You are on page 1of 12

Expert Systems With Applications 98 (2018) 93–104

Contents lists available at ScienceDirect

Expert Systems With Applications


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

A new hybrid heuristic algorithm based on bacterial foraging


optimization for the dynamic facility layout problem
Betül Turanoğlu∗, Gökay Akkaya
Department of Industrial Engineering, Atatürk University, Erzurum 25240, Turkey

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

Article history: One of the most important features of modern production is that the demand for the product is con-
Received 2 September 2017 stantly changing. The problems showing demand variability in certain time periods are called dynamic
Revised 5 January 2018
facility layout problems (DFLP). The establishment of a balance between handling and moving costs is
Accepted 9 January 2018
attempted with such problems. Because the DFLP are complex combinatorial optimization problems, clas-
Available online 10 January 2018
sical optimization techniques may not always be sufficient to solve them. Furthermore, it is necessary to
Keywords: find a solution within a reasonable computation time. Therefore, researchers tend to use heuristic meth-
Dynamic facility layout ods. In the literature, various heuristic methods, such as the genetic algorithm, tabu search and ant-colony
Bacterial foraging optimization algorithm, were used to solve the DFLP. The paper introduces the use of bacterial foraging optimization
Simulated annealing (BFO) to solve the DFLP. In this study, a new hybrid heuristic algorithm, called simulated annealing which
Hybrid heuristic methods is based on bacterial foraging optimization (SABFO), is proposed for the DFLP. Also, the parameter tuning
of the SABFO is made by Taguchi method. The proposed algorithm is tested on the most-used test prob-
lems in the literature, and satisfactory results are obtained at reasonable computation times. The study
is important to show how BFO that is a new heuristic approach is applied to the DFLP.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction In a facility layout problem, if the interdepartmental material


flows are assumed to be constant during all planning periods, this
Today, industrial enterprises make facility layout in order to ef- is called the static facility layout problem (SFLP). However, inter-
fectively and efficiently use machinery, equipment and manpower departmental material flows may vary by reasons such as changes
resources, which are one of the existing production factors. Facility in product design, removal or addition of a product from/to the
layout studies are studies that maximize the rate of machine uti- production line, renewal of the production equipment, changes
lization, minimize unnecessary movements, save time, enable em- in the production amount depending on the change of demand
ployees to work more efficiently and thus reduce costs. Deciding (Sahin, 2008). These situations in real life make it necessary to
on the physical organization of a production system can be de- consider facility layout as the DFLP. The purpose of the DFLP is to
scribed as a facility layout. This combinational organization prob- find a facility layout that minimizes the total of material handling
lem arises in different production activities. The most important cost and relocation cost of departments for each planning period.
thing here is to arrange manufacturing layout. In production sys- For this reason, it attempts to establish a balance between ma-
tems, the manufacturing system is a critical element. terial handling costs and relocation costs. Relocation of depart-
The material handling cost in facility layout is the most impor- ments within a facility can lead to such costs as some produc-
tant measure for determining the effectiveness of an arrangement tion losses, production control, time loss and training. However, if
and it is the most used criterion in comparision of location op- the income from the demand change is sufficiently high, rearrange-
tion. The material handling cost constitutes 20% to 50% of the total ment is economically and reasonably considered. If rearrangement
operating cost and also 15% to 70% of the total cost of a product costs are lower than material handling costs, a new facility layout
(Sahin, 2008). Therefore, it is focused on minimizing total material can be made (Ulutas & Islier, 2015).
handling cost in the facility layout problem. In this study, a new hybrid heuristic algorithm, based on the
BFO, is proposed for the DFLP. The proposed approach has been
tested using test problems, and quite good results have been ob-
∗ tained at reasonable times. Furthermore, in the study, the BFO that
Corresponding author.
E-mail addresses: b.turanoglu@atauni.edu.tr (B. Turanoğlu), is a new heuristic method is used for the first time in the DFLP
gakkaya@atauni.edu.tr (G. Akkaya). solution, and its applicability to this problem type is shown.

https://doi.org/10.1016/j.eswa.2018.01.011
0957-4174/© 2018 Elsevier Ltd. All rights reserved.
94 B. Turanoğlu, G. Akkaya / Expert Systems With Applications 98 (2018) 93–104

