Professional Documents
Culture Documents
Single Row Facility Layout Problem Journal Review
Single Row Facility Layout Problem Journal Review
Intelligent
Manufacturing
System
Midterm Report
10274014
10274015
10274054
10274059
4-30-2014
1
1.introduction ................................................................................................................. 3
1.1. Single Row Facility Layout Problem (SRFLP) Definition ........................ 3
1.2. Estimation of Distribution Algorithm (EDAs) .......................................... 3
1.3. Study Objectives and Problem Definition ................................................. 3
2. mathematical formulation ...................................................................................... 5
2.1. Single Row Facility Layout Problem (SRFLP) ......................................... 5
2.1.1. Notations ........................................................................................ 5
2.1.2. Assumptions ................................................................................... 5
2.2. Enhanced Single Row Facility Layout Problem ........................................ 6
2.2.1. Notations ........................................................................................ 6
2.2.2. Assumptions ................................................................................... 7
3. Problem solving methods ....................................................................................... 8
3.1. Assumptions & Notations .......................................................................... 9
3.2. extended Chromosomes Genetic Algorithm (eACGA) ........................... 10
3.3. EDAhybrid ............................................................................................... 12
3.4. EDA Part .................................................................................................. 12
3.5. PSO Part ................................................................................................... 16
3.6. Tabu Search Part ...................................................................................... 16
4. results analysis ..................................................................................................... 19
4.1. Parameter setting ...................................................................................... 19
4.2. EDAHybrid for SRFLP............................................................................ 20
4.3. EDAHybrid for enhanced SRFLP ........................................................... 21
5. conclusion and improvement ............................................................................... 22
TABLE OF CONTENTS
2
5.1. Improvement ............................................................................................ 22
6. References ............................................................................................................ 24
3
1. INTRODUCTION
1.1. Single Row Facility Layout Problem (SRFLP) Definition
SRFLP is a method where multiple products with different production volumes and
different process routings need to be manufactured. The objective of SRFLP is to set
up facilities so that sum of distances between all facility pairs can be minimized.
1.2. Estimation of Distribution Algorithm (EDAs)
EDAs are stochastic optimization techniques that explore that space of potential
solutions by exploiting inter variable dependency and sampling probabilistic models of
promising candidate solutions problem definition (Hauschild & Pelikan, 2011). EDA
are predicted to be potentially effective to solve SRFLP (Ou-Yang & Utamima, 2013).
As more and more iteration is to be run, EDAs will suffer diversity loss that grows
rapidly; however, EDAs improve the solution quality in the first few runs efficiently.
1.3. Study Objectives and Problem Definition
The proposed study by (Ou-Yang & Utamima, 2013) creates a new Hybrid Estimation
of Distribution Algorithm (EDA Hybrid), an algorithm which consist of hybridization
of Estimation of Distribution Algorithm (EDAs), Particle Swarm Optimization (PSO),
and Tabu Search Algorithm to surmount the basic and enhanced SRFLP. PSO is used
as metaheuristic algorithm for maintaining the diversity of EDA. Tabu Search explores
the global best value achieved in every iteration. Other hybridization algorithm,
extended Artificial Chromosomes Genetic Algorithm (eACGA) is built as a benchmark.
4
Objectives of this research are as follows (Ou-Yang & Utamima, 2013):
1. Develop EDA hybrid, a new meta-heuristic algorithm which is the hybridization
of Estimation of Distribution Algorithm Particle Swarm Optimization and Tabu Search
Algorithm.
2. To design an enhanced Single Row Facility Layout Problem that considers more
constraints which include not only flow, length, and clearance space, but also the
installation cost and safety reason.
3. To apply EDAhybrid algorithm to solve the basic and enhanced SRFLP.
5
2. MATHEMATICAL FORMULATION
2.1. Single Row Facility Layout Problem (SRFLP)
The research proposed (Ou-Yang & Utamima, 2013) considers SRFLP with different
sizes of facilities. The objective function is to minimize which stands for sum of the
distances between all facility pairs. ABSMODEL, proposed by Heragu and Kusiak
(1991), is well known model for solving SRFLP. ABSMODEL is illustrated in Eq. (1).
=
=+1
1
=1
. :
1
2
(
) +
(1)
0; = 1,2, , 1; = +1, ,
=
1
2
(
) +
(2)
2.1.1. Notations
] are given.
=1
=1
=+1
1
=1
s. t. :
1
2
(
) +
(3)
0; = 1,2, , 1; = +1, . ,
2.2.1. Notations
, a sequence X contains
permutation numbers of all the labels. X can be represented by assigning facility X[i]
to location i, where [1, ] . Different sequences in X represent different layout
solutions. The representation of chromosome is shown in Table 1.
i 1 2 3 4 5
X[i] 5 3 2 1 4
Table 1 Chromosome Illustration
Source:(Ou-Yang & Utamima, 2013)
Table 1 shows the example of chromosome representation with 5 facilities.
