Vous êtes sur la page 1sur 6

4th IEEE Conference on Automation Science and Engineering

Key Bridge Marriott, Washington DC, USA


August 23-26, 2008

A Hybrid ACO/PSO Heuristic to solve Single Row


Layout Problem
Y.T. Teo S.G. Ponnambalam
School of Engineering School of Engineering
Monash University Sunway Campus Monash University Sunway Campus
Petaling Jaya, Malaysia Petaling Jaya, Malaysia
yuangteck@gmail.com sgponnambalam@eng.monash.edu.my

AbstractIn this paper, a hybrid ACO/ PSO heuristic to The commonly used machine layouts in FMS are single
solve single row layout problem is proposed. As opposed to row layout, double row layout, loop layout, cluster layout
previous researches, the clearances and size of machines are and circular layout. Hassan [2] pointed out that approaches
treated as variables by using non-linear 0-1 mathematical suggested in past literature were mostly dealt with block
model adopted from literature. ACO is used as constructive layout and are not suitable to be implemented in new types
heuristic with a new pheromone update developed to achieve of layout in FMS, such as single row layout. Therefore, it is
better performance on proposed algorithm. PSO, on the other essential to develop new models and solution methodologies
hand, is used as an improvement heuristic to guide the ants to to suit the modern manufacturing systems.
reach the best solution. To further improve the quality of the
solutions, 2-Opt local search is added. The proposed hybrid This paper emphasizes on developing a new solution
ACO/PSO heuristic is evaluated with various sets of problems methodology to solve the single row layout problem. A
available in the literature, and the results are compared hybrid ant colony optimization (ACO)/particle swarm
with those reported in the literature. optimization (PSO) is proposed in this paper to determine
the best arrangement of machines in a single row layout
KeywordsACO, PSO, Single Row Layout, Hybrid. suitable for FMS. The performance of the proposed
algorithm is evaluated with a set of problems available in the
I. INTRODUCTION literature and the proposed hybrid ACO/PSO performs better
Issues on facilities layout in manufacturing industries over the models reported in the literature. The mathematical
have been a common problem in the past two decades and model proposed by Solimanpur et al. [3] is adopted in this
are well known among researchers. Optimizing facilities project which is suitable for FMS environment.
layout in a manufacturing plant is basically to arrange the
machines in the workshop area with the best or optimized II. LITERATURE REVIEW
sequence to increase the performance of the factory. Single row layout is well known as a NP-complete
Researchers have found out that it is possible to reduce the problem [4]. The problem is not as simple as permutations of
manufacturing costs to up to 30% by a good arrangement of a fixed number of machines, but certain criteria also needed
machines because it was estimated that 20-50% of the costs to be satisfied in the operation sequence of all the products.
are due to the handling of work pieces [1]. Therefore, Furthermore, if one would like to generate the permutation
looking into such high percentage of return, many researches for a large problem with 30 machines, it would take
had done to provide the best solution to the problem. generations to find out all the possible combinations before
Flexible manufacturing system(FMS) is one of the solving the problem itself. Accurate mathematical solutions
automated manufacturing system where production system do not exist for such problem [5], therefore many heuristics
are formed by a set of automated machines and flexible are proposed to solve single row layout problem. Most of the
material-handling system such as robots, automated guided past literature formulated the single row layout problem as
vehicle and others. Due to the incorporation of advanced Quadratic Assignment Problem (QAP). QAP is one of the
material-handling system, facility layout in FMS has combinatorial optimization problems branching from
changed significantly. The enthusiasm of arranging these optimization or operation research in mathematics,
facilities in the right order has increased with the particularly facilities location problem [6].
implementation of FMS.
Kumar et. al.[7] suggested a constructive (greedy)
heuristic which looked into the strong relationships between
Y. T. Teo was with School of Engineering, Monash University mutually distinct pairs of facilities, minimizing the material
Sunway Campus, Malaysia. This work was done when he was with handling cost. They did not take into account the dimension
Monash University Sunway Campus. He is presently with Western Digital of the facilities in the layout construction. Due to QAP
Malaysia (E-mail: yuangteck@gmail.com). formulation, the size of the locations are predefined as equi-
S. G. Ponnambalam is with School of Engineering, Monash University
Sunway Campus, Malaysia. (Phone: +60-3-55146207; Email:
distant which lie on a straight line. Braglia [8] used the
sgponnambalam@eng.monash.edu.my). backtracking minimization as the objective of the problem.
He adopted the mathematical formulation found in [9] and