The paper is structured as follows: in Section 2, a mathematical modifying CRAFT and by cutting the plane algorithm, the branch-
model of the DFLP and literature review is expressed. The struc- bound algorithm, the dynamic programming algorithm and the
ture and steps of the BFO algorithm are then clarified in Section 3. tree algorithm. The performance of these five algorithms was com-
Section 4 explains the details of the proposed simulated annealing pared by generating new test problems. Conway and Venkatara-
based on the bacterial foraging optimization (SABFO) algorithm. manan (1994) developed a genetic algorithm-based heuristic for
The results of the experimental study and a comparison with other the DFLP. In the algorithm, the first generation was randomly gen-
algorithms in the literature are discussed in Section 5. Finally, the erated, and the selected generations using the roulette technique
last section present a summary of the paper and future works. were crossed according to the determined methods. The algorithm
was tested on Rosenblatt’s (1986) test problems. Balakrishnan and
2. Problem formulation and literature review Cheng (20 0 0) proposed a new genetic algorithm called the “Nested
Loop Genetic Algorithm”. This algorithm mainly consists of two
Rosenblatt first defined and addressed the DFLP in 1986. In this parts as the outer loop and the inner loop. The crossover and mu-
study, the author used dynamic programming to find the optimal tation operators were used in the inner loop. Conformance testing
facility layout. By using optimal arrangements of each period in a was developed to prevent the occurrence of noncompliant gener-
relation function, finding a single optimal solution for the whole ations as a result of the crossover. The solution space was also
problem was attempted. Because there was computation difficulty extended by the mutation process. In the outer loop, randomly
even in small the DFLP problems, two different heuristic algo- generated facility layouts were replaced by some of the weak
rithms were developed. families of the population. Baykasoğlu and Gindy (2001) were
In the DFLP, a new facility layout is obtained for each pro- the first users of the simulated annealing algorithm to solve
duction period. The mathematical model of DFLP is as follows the DFLP, and quite good results were obtained in large prob-
(Balakrishnan & Cheng, 20 0 0): lems. Balakrishnan, Cheng, Conway, and Lau (2003) developed
two different hybrid genetic algorithms. Erel, Ghosh, and Si-

T 
N 
N 
N 
T 
N 
N 
N
Min Atijl x(t−1 )ij xtil + ftik djl cik xtij xtkl (1) mon (2003) used dynamic programming and simulated anneal-
t=2 i=1 j=1 l=1 t=1 i=1 j=1 l=1 ing methods for the DFLP. Baykasoglu, Dereli, and Sabuncu (2006)
made the first attempt for the DFLP with budget constraints.

N
McKendall and Shang (2006) presented hybrid ant systems for
xti j = 1, i = 1, 2, . . . , N, t = 1, 2, . . . , T (2) the DFLP. Ulutas and Islier (2009) developed an heuristic-based
j=1 clonal-selection algorithm for the DFLP. The proposed algorithm
was tested on three different datasets, and the better results were

N
xti j = 1, j = 1, 2, . . . , N, t = 1, 2, . . . , T (3) obtained in large problems. Şahin and Türkbey (2009) proposed
i=1 a hybrid metaheurisitc method called TABUSA. Moslemipour, Lee,
and Rilling (2012) reviewed the intelligent approaches for dy-
xti j ∈ {0, 1}, ∀i, j, t namic and robust layout problems. Mazinani, Abedzadeh, and Mo-
hebali (2013) proposed a genetic algorithm approach for the DFLP.
Where;
Pourvaziri and Naderi (2014) developed a hybrid multipopulation
N: Number of departments genetic algorithm. Here, the solution area was divided into parts
T: Number of periods to provide the diversity of the algorithm. Furthermore, a simu-
ftik : Material flow between departments i and k in the period t. lated annealing-based strong local search mechanism was devel-
djl : Distance between locations j and l oped. The operators in the algorithm were designed to search only
cik : Unit cost between departments i and k. the appropriate solution. Chen, Jiang, Wahab, and Long (2015) de-
Atijl : Relocation cost of department i from location j to location veloped new mathematical models for the facility layout prob-
l in period t. lem in a logistics park. García-Hernández, Palomo-Romero, Salas-
Xtij : Variable having value of 1 if department i is assigned to Morera, and Arauzo-Azofra (2015) proposed a hybrid genetic al-
location j in period t, otherwise variable having value of 0 gorithm for capturing decision maker knowledge into the un-
equal area facility layout problem (UA-FLP). Palomo-Romero, Salas-
The objective function of the model (1) attempts to minimise Morera, and García-Hernández (2017) presented an island model
the total relocation costs and material handling costs. Constraint genetic algorithm for the UA-FLP. Kang and Chae (2017) developed
(2) means that each department is assigned to only one location a harmony search-based heuristic algorithm to solve the UA-FLP.
in each period. Constraint (3) means that only one department is Pourvaziri and Pierreval (2017) suggested an analytical approach
assigned to one location in each period. based on open queuing network theory for the DFLP. Guo, Cheng,
The DFLP are classified as NP-Hard problems. The number of and Wang (2017) proposed a hybrid evolutionary algorithm for
possible facility layout alternatives of DFLP having N number of de- two-stage capacitated facility location problem. Xiao, Xie, Kulturel-
partments and T number of periods is (N!)T . This shows computa- Konak, and Konak (2017) presented an evolution algorithm with
tion difficulty even in small problems. Therefore, different heuris- linear programming for the DFLP.
tic algorithms were proposed in the literature, which yield results
close to the best in a shorter time than analytical solution ap- 3. Bacterial foraging optimization (BFO)
proaches. Urban (1992) used lower-bound procedures to improve
the computational performance of the DFLP. He obtained these BFO is a new population-based heuristic method that imitates
procedures on the basis of the lower-bound methods developed the foraging optimization process of Escherichia coli (E. coli) bacte-
for the SFLP. Balakrishnan, Jacobs, and Venkataramanan (1992) pro- ria. The foraging mechanism of these bacteria consists of a series
posed a new model by adding a budget constraint to the DFLP, of movements towards nutritional sources. The method is based on
including rearrangement costs. The proposed model was solved the principle of assessing changes from one location to another to
by using the shortest-path algorithm, and the results of the test be a reference to the next relocation movement of E. coli bacte-
problems were compared to the dynamic programming approach. ria. The relocation between two locations is called a “movement”,
Lacksonen and Enscore (1993) utilised the quadratic assignment and each movement has a specific direction and step length. E.
problem formulation for the DFLP. They attempted to solve DFLP by coli bacteria approach nutritional sources step by step, depending
B. Turanoğlu, G. Akkaya / Expert Systems With Applications 98 (2018) 93–104 95

