determine the minimum
number of matches and their corresponding heat loads. Finally, in the third step, heat loads
and matches are fixed and the area cost is minimized by the solution of an NLP model
(Floudas et al., 1986) to determine the optimal network configuration. The limitation of a
sequential synthesis method is that different costs associated with the design cannot be
optimized simultaneously. In other words, tradeoffs between the different costs, as shown in
Fig. 1, cannot be accounted for accurately. Specifically, as shown in Fig. 1, trade-offs
between utility cost, fixed charges for the number of units, and heat transfer area cost should
be determined simultaneously (Yee and Grossmann(1990)).
Yee & Grossmann (1990) has proposed a superstructure representation to synthesis a
model which accounts for all the costs simultaneously yet requiring very reasonable solution
times. The problem can be formulated as an MINLP which has the desirable feature that all
the constraints are linear. The solution scheme determines the network which exhibits least
annual cost by optimizing simultaneously for utility requirement, the number of units, and the
heat transfer area. The superstructure is a stage-wise representation where within each stage
exchanges of heat can occur between each hot and cold stream.

Fig.1 Trade-off between costs in design

5
This document is created with trial version of Document2PDF Pilot 2.5.82.

4.1. Synthesis of HEN by MO-MINLP

In this work, we extend the work of simultaneously considering minimization of the total
utility consumption, number of matches and the heat transfer area as multiple design
objectives. The HEN synthesis problem has been formulated as the one of multi-objective
mixed-integer nonlinear programming (MO-MINLP). This formulation enables the
investigation of the effect of various cost components on the total annual cost and thus helps
in choosing the best configuration.
4.1.1. Problem statement
In order to address the heat integration problem, it will be assumed that given are a set of
hot process streams HP to be cooled and a set of cold process streams CP to be heated.
Specified are also a set of hot utilities HU and a set of cold utilities CU and their
corresponding temperatures. The solution defines the network by providing the following
(Yee et al. (1990)):
1. Utilities required.
2. Stream matches and the number of units.
3. Heat loads and operating temperatures of each exchanger.
4. Network configuration and flows for all branches.
5. Area of each exchanger.
4.1.2. Superstructure model representation and formulation
The proposed representation for heat integration is a stage-wise superstructure which
allows for different possibilities and sequences for matching streams. At each stage, hot and
cold streams are split to allow the potential existence of a heat exchanger to match any hot-
cold pair of streams. Before a stream enters a new stage, its streams of the preceding stage are
remixed isothermally. For each stream, only an overall heat balance must be performed
within each stage. Also, for simplicity, the hot cold utilities are placed at the outlet of the
superstructure. For details of representation see (Yee et al. (1990)).An example of a
superstructure involving two hot and two cold streams is shown in Fig. 2.
Binary variables are introduced to designate the existence of each potential heat exchanger
in the superstructure, and to model fixed cost charges for the exchangers. Continuous
variables are assigned to temperatures and heat loads. The general model involves overall
heat balances for each stream, stream energy balances at each stage, assignment of known
stage temperatures, calculation of hot and cold utility loads, logical constraints and
calculation of approach temperatures. The feasible space of the problem can be defined by a
set of linear constraints as will be shown later. The nonlinearities, of the model, involving the
calculation of areas using stage temperatures, are isolated in the objective function (Yee et al.
(1990)).
In order to formulate the MINLP model for number of hear exchanger, area and energy
targeting for the proposed superstructure described previously, the following definitions are
necessary(Yee & Grossman (1990)) :

6
This document is created with trial version of Document2PDF Pilot 2.5.82.

Fig. 2. Two-stage superstructure.