978-1-4244-2023-0/08/$25.00 2008 IEEE. 597


implemented the Simulated Annealing (SA)-based heuristic c) There are no duplication of machines allowed in a
proposed by Kouvelis and Chiang [10] as the basis of the single complete solution.
proposed solution and added Genetic Algorithm (GA) to d) The precedence constraint between machines also
improve the quality of the solution. will not be considered.
e) The solutions created are regardless of the input or
Braglia [11] considering traveling time instead of
output of the parts from the system, therefore
traveling distances because material handling devices minimizing the complexity of the problem.
(MHD) generally are not consistent in velocity and proposed
a constructive heuristic algorithm based on NEH heuristic IV. NOTATIONS:
which was developed for the flow shop scheduling problem
by Nawaz et. al.[12]. The notations used are explained in TABLE I.

Djellab and Gourgand [13] proposed an iterative TABLE I. NOTATIONS


construction procedure which consists of a number of stages Notations Description
to minimize the total time required by MHD to transport the M Total number of machines or locations
part types between machines. fij Number of trips from machinei to machinej
Ficko et. al.[5] proposed a genetic algorithm to solve the f(h)(l) Number of trips from machinei at locationh to machinej
at locationl
single and multiple rows layout problems with variable
Li Length of the machinei
distance between machines to minimize the distance
L(h) Length of the machinei at locationh
between them. They tested their model with an example of
sij Clearance required between machinei and machinej
FMS with 14 devices, but they did not include any
s(h)(l) Clearance required between machinei at locationh and
description on the problem itself which restricted us from machinej at locationl
comparing. dhl Distance between the centroids of locationh and
locationl.
An ant algorithm was proposed by Solimanpur et. al.[4]
il Heuristic information of placing machinei to locationl
for he single row layout problem. They presented a non-
il Pheromone trail of placing machinei to locationl
linear 0-1 mathematical model for the NP-hard problem
pilk Probability information of antk to place machinei at
which is more suitable for single row layout problem in locationl
FMS environment. Ptk Position of particlek in PSO at time t
Hassan [2] reviewed the formulations and solution Vtk Velocity of particlek in PSO at time t
methodologies for the machine layout problems of all the N Number of ants or particles
literatures from year 1963 to 1992. He presented the design A_stop ACO termination condition
issue of machine layout problems and also the optimization P_stop PSO termination condition
approach for developing the machine layout. The potential , Weightage for local pheromone update
Weightage for global pheromone update
areas for research in the machine layout problem are also
Importance of pheromone change in selection
presented. Singh and Sharma [14] presented the state of the probability equation
art review of papers on facility layout problems. They
C0,C1, C2 Parameters used in PSO velocity update equation.
identified the trend towards using meta-heuristics to solve
facility layout problem. Their survey includes the research
reported using meta-heuristics and other approaches V. MATHEMATICAL FORMULATION
published from past two decades. They concluded that future The non-linear 0-1 mathematical formulation needed in
researches could be developing meta-heuristic models for this paper is adopted from the Solimanpur et. al.[3] as
the facility layout problem. detailed below.
It is identified that exploiting the features of the swarm M 1 M
l 1 l 1 L L (1)
intelligent based approach is not explored in solving single Min ( f (h )(l ) + f (l )(h ) ) d (h )(l ) d (h )(l ) = Lt + s(t )(t +1) + (h ) + (l )
h=1 l =h+1 t =h+1 t =h 2 2
row layout problem. Hence, there is a scope of developing
hybrid swarm based intelligent model for single row layout s.t.
problem. where is denoted as the particular permutation and is
III. ASSUMPTIONS AND CONSTRAINTS the set of all permutation. (h) represents the machine at
location h in permutation .
The assumptions presented in the mathematical model
are as follows: VI. PROPOSED HYBRID ACO/PSO HEURISTICS
a) The machines are assumed to be in rectangle shape Many heuristic, meta-heuristic or even exact solution
with different sizes. Distances between machines are methods had been proposed to solve the single row layout
measured with respect to the centroids of each problem in the past two decades. For a conventional
machine. heuristic search, the algorithm is simple, which an initial
b) The clearance required between machines is variable state is defined in a solution space and search is performed
where different pair of machines have different through the neighborhood to find the optimum point.
clearance requirement. However, the draw back of this conventional heuristic search