on their control mechanisms. Biological studies on E. coli bacteria bling movement is as follows (Bermejo et al., 2015):
show that the foraging process consists of four steps. These steps
(i )
can be listed as follows (Passino, 2002): φ (i ) =  (5)
(i )T (i )
 Searching for a reasonable nutritional location,
 Deciding whether to enter the current reasonable nutritional Where, (i)refers to a vector randomly determined in the range
location, of [−1,1] for the direction of ith bacteria.
 Performing rigorous foraging if a new nutritional location has When E. coli bacteria reach the nutrient, they release a chem-
been entered, ical substance that has a stimulant effect on other bacteria. This
 Deciding whether to migrate to a new and better nutritional substance leads other bacteria to move towards the location of the
location or stay in the same nutritional location after the con- bacteria that has found the nutrient. This is known as swarm in-
sumption of some of the nutrients available in the current nu- telligence or swarm behavior. The effect of swarm behavior in the
tritional location. algorithm is used upon request. Swarm behavior is modelled as
follows (Bermejo et al., 2015; Panda & Naik, 2015):
The BFO was first proposed by Passino (2002) for dis-
tributed optimization and control. Lately, it seems to be used 
S
Jcc (θ , P ( j, k, l )) = Jcc (θ , θ i ( j, k, l ))
to solve various engineering problems. For example; it has
i=1
been successfully applied to such engineering problems as har-
S 
  p 
monic prediction (Mishra, 2005), prediction of stock indices 2
= −dat t ract exp −wat t ract (θm −θmi )
(Majhi, Panda, Majhi, & Sahoo, 2009), multi-objective optimization m =1
i=1
problems (Guzman, Delgado, & Carvalho, 2010), image segmenta-
S 
  p 
tion (Bermejo, Cordon, Damas, & Santamaria, 2015; Sanyal, Chat- 2
+ −hrepell exp −wrepell (θm −θmi )
terjee, & Munshi, 2011; Sathya & Kayalvizhi, 2011a, b), optimal ac- m=1
i =1
tive power rescheduling of generators (Venkaiah & Kumar, 2011),
numerical optimization (Chatzis & Koukas, 2011), optimal multi- Jcc (θ ,P(j, k, l)): Total cell-to-cell attractant value added to objec-
level thresholding (Sathya & Kayalvizhi, 2011a, b), modelling of ma- tive function in jth chemotaxis step
nipulator systems (Supriyono & Tokhi, 2012), environmental eco- P(j, k, l) = {θ i (j, k, l)|i = 1, 2, ..., S}: The position of ith bacterium
nomic dispatch (Pandit, Tripathi, Tapaswi, & Pandit, 2012), vehicle at jth chemotactic step, kth reproduction step, and lth elimination-
routing problems (Hezer & Kara, 2013), optimal placement of dis- dispersal events.
tributed generation (Devi & Geethanjali, 2014), RFID network op- S: Total number of bacteria
timization (Chen, Zhu, Hu, & Ma, 2014), multicycle transient fault- p: The size of the problem
tolerant datapath (Sengupta & Bhadauria, 2015), optimization of ra- dattract : Coefficient of the cell-to-cell attractant signal
dial distribution systems (Mohammadi, 2015), controller design for wattract : Spreading speed of the cell-to-cell attractant signal
electrohydraulic systems (Mandal et al., 2015), linear discriminant hrepell : Coefficient of the cell-to-cell repellant effect
analysis (Panda & Naik, 2015) and global optimization (Kim, Abra- wrepell : Cell-to-cell spreading speed
ham, & Cho, 2007; Zhao & Wang, 2016).
BFO consists of three main phases: chemotaxis, reproduction • Reproduction: After a foraging period, it is obvious that the for-
and elimination-dispersal (Passino, 2002): aging strategies of some bacteria clearly fail. These bacteria are
removed from the population because of the low probability of
• Chemotaxis: The bacteria move forward if E. coli bacteria ro- obtaining adequate nutrients. To keep the population number
tate counter-clockwise via their flagellas, whereas the bacteria constant, bacteria with good foraging strategies are reproduced
slow down and are released wherever they are if flagellas rotate to replace those removed in the same amount. First, each bac-
clockwise. The foraging of E. coli bacteria is based on changes terium is sequenced according to its assessment criteria in its
between the last two behaviors, and these two behaviors result location. Then, the bacteria in the last half of the sequence are
in the so-called chemotactic step of the bacteria. During the for- removed from the population, and each bacterium in the first
aging process of the bacteria, the rotating of flagellas is based half is reproduced so that they will be in the same location as
on assessing the current bacterial environment. It is then de- the originals. This process is the imitation of a kind of bacteria
cided whether the current location will be changed and, if so, mitosis.
how it will be changed in light of some parameters (direction • Elimination-dispersal: The behaviors of the bacteria are af-
and step length of the next movement). The formula for chang- fected significantly by the environmental conditions they live
ing direction in the BFO is as follows: in (temperature, fast water flow, etc.). These factors may cause
the death of the bacteria in that area or the relocation of some
θ i ( j + 1, k, l ) = θ i ( j, k, l ) + C (i )φ (i ) (4)
of them by moving. Although elimination and dispersal may
Where; have an adverse effect on chemotaxis, they may also have a
positive effect by allowing bacteria to move to the richer ar-
i: Bacterial index, eas. The elimination-dispersal event is carried out depending
j: Chemotaxis index, on a certain probability (Ped ). If a bacterium is exposed to an
k: Reproduction index, elimination-dispersal event, that bacterium is eliminated and a
l: Elimination-dispersal index, new bacterium is reproduced. This means the relocation of the
θ i (j,k, l): the current location of ith bacterium, said bacteria.
θ i (j + 1, k, l): the next location of ith bacterium,
C(i): the swimming or tumbling step length of ith bacterium, BFO can be explained step by step as follows (Passino, 2002):
ϕ (i): the tumbling movement of ith bacterium. Step 1: The initialization of all stated parameters: n, S, Nc , Ns ,
Nre , Ned , Ped , C(i) (i = 1,2,…,S). Where;
During chemotaxis, bacteria will continue to swim in the same n: the dimension of the search space,
direction or tumble, changing direction, depending on the nutri- S: the number of bacteria in the population,
tional concentration. The mathematical representation of the tum- Nc : the number of chemotactic steps,
96 B. Turanoğlu, G. Akkaya / Expert Systems With Applications 98 (2018) 93–104