Objective functions for HEN

The multi-criteria optimization approach has been adopted for HEN synthesis and the
problem is formulated as a multi-objective mixed-integer nonlinear programming (MO-
MINLP). Minimizing utility, minimizing number of matches and minimizing the exchangers
areas can be simultaneously considered as conflict objectives for synthesis of the network
structure. The objective functions could be stated as follows:
min f   qcu   qhu (9)
1 i j
i  HP j  CP

min f     z   zcu   zhu (10)

2 i, j , k i j
i  HP j  CP s  ST i  HP j  CP

  
min f
3    qi, j , s /(U ij AMTDi, j , s )
iHP jCP sST  

  qcu /(U AMTD ) (11)

iHP  i i, cu i, cu 

  qhu /(U AMTD )

jCP 
 j hu , j hu , j 

where
AMTD i , j ,s  ( dti , j ,s  dt )/2
i , j ,s 1


AMTD i,cu  dtcui  (Ti,out  Tcu,in ) / 2 
AMTD
hu, j

 dthu j  (T
hu ,in
 T j ,out / 2 
1 1 1 1 1 1 1 1 1
  ,   and  
U i, j hi hj U h hj U i ,cu hi hcu
hu , j hu

7
This document is created with trial version of Document2PDF Pilot 2.5.82.

The constraints for HEN

Overall heat balance for each stream
(T T ) FCp    q  qcu , i  HP
i , in i , out i i, j , s i
s  ST j  CP
(12)
(T T ) FCp    q  qhu , j  CP
j , out j , out j i, j , s j
s  ST i  HP

Heat balance at each stage

(t t ) FCp   q , i  HP , s  ST
i, s i, s  1 i i, j , s
j  CP
(13)
(t t ) FCp   q , j  CP , s  ST
j, s j, s  1 j i, j , s
i  HP

Assignment of superstructure inlet temperatures

t T , i  HP
i ,1 i , in
(14)
t T , j  CP
j , NOK  1 j , in

Feasibility of temperatures
t t , i  HP , s  ST
i, s i, s  1
t t , j  CP , s  ST
j, s j, s  1
(15)
T t , i  HP
i , out i , NOK  1
T t , j  CP
j , out j ,1

Hot and cold utility load

(t T ) FCp  qcu , i  HP
i , NOS  1 i , out i i
(16)
(T t ) FCp  qhu , j  CP
j , out j ,1 j j

Logical constraints
q  z 0, i  HP , j  CP , s  ST
i, j , s i, j i, j , s
qcu   zcu  0 , i  HP (17)
i i i
qhu  zhu 0, j  CP
j j j

Where the corresponding upper bound  can be set to the smallest heat content of the
two streams involved in the match.

8
This document is created with trial version of Document2PDF Pilot 2.5.82.

 i  FCp i (Ti ,in  Ti ,out ) , i  HP

 j  FCp j (T j ,out  T j ,in ) , j  CP (18)

 i , j  min( i ,  j ) , i  HP , j  CP

z , zcu , chu  0,1 , i  HP , j  CP , s  ST

i, j , s i j

Calculation of approach temperatures

dt t t   (1  z ), i  HP , j  CP , s  ST
i, j , s i, s j, s i, j i, j , s
dt t t   (1  z ), i  HP , j  CP , s  ST
i, j , s  1 i, s  1 j, s  1 i, j i, j , s
(19)
dtcu  t T   (1  zcu ) , i  HP
i i , NOS  1 cu , out i

dthu T t   (1  zhuj ), j  CP
j hu , out j ,1

i , j  t i ,in  t j ,in , i  HP , j  CP
Where
  max i , j

Also, in order to avoid infinite areas, small positive bounds are specified for the approach
temperature variables dt; that is:
dt  T , i  HP , j  CP , s  ST
i, j , s min
dtcu  T , i  HP (20)
i min
dthu  T , j  CP
j min

5. Application of PSO and GAMS on MO-MINLP

PSO and GAMS methods have been applied to the MO-MINLP formulation of an
example which is taken from (Yee T. F et al, 1990). The problem involves two hot streams
and two cold streams along with hot and cold utility. Also specified is a heat recovery
approach temperature (HRAT) of 10 K. The target is to minimize utilities, number of units and the
total heat transfer area of the HEN and hence minimizing the overall cost /year. The data are
presented in Table 1.