598
is that it is easily trapped in local optima, or the solution may generate the solutions closer to the best solution
be revisited in every cycle. Therefore, meta-heuristics comes found.
into action where it allows more diverse search in the 13. Repeat Step 2 to Step 12 until the A_stop is met.
solution space to prevent from being trapped into local
optima. B. Ant Colony Optimization:
ACO is an improved version of the Ant System which
In this paper, two meta-heuristic algorithms are used to was developed by Colorni, Dorigo and Maniezzo in
solve the single row layout problem, which are ant colony 1991[15,16]. This algorithm simulates the behavior of the
optimization (ACO) and particle swarm optimization (PSO). ants in the real world.
The two optimization tools are hybridized to form a new
algorithm to solve the problem. The proposed hybrid Ants travel in accordance to the pheromone intensities of
optimization tool is split into two stages, where ACO is used the trails. Each ant will deposit some substances call
in the first stage followed by PSO used in the second stage. pheromone along the path they have passed through. This
Typically the assumptions made are, the locations are known causes a positive feedback loop where the pheromone trails
in advance and the distance plus clearance are all predefined increases whenever more ants passed through. This allow
numerical value or some treated them as equi-distance. No other ants to follow their route and eventually the better
duplication of machines are allowed in a single solution, route will have more pheromone deposited and all the ants
therefore a machine that has already been assigned to a will favor that particular route to reach their food source
particular location cannot be assigned to other locations. from their nest or vice versa.
In addition to ACO and PSO, a 2-opt local search is 1) Construction of complete feasible solution:
applied to further improve the quality of the solutions. A 2- ACO is used as the initial feasible solution constructor in
opt local search which is also known as pairwise interchange the proposed ACO/PSO heuristic.
heuristic is basically swapping two machines at a time. After
Initially, each ant is placed randomly on different
each swap, the solution is evaluated again and compares the
machines at the starting point. Then the ants will construct
new objective function value (OFV) with the current best
the solution based on the probability of placing the next
OFV. If a better solution is found, the current best solution
machine onto the next location. For example, there are 5
will be replaced.
machines in a single row layout problem; antk is placed at
A. Stepwise Procedures of Proposed Hybrid ACO/PSO machine1 at location1. Then the ant will select the next
Heuristics machine from the list (2, 3, 4, 5) and place the machine to
The proposed solution can be described in 2 stages with next location, location2. This situation is denoted by place(i ,
13 steps as shown below: l), where the machinei will be placed at locationl. For this
case, i takes a value from 2 to 5 and l is 2. The machine is
1. Initialization: N, A_stop, P_stop, , , , C0, C1, C2, placed in the location based on the probability information
and initial pheromone il(0) calculated. Higher the probability the higher chance of
Stage 1 placing that machine to a location under considerations.
2. An ant is randomly placed on the set of machines.
3. The ant will select the next feasible machine to place Three lists will be attached to each ant where first list is
at the next location based on the probability the machines that the ant need to assign (for feasible solution
information shown in equation 8. Repeat Step 3 until construction), second list is the locations that the ant is going
a complete solution is constructed. to place the machine, and lastly the third list is a list that
4. Update the pheromone trail using the local shows the place selected.
pheromone update shown in equation 5. Continue with the example mentioned above, the ant is
5. Repeat step 2 to 4 for N ants to generate a swarm of placed initially at machine1, therefore List1 and List2 will
particles. have [2,3,4,5]. List3 will have [place(1,1)] at the moment.
Stage 2 Lets say if the ant choose machine3 to place at location2 now,
6. Evaluate a particle from the swarm, the complete then List1 will have [2,4,5], List2 will be [3,4,5] and List3 will
solution generated by the ant. be [place(1,1), place(3,2)]. By implementing the lists as
7. Update the velocity attached to that particle based on mentioned, only feasible solutions are constructed.
the equation 10.
8. Update the position of the particle using the updated 2) Heuristic Information:
velocity in Step 7. Each ant in ACO can include some heuristic information
9. Apply 2-opt local search to that particle to improve while assigning the machine to the locations specified. In our
the quality of the solution. convention, 12 is denoted as the heuristic information
10. Repeat Step 6 to 9 for N numbers of times. pertaining to place(1,2), or in other words the desirability of
11. Repeat Step 6 to 10 until P_stop is met, otherwise the ant of placing machine1 to location2. In this paper, the
proceed to Step 12. heuristic information is updated accordingly whenever a new
12. Update the best result obtained so far and the machine is placed in the next location successfully. The
pheromone trail using global pheromone update heuristic information is measured based on the partial
shown in equation 6 to encourage the ants to contribution of the place to the objective function. This
means the smaller the distances traveled by MHD between