X[2] = 3 means assigning facility 3 to location 2.
10
3.2. extended Chromosomes Genetic Algorithm (eACGA)
Start
Initialize all variables
Initialize population
of chromosomes
randomly
g mod 2=0
Selection
Decide cross over
rate and mutation
rate to be 0.7 and
0.3 respectively.
Yes No
Univariate and
Bivariate
Probabilistic
Modelling
Sampling
EDA GA
Crossover and
Mutation process
Population
Replacement
Fitness Calculation
Set elitism strategy
to be 10%
g = max
generation?
Stop
Yes
No
Figure 2. eACGA Flowchart
11
The eACGA method proposed by (Ou-Yang & Utamima, 2013) procedures starts
with initialization of all variables. Afterwards we will initialize the population of
chromosomes randomly. If g, the current generation of iteration is able to be divided by
2, then EDA procedure is executed, otherwise GA procedure is executed.
The EDA procedure is started with the selection of chromosomes with better
fitness values. Afterwards the selected chromosomes that are grouped is being applied
probabilistic models form by the ordinal (univariate model) and dependency (bivariate
model) matrices.
GA procedure start with assigning the crossover and mutation rate to be 0.7 and
0.3 respectively. The crossover rate and mutation rate decide whether chromosomes
mate and mutate, respectively. Roulette wheel selection is used to choose the parental
chromosomes. Two point crossover operator mates two chromosomes with better
fitness values to be the parents for the next iteration. The mutation then probably occurs
and chooses between three mutations operators that flip, swap, or slide, and that is to
be undertaken.
The chromosomes solutions afterwards will replace the previous iteration
populations. The fitness value of each chromosomes is calculated. Develop an elitism
strategy with the rate of 10%. If the generation have not exceeds the maximum iteration,
then the group of the new generation chromosomes are to be processed into the next
iteration.
12
3.3. EDAhybrid
Figure 3. EDAHybrid Pseudocode
Source: (Ou-Yang & Utamima, 2013)
3.4. EDA Part
In every generation g, a group of C particles which have better fitness are selected. The
particles are labeled as X
1
,
2
, . ,
[]
in equation (4) shows the importance of facilities in the sequence. It represents
how many time facility i is placed at position [i] at current generation. A
i[i]
k
is set to
facility 1 if facility i is placed at position [i], otherwise it is set to 0.
[]
= A
i[i]
k
=1
(4)
where i=1,.,n; k=1,.,C.
Source:(Ou-Yang & Utamima, 2013)
The bivariate or dependency probabilistic model
is set to 1 if facility
j is placed next to facility I, otherwise it is set to 0..
=1
(5)
where i,,j=1,.,n; k=1,.,C.
Source:(Ou-Yang & Utamima, 2013)
Let
[]
be the probability value of assigning facility i at position [i]. Selecting
facility i has better probability value than other facilities when both probabilistic models
statistical information is used.
[]
is added by
[]
with
[]
+
(
[]
+
(6)
where P
i[i]
: the probability value of assigning facility i at position [i], [i] = 2, 3, . . ., n;
i,j = 1, 2, . . ., n, = set of unassigned facilities. The flowchart shown in Fig. 4
demonstrates the assignment flowchart for placing facilities into location 2 until n.
Where:
= .
= .
= (0,1)
=
=
=
15
Start
Sampling the
population
Set assigned
facilities =
k n
Set using
univariate random
(0,1) distribution
Set assigned
facilities to location
to proportional
location i
Set element index of
the facilitys
position=k. k=2
Yes
Select a facility
which satisfy
<P
i[i]
,where i
End No
Figure 4. Flowchart of assignment of facilities
16
3.5. PSO Part
PSO procedure starts when current generation is odd and begins with initialization of
all related parameters. Next, the velocities calculation of every particle is operated. If
the velocity is larger, the particle is more likely to change to a new permutation
sequence. The velocity update formula remains the same like provided in previous
research, while the particle update process is changed. EDAhybrid uses permutation
based particle updating based on concept from previous research.