9
This document is created with trial version of Document2PDF Pilot 2.5.82.

Table (1) Example (Yee T. F et al, 1990)

index Stream TIN(K) TOUT(K) FCp h Cost
(kWK-1) (kWm K ) ($kW-1yr -1)
2 -1

1 H1 395 343 4 2.0 -

2 H2 405 288 6 0.2 -
3 C1 293 493 5 2.0 -
4 C2 353 383 10 0.2 -
hu S1 520 520 - 2.0 140.2
cu W1 278 288 - 2.0 7.1
-1 2
Cost of Heat Exchangers and coolers ($ yr ) = 6250 + 83.26* [Area (m )]
Cost of Heaters ($ yr-1) = 6250 + 99.91 *[Area (m2)] (Zamora & Grossman, 1998)
5.1. Parameters for MO-MINLP and PSO algorithm:
The algorithm needs some parameters to be initialized:
c1  1.2, c 2  2.8, w max  0.8, w min  0, T pop  70, N iter  1000 ,   100000
5.2. Results of the example:
Different weighting objective parameters have been tried to investigate the effect of the
various cost components on the total annual cost and the problem has been solved using the
GAMS method. The results are shown in the Appendix. The best weighting objective
parameters were w1  0.01, w 2  0.98, w 3  0.01 , so it is used with the PSO method.
The results of using the PSO and the GAMS methods are depicted in table (2) together with the
published result (Yee et al, 1990). The deduced HEN for the PSO; GAMS and the published results is
shown in figs. (3); (4) and (5) respectively.
As shown in table (2) and Figs.(3,4,5), the HEN designed by PSO has the best results with respect
to overall cost/year where it has minimum number of units as those obtained from GAMS and less
than those in the published results . Moreover the hot and cold utilities are similar to the published
results lower and than the values obtained from GAMS.

Table (2): solution of the example by the different methods

Solution of Number of Sum of cold Sum of hot Area Overall cost
-1
example units utilities utilities m2 ($ yr )
PSO method 6 230 620 257.3525 147628.8
GAMS Program 6 250 640 245.6277 149599.8
Published Paper 8 230 620 245.7026 159179.5

6. Conclusion

Process synthesis problems can be mathematically represented as Multi-objective Mixed-

Integer Nonlinear Programming (MO-MINLP) models, which are often irregular, large and
non-convex and difficult to get the overall optimum by traditional method. In this paper, the
MO-MINLP problem was modified by applying the weighting method and using Penalty
function to eliminate the implicit constraints and introducing a sigmoid ([]) function to deal
with the discrete variables. With these modifications the PSO algorithm, which originally
can only be used in problems with continues variables with max/min limitations, can solve
the MO-MINLP problems with equations and inequalities constraints.

10
This document is created with trial version of Document2PDF Pilot 2.5.82.

The PSO algorithm is simple in that there are very few parameters to adjust. It can achieve
the optimal or near-optimal solutions in a rather short time without enormous iterative
computations in digital implementation.
So far, there is no method for determining the global optimal solution to the general
nonlinear programming problem. Because PSO has very deep intelligent background, it is
suitable for science computation and engineering applications.
We conclude that the PSO method and the developed program can be used to solve large
scale MO- MINLP problem. The above example of 60 constraints (16 equalities & 44
inequalities) and 48 variables (12 binary & 36 continuous) demonstrates the validity of the
PSO method compared to other methods.
It is recommended to use PSO method to design HENs with optimum conditions. Still
there are other issues needed for further researches such as trying other evolutionary
algorithms and their combinations of for solving engineering problems.