599
the assigned machines, the better the desirability of the other words, all the ants will execute local pheromone
place, because is a function of inverse distance. The update.
partial contribution of the move was adopted from 5) Global pheromone update
Solimanpur et. al.[3]. In contrast to local pheromone update, only one ant is
Partial contribution of place(i,l) to the fitness function: allowed to execute global pheromone trail update, which is
the best ant that constructed the best solution among the
l 1

( f ( ) + f i (h ) ) d hl
constructed solutions
h i (2)
h 1 il il + ilBS (6)
where where
l 1 l 1 L (h ) Li
d (h )(l ) = L ( ) + s ( ) (
u = h +1
u
u =h
u u +1) +
2
+
2
(3) z
ilBS = * 1 il (7)
(l) = i z
Therefore the heuristic information pertained to place(i,l) z is denoted as the average fitness of all the solutions
is formulated as below: generated, or in our case the average fitness of all the
particles in the swarm and z* is the best fitness found so far.
1 The ratio is then minus off by 1 to obtain only the decimal
il = (4)
l 1 numbers. The change of pheromone in global pheromone
( f ( )
h 1
h i + f i (h ) ) d hl update is formulated such that it is a function of the original
pheromone trail. This can eliminate the problem on either
The number 1 added with the partial fitness in the first term dominating the second term in equation 6 or
denominator is to avoid division by 0. vice versa. By using the pheromone update in equation 7, it
encourages the best ant to deposit more pheromone on the
3) Pheromone Trail Update: trails if the average result is further away from the best found
There are two kinds of pheromone trail updates: result, minimizing the gap between the best fitness. For
a) Local pheromone trail update example, if z is much larger than z*, this will increase the
pheromone deposition due to the higher of il . However,
BS
b) Global pheromone trail update
if the best found fitness is somehow nearer to the average
4) Local Pheromone Update
fitness of all solutions, il will have a smaller value,
BS
Local pheromone trail update will be applied after an ant
has constructed a complete solution. which in turns less pheromone will be deposited to the
pheromone trails. The advantage of global updating rule is
z that it reduces the computational effort of pheromone update.
il (1 ) il + il (5)
z 6) Selection Probability
where il is the pheromone trail pertained to place(i,l) The probability of assigning which machine to a location
depends on two parameters, the heuristic information il,
Local pheromone trail update is to apply pheromone
evaporation to the pheromone trail (First term of equation 5), and the trail level il. The proposed ACO obtains the
however the evaporation rate is reduced based on the fitness probability using the equation 8 which is proposed by
of the current solution compare with the best result found so Maniezzo [17,18].
far. z* denoted the fitness of the best solution has found so
far, and z is denoted as the fitness of current solution il + (1 ) il (8)
pilk = ; i List1
generated. The ratio is then multiplied with the pheromone iList ( il + (1 ) il )
1
trail so as to replace the pheromone trails that have been
evaporated in the first term. This is to guide the ants to move where List1 contains the set of feasible neighborhood.
towards the global optimum solution. The nearer the fitness
An advantage of this equation over the original equation
of the current solution generated is to the best found so far,
from Ant System (AS) [17] is that it is a much simpler
the lower the pheromone trail is being evaporated. Besides,
operation by using summations instead of multiplications for
the evaporation enables greater exploration in the solution
combining the heuristic information and pheromone trail.
space by the ants without compromising the possibility of
Besides, only one parameter, is used in the equation.
losing track to the global optimum point. in the equation is
governs the importance of pheromone trail over the heuristic
a parameter which determines how important is the
information in equation 8.
pheromone change in local pheromone update as well as
controlling the rate of evaporation. Higher will cause the C. Particle Swarm Optimization
pheromone trail to evaporate at higher rate and emphasize In second stage, PSO is used to improve the solutions
more on the second term. Local pheromone update is created by the ants in first stage. PSO was originally
executed every time a complete solution is constructed, or in designed by Eberhart and Kennedy [19,20,21] which