3.6. Tabu Search Part
Start
TabuSolution=BestSolution
i<=n
Construct Swaplist
Calculate Cost
For every move
Move is tabu
&
Moves cost > tabusolution
Give move penalty
cost
Update tabulist
Calculate minimum cost
of currentSolution
currentSolution < tabuSolution
tabuSolution = currentSolution
tabuSolution < BestSolution
BestSolution = tabuSolution
End
Y
Y
Y
N
N
N
N
Y
Figure 5 Tabu Search Algorithm
17
Figure 6. Tabu Search Pseudocode
Source:(Ou-Yang & Utamima, 2013)
The best solution from individual generated by EDA or PSO part in every
generation is the input for Tabu Search. Fig. 6 provides the pseudocode of Tabu Search
in EDAhybrid. The Tabu Search procedure is performed for n (equal to the number of
facilities) generations. Swaplist is constructed in every tabu generation; this process is
shown in line 3 of Fig. 4. Swaplist consists of three columns, the first and second
columns are the specified facility that will be swapped and the last column provides the
cost of that move.When calculating a moves cost, the algorithm checks whether the
move is taboo or not by looking in tabulist. If the move is listed in tabulist and the
moves cost is greater than tabuSolution, a penalty cost is added; otherwise the move is
accepted without a penalty cost. The algorithm records the move with minimum cost in
line 8. Tabulist is updated in line 9. If the currentSolution is better than tabuSolution,
then tabuSolution is updated. The iterations continue to run until n generations.
18
At the end of this local search, the algorithm checks whether tabuSolution, as the
output of Tabu Search, can find better solution than the current best solution. This part
is shown in line 14 until 16. If tabuSolution succeeds to get lower cost, then the current
best solution is replaced with tabuSolution.
19
4. RESULTS ANALYSIS
4.1. Parameter setting
The problem set some parameter which is general, such as population size and
generation size. Population and generation size here is based on the size of the problem.
The larger the number of facilities, the more population and generation are needed to
get an optimal solution. The number of facilities here is represented by n. The number
of generation and population size here can be described in the table below:
Number of
facilities n
Problem Size Generation Size
4-15 facilities 4n 10n
20 facilities 5n 350 generations
30 facilities 6n 800 generations
Table 2 Generation and Problem Size in Regards of number of Facilities
The elitism used in the proposed problem by (Ou-Yang & Utamima, 2013) are
10% which cited from the previous research. To determine the PSO parts parameter the
researchers use such equation to determine:
w = (maxgen currentgen)/maxgen (7)
Source: (Ou-Yang & Utamima, 2013)
Regarding the accelerators c
1
, c
2
the researchers also site from the previous
research which is [2.05,2.05] respectively. Tabu search in here is used as the local
search, so that the algorithm is set to become faster with n generations, swaplist size
equal to 2(n 1) and tabu tenure is [n/5].
20
4.2. EDAHybrid for SRFLP
In the research by (Ou-Yang & Utamima, 2013), the researchers uses 15 problem and
compare EDAHybrid with another algorithm. The researchers in here measure the error
rates percentages using such function:
error(%) =
min
min
100% (8)
The researchers also using Matlab function for standard deviation to compute the
standard deviation. The obtained value will be returned onto standard deviation using
function in Eq.(9).
s = (
1
1
(
)
2
=1
)
1
2
(9)
Several table below are the results of the simulation using Matlab code obtained from
the researchers results (Ou-Yang & Utamima, 2013):
21
4.3. EDAHybrid for enhanced SRFLP
22
5. CONCLUSION AND IMPROVEMENT
The paper proposed by (Ou-Yang & Utamima, 2013) delivers a new metaheuristic
algorithm named Hybrid Estimation of Distribution Algorithm (EDAHybrid), which
consist of hybridization of EDA, PSO, and Tabu Search. The proposed algorithm runs
with EDA and PSO alternatively every two generations and then Tabu Search as a local
search is added at the end of every iteration.
Computational results of 15 benchmark problems shows that the performance of
the proposed algorithm always achieved near optimum solution in basic SRFLP. The
error rates of the proposed algorithm is proven to be better than eACGA, EDA, PSO,
and GA. The proposed metaheuristic develop two probabilistic model in the
EDAHybrid Algorithm which is univariate and bivariate model. EDAHybrid proves to
perform better than eACGA because it adds Tabu Search as a local search. Enhanced
SRFLP problem is designed to make basic SRFLP closer to real case by adding more
constraints. Compared to other algorithms, EDAHybrid can get the lowest minimum
and error rates in enhanced SRFLP.
5.1. Improvement
In the proposed paper by (Ou-Yang & Utamima, 2013) they always stated that
their algorithm achieved an optimum value, however it is known that in NP-Complete
Problem there is no possibility that metaheuristic can reach an optimum solution.
The proposed research can be improved in the way the metaheuristic worked. We
can use another algorithm to combine with EDAHybrid instead of PSO, we can use
23
another algorithm such as the recently proposed bat algorithm. Another method is to
tweak the parameter of the algorithm to find another possible better solutions.
24
6. REFERENCES
Hauschild, M., & Pelikan, M. (2011). An introduction and survey of estimation of
distribution algorithms Swarm and Evolutionary Computation (Vol. 1, pp.
111-128).
Ou-Yang, C., & Utamima, A. (2013). Hybrid Estimation of Distribution Algorithm
for solving Single Row Facility Layout Problem. Computers & Industrial
Engineering, 66(1), 95-103. doi: 10.1016/j.cie.2013.05.018