11
This document is created with trial version of Document2PDF Pilot 2.5.82.

Fig. (3) HEN design of the example by using PSO method

Fig. (4) HEN design of the example using by GAMS program

Fig. (5) Published HEN design of the example

12
This document is created with trial version of Document2PDF Pilot 2.5.82.

(i) Indices:
i = hot process or utility stream,
j = cold process or utility stream,
s = index for stage 1,..., NOS and temperature location 1,…, NOS + 1
cu = cold utility
hu = hot utility
in = inlet
out = outlet
(ii) Sets:
HP = {i │ i is a hot process stream},
CP = (j │ j is a cold process stream},
HU = hot utility,
CU = cold utility,
ST = {s │ s is a stage in the superstructure, s = 1, . , NOS};
(iii) Parameters:
Ti , in , Ti , out = inlet and outlet temperature of hot stream,
T j , in , T j , out = inlet and outlet temperature of cold stream,
FCp = heat capacity flow rate,
U i , j ,U i ,cu , U hu , j = overall heat transfer coefficients,

NOS =Total number of stages

i, j = an upper bound for heat exchange, can safely be set to the smallest
heat content of the two streams involved in the match
i, j = an upper bound for temperature difference; which are calculated
based on the inlet and outlet temperatures of the superstructure,
T = exchanger minimum approach temperature difference (EMAT).
min
(iv) Variables:
dti,j,s = temperature approach for match (i, j) at temperature location s,
dtcui = temperature approach for the match of hot stream i and cold utility,
dthuj = temperature approach for the match of cold stream j and hot utility,
qi,j,s = heat exchanged between hot process stream i and cold process stream j in stage s,
qcui = heat exchanged between hot stream i and cold utility,
qhuj = heat exchanged between hot utility and cold stream j,
ti,s = temperature of hot stream i at hot end of stage s,
tj,s = temperature of cold stream j at hot end of stage s,
zi,j,s = binary variable to denote existence of match (i,j) in stage s,
zcu i = binary variable to denote that cold utility exchanges heat with hot stream i,
zhuj = binary variable to denote that hot utility exchanges heat with cold stream j,

List of Abbreviation

DM Decision Maker.
HENS Heat Exchanger Network Synthesis
HRAT Heat Recovery Approach Temperature
PSO Particle Swarm Optimization
NLP Nonlinear Programming.
LP Linear Programming
MINLP Mixed Integer Nonlinear Programming Heat exchanger network
MO- MINLP Multi-objective Mixed Integer Nonlinear Programming
GAMS General Algebraic Modeling System.
(one of the first and most widely used algebraic modeling languages)