600
simulates the behavior of living entities that work in a collected from different sources. Problem 2 and 6 are from
swarm, such as ants. The basic elements in PSO are position Love and Wong [24], problem 3, 4 and 5 are from Simmons
and velocity. In this context, each solution created by the [25] and lastly problem 7 and 8 are from Nugent at. al.[22].
ants is represented as one particle. Each particle has a Dimensions of all the machines are from Heragu and Kusiak
position or in other words their fitness in the solution space. [26]. Problem 1 in second set of problem is not considered
Each is then attached with a velocity which the particle will for comparison due to insufficient information. The
move towards the direction of the velocity vector pointing clearances between pair of machines in problem set 1 are
with the distance it specified. The particles will continue reported in literature as 0.01, therefore for comparable
moving and updating the velocity until they reach a common results, the clearance, s in equation 1 is also set to 0.01. For
goal or the optimum solution as all the particles agreed. problem set 2, the clearances are set to 0 as reported by
Kumar et. al.[7].
Each complete solution constructed in ACO is used as a
particle in PSO. For example, [1,2,4,5,3] representing one For the third set of problem, the problems are identical to
feasible solution refers to a particle. Each particle in PSO is the problem set 1, but clearances between machines are set
evaluated based on the objective function and the best to the distances reported in Nugent at. al.[22]. Due to the
solution is determined based on the position discovered so reason that no past papers have ever considered this
far in the iterations, which is called personal best (Pbest). assumption before, the results are only comparable with the
Then all the particles are compared to identify which particle results reported in Solimanpur et. al.[3]. The proposed
has the best solution, or the so called global best (Gbest). heuristic is coded in Matlab program and run on a Pentium 4
Therefore in each iteration, the velocity of the particle is (Prescott) of 3.36GHz with 1MB L2 cache.
recalculated according to the Pbest and Gbest and then
updates the current position of the particle to a new position TABLE II shows the results for the first set of problem in
according to the calculated velocity. The new position is comparison with the results reported in literatures. The time
calculated using equation 9. recorded in the table is the CPU time in seconds. However,
the CPU time recorded for each problem in TABLE II and
Pt k+1 = Pt k + vt (9) TABLE III are not suitable for comparison since the earlier
researches need different computing systems.
The velocity is represented as the transposition of the
machines between locations. The length of velocity is For problem set 1, the results for the proposed method as
determined by the numbers of transposition presented. For well as past heuristics are tabulated in TABLE II. All the
example, [(1,2), (4,2)] is denoted as a velocity of length 2 other past heuristics in TABLE II were obtained on a Sun
and the velocity consists of transposition of location1 with 3/260 Computer except method proposed by Solimanpur et.
location2 and location4 with location2. al.[3] which was obtained on Pentium 3 550MHz PC. The
OFVs in bold show the optimum results for each of the
The velocity update in PSO is shown in equation 10.
problem. The results for the proposed hybrid ACO/PSO are
( k
vt = C0 vt 1 + C1 rand1 PPBest ) (
Pt k + C2 rand2 PGBest
k
Pt k ) tabulated in the last column. The proposed ACO/PSO is able
to achieve best results obtained so far.
(10)
TABLE II. THE COMPARATIVE RESULTS FOR PROBLEM SET 1
In equation 10, vt is governed by three vectors, which are
previous velocity, vt-1, velocity of current position to
personal best position and lastly the velocity of current
position to global best position. ( PPbest
k
Pt k ) is denoted as
the transposition needed to change from the current position,
Pt k to the personal best, PPbest
k
. ( PPbest
k
Pt k ) is denoted as
the transposition needed to change from the current position,
Pt k to the global best, PGbest
k
. The results for the second set of problem are tabulated in
TABLE III. The proposed hybrid ACO/PSO is able to
Each of the vectors consists of transposition and the achieve the best solutions so far reported, except for problem
length is controlled by the parameters C0, C1, C2 as well as 8. The first problem in problem set 2 is not experimented
random number, rand. The random number generated should due to the incomplete information of the data. The data were
lie in the range of 0rand1 for both second and third vector published in a French paper. The reported result for problem
in the velocity update. All the vectors will be summed up to number 8 by Solimanpur ed. al.[3] is extremely low
form a new velocity vector for position update. compared to the other methods. We suspect that this could
be a typo error. We are in correspondence with the authors
VII. PERFORMANCE ANALYSIS for clarification. The proposed method by Heragu and
Three sets of problems are commonly used by the Kusiak [26] was run on an AMDAHL 5870 Computer.
researchers to evaluate the performance of their proposed Heragu and Alfa [27] obtained the CPU time on an VAX
method. Problem Set 1 by Nugent at. al.[22] consists of eight 6420 mainframe computer. Kumar[7] tested their proposed
problems and the size of the machines are found in Heragu heuristic on a Sun 3/260 computer and Solimanpur et. al.[3]
and Kusiak [23]. Problem Set 2 consists of eight problems

601
evaluated their proposed method on a Pentium III 550MHz [2] Hassan M.M.D. Machine layout problem in modern manufacturing
facilities. International Journal of Production Research. 1994;
PC. 32:2559-2584.
[3] Solimanpur M., Vrat P., Shankar R. An ant algorithm for the single
TABLE III. THE COMPARATIVE RESULTS FOR PROBLEM SET 2 row layout problem in flexible manufacturing systems. Computer &
Operations Research. 2005; 32:583-598
[4] Beghin-Picavet. M., Hansen, P. Deux problemes daffectation non
lineaires. RAIRO Recherche Operationelle. 1982; 16:263-276.
[5] Ficko M., Brezocnik M., Balic J. Designing the layout of single- and
multiple-rows flexible manufacturing system by genetic algorithms.
Journal of Materials Processing Technology. 2004; 150-158.
[6] Quadratic assignment problem Wikipedia, the free encyclopedia.
[Online] 2007 [cited 2007 October 1]. Available from:
http://en.wikipedia.org/wiki/Quadratic_assignment_problem
[7] Kumar K.R., Hadjinicola G.C., Ting-Li L., A heuristic procedure for
the single-row facility layout problem. European Journal of
TABLE IV shows the OFVs and the computational time Operational Research, 1995; 87:65-73.
in average of five cycles for each of the problem in problem [8] Braglia M. Optimisation of a Simulated-Annealing-based Heuristic
for Single Row Machine Layout Problem by Genetic Algorithm.
set 3. The results of the proposed hybrid ACO/PSO in last International Transaction in Operational Research. 1996; 3:37-49
column are compared with 2-opt algorithm and algorithm [9] Sarker B.R., Wilhelm W.E. & Hogg G.L. Backtracking and its
proposed by Solimanpur et al.[3]. As shown, the proposed amoebic properties in one-dimensional machine location problems.
Journal of Operational Research Society. 1994; 45:1024-1039.
method only hit the optimum results in problem 1, 2, 6 and 8. [10] Kouvelis P., Chiang W.C. A simulated annealing procedure for
However, we see inferiority in solving the other problems. single row layout problems in flexible manufacturing systems.
The proposed hybrid ACO/PSO performs better than a 2-opt International Journal of Production Research. 1992; 30:717-732.
[11] Braglia M. Heuristics for single-row layout problems in flexible
local search in all the problems except problem 4. The manufacturing systems. Production Planning & Control. 1997;
inferiorities are mainly due to the inconsistency of results 8:558-567.
[12] Nawaz N., Enscore Jr. E. E., Ham I. A heuristic algorithm for m-
obtained in different iteration. Heuristic methods involved machine, n-job flow-shop scheduling problem. Omega, 1983; 11:91-
random aspect; results obtained from same sets of 95.
configuration might not be the same even with the same [13] Djellab H., Gourgand M. A new heuristic procedure for the single-
row facility layout problem. International Journal of Computer
experimental machines and same environment. Therefore, Integrated Manufacturing. 2001; 14:270-280
the results reported in TABLE IV have yet to be used to [14] Singh S. P., Sharma R. R. K. A review of different approaches to
conclude that the method proposed by Solimanpur et al. [3] facility layout problems. International Journal, Advance
Manufacturing Technology 2006; 30: 425-433
performs better than proposed hybrid ACO/PSO. [15] Colorni, A., Dorigo, M. and Maniezzo, V. Positive Feedback as a