13
This document is created with trial version of Document2PDF Pilot 2.5.82.

References:
1. Aaltola J. (2002), “Simultaneous synthesis of flexible heat exchanger network”,
App. Therm. Eng., 22(8), 907–918.
2. Aruldoss, T.; Albert Victoire and A. Ebenezer Jeyakumar (2005),
“Deterministically guided PSO dynamic dispatch considering valve-point effect.”,
Electric Power Systems Research 73, 313-322
3. Bi Rongshan and Yang Xia (2003), “ Using Particle Swarm Optimization for
Mixed-Integer Nonlinear Programming in Process Synthesis”, (Institute of
Computer and Chemical Engineering, QingDao University of Science and
Technology, QingDao City, ShanDong Province, China 266042) 10.
4. Chen C. L. and p.s. Hung (2005), “Multicriteria synthesis of flexible heat-
exchanger networks with uncertain source-stream temperatures.” Chem. Eng.
Processing 44, 89-100.
5. Eberhart, R. C., Simpson P. and Dobbins R. (1996), “Computational Intelligence
PC Tools”. Academic Press.
6. Emad,E; Hegazy, T. and Grierson, D, (2005), “Comparison among five
evolutionary-based optimization algorithms”; Advanced Engineering Informatics
19, 43–53.
7. Floudas C.A. (1995), “Nonlinear and Mixed-Integer Optimization: Fundamentals
and Applications”, Oxford Univ. Press, New York.
8. Floudas, C. A.; Ciric, A. R. and Grossmann, 1. E. (1986), “Automatic synthesis of
optimum heat exchanger network configurations”. AIChE JI 32, 276-290.
9. Gundersen T. and I. E. Grossmann (1988), “Improved optimization strategies for
automated heat exchanger network synthesis through physical insights.” Presented
at the Annual AIChE Meeting, Washington, D.C.
10. Kannan S.; S. Mary Raja Slochanal; Subbaraj, P. and Narayana Prasad Padhy
(2004), “Application of particle swarm optimization technique and its variants to
generation expansion planning problem”, Electric Power Systems Research 70,
203–210
11. Kassem, M. (1995), “Interactive stability of multiobjective nonlinear programming
problems with fuzzy parameters in the constraints”. Fuzzy Sets and Systems
73,235-243
12. Kennedy, J. and Eberhart, R. C. (1995). “Particle Swarm Optimization”, Proc.
IEEE International Conference on Neural Networks (Perth, Australia), IEEE
Service Center, Piscataway, NJ, IV: 1942-1948.
13. Lovbjerg M. (2002), “Improving particle swarm optimization by hybridization of
stochastic search heuristics and self-organized criticality”. Masters Thesis, Aarhus
Universitet, Denmark.
14. Miettinen, K. (2002), “Nonlinear Multi-objective Optimization.” Kluwer
Academic Publishers.
15. Omar. S. Soliman (2005), “An Interactive Particle Swarm Optimization for General
Fuzzy Non-Linear Goal Programming”, Third International Conference on
Informatics and Systems, Mar 19-22, Cairo University, Faculty of Computers and
Information, Giza, Egypt
16. Papoulias S. A. and Grossmann, I. E. (1983), “A structural optimization approach
in process synthesis-II. Heat recovery networks” , Comp. Chem. Engng. 7, 707-721
17. Parsopoulos, K. E. and Vrahatis, M. N. (2005), “Recent approaches to global
optimization problems through Particle Swarm Optimization”, Natural Computing
1, 235-306

14
This document is created with trial version of Document2PDF Pilot 2.5.82.

18. Yee T. F. and Grossmann, 1. E. (1990), “Simultaneous optimization models for heat
integration-II. Heat Exchangers Network Synthesis.” , Computers Chem. Engng. 14,
1165-1184.
19. Yee T. F.; Grossmann, 1. E. and Kravanja, Z. (1990), “Simultaneous optimization
models for heat integration-I. Area and energy targeting and modeling of multi-
stream exchangers.” Computers Chem. Engng 14, 1151-1164.
20. Ying, J. T.; Baodong Xu and Dingwei W. (2005), “An Application of Swarm An
Application of Swarm Optimization to Nonlinear Programming.”, Computers and
Mathematics with Applications 49, 1655-1668
21. Zamora J. M. and Grossman I. E. (1998), “A global MINLP optimization algorithm
for the synthesis of heat exchanger networks with no stream splits”, Comp. chem.
Engng. Vol. 22, No. 3, pp. 367-384.

Appendix

Weighting table
No. w1 w2 w3 f1 f2 f3 Cost $
(1) 0.272 0.004 0.724 266.2 7 850 154914.4
(2) 0.066 0.929 0.005 82.88 4 1794 190288.4
(3) .01 .98 .01 245.6 6 890 149599.8
(4) 1/3 1/3 1/3 255.2 8 850 159966.2
(5) 0.05 0.9 0.05 266.2 7 850 154615.4

The weight could be determined by:

(1) using the equation (Kassem, M. A., 1995)
k
w i  ( f i max  f i min ) /  ( f jmax  f jmin ) i  1,.., k
j 1

(2) using pay off table (Kaisa, 2002)

where
1 zinad  zi e
ei   nad
, wi  k i , z is ideal objective vector
zi zi
 ei i 1
nad
and z is nadir objective vector

(3) Decision Maker suggestion

15