Search Strategy. Technical Report. 1996; No. 91-016, Politecnico di
TABLE IV. THE COMPARATIVE RESULTS FOR PROBLEM SET 3 Milano, Italy
[16] Colorni, A., Dorigo, M. and Maniezzo, V. The Ant System: An
Autocatatlytic Process. Technical Report. 1996; No. 91-016,
Politecnico di Milano, Italy
[17] Dorigo, M. and Sttzle, T. Ant Colony Optimization. London,
England: The MIT Press. 2004. Pg. 76-80.
[18] Maniezzo, V. Exact and approximate nondeterministic tree-search
procedures for the quadratic assignment problem. INFORMS
Journal on Computing. 1999; 11:358-369
[19] Eberhart, R. C., and Kennedy, J. A new optimizer using particle
swarm theory. Proc. Sixth International Symposium on Micro
Machine and Human Science (Nagoya, Japan) 1995; No. 39-43,
IEEE Service Center, Piscataway, NJ
[20] Kennedy, J., and Eberhart, R. C. Particle swarm optimization. Proc.
IEEE International Conference on Neural Networks (Perth, Australia)
VIII. CONCLUSION: 1995; No. IV: 1942-1948, IEEE Service Center, Piscataway, NJ
[21] Kennedy, J. The particle swarm: social adaptation of knowledge
A hybrid meta-heuristic by combining ACO and PSO is Proc. IEEE International Conference on Evolutionary Computation
(Indianapolis, Indiana) 1997; No. 303-308, IEEE Service Center,
proposed to solve the single row layout problem. A 2-opt Piscataway, NJ
local search is also implemented to improve the quality of [22] Nugent C.E., Vollman T.E., Ruml, J., An experimental comparison
the solution. The mathematical model of the single row of techniques for the assignment of facilities to locations. Operations
Research, 1968; 16: 150-173
layout problem adopted in this research considered the [23] Heragu S.S., Kusiak A. Machine layout problem in flexible
dimensions and clearance between machines as a variable manufacturing systems. Operation Research. 1988; 36:258-268.
[24] Love R.F., Wong J.Y., On solving a one-dimensional space
which most of earlier researchers ignored. This leads to a allocation problem with integer programming. INFOR, 1976; 14(2):
better representation of a realistic problem. Different sets of 139-143
problems are used to evaluate the proposed ACO/PSO [25] Simmons D.M., One-dimensional space allocation: an ordering
algorithm. Operations Research, 1969; 17(5): 812-826
heuristic with other algorithms reported in the literature. It is [26] Heragu S.S., Kusiak A., Efficient models for the facility layout
found that the proposed ACO/PSO performs better over the problem. European Journal of Operational Research. 1991; 53(1): 1-
heuristics reported in the literature. It is also noted that the 13
[27] Heragu S.S. and Alfa A.S. Experimental analysis of simulated
proposed heuristics could provide best solution for larger annealing based algorithms for the layout problem. European
problems than smaller one. Journal of Operational Research. 1992; 57(2): 190-202
[28] Hall KM. An r-dimensional quadratic placement algorithm.
Management Science 1970; 17(3):219-29
REFERENCES [29] Neghabat F. An efficient equipment layout algorithm. Operations
[1] Tompkins J.A., White J.A., Bozer Y.A., Frazelle E.H., Tanchoco Research 1974;22:622-8.
J.M., Trevino J., Facilities Planning. New York: Wiley; 1996.

602

Vous aimerez peut-être aussi