Ns : the swimming length, • Bacteria: It is a suitable solution for DFLP. Fig. 2 shows the
Nre : the number of reproduction steps, coded version of this solution.
Ned : the number of elimination-dispersal steps, • J(i, j, k, l): The objective function value of ith bacterium at jth
Ped : the elimination-dispersal probability, chemotactic step, kth reproduction step, and lth elimination-
C(i): the step size of ith bacterium. dispersal events. So, it is the total cost of ith facility layout.
Step 2: Elimination-dispersal loop: l = l + 1 • θ i (j + 1, k, l): The position of ith bacterium at (j + 1)th chemo-
Step 3: Reproduction loop: k = k + 1 tactic step, kth reproduction step, and lth elimination-dispersal
Step 4: Chemotaxis loop: j = j + 1 events. In other words, it is the new layout obtained by modi-
Step 4.1: For each bacterium (i = 1,2,…,S), initialize a chemo- fying the existing layout (θ i (j,k, l)).
tactic step as follows.
• J(i, j + 1, k, l): It is the value of the new objective function cal-
Step 4.2: Calculate objective function J(i,j,k,l). If the effect of culated according to θ i (j + 1, k, l).
swarm behavior will be used, add value Jcc (θ ,P(j, k, l)) to
this value. 4. Proposed hybrid heuristic algorithm for DFLP: SABFO
Step 4.3: Set Jlast = J(i,j,k,l).
Step 4.4: Create a random vector (i) ∈ Rn in the range of The proposed SABFO algorithm is basically composed of two
[−1,1], including m (i), m = 1, 2, ..., n. steps. In the first step, an initial solution is obtained with the BFO.
Step 4.5 (Moving): Using Eq. (4), move as much as step C(i) Starting an algorithm from a good solution in heuristic methods
in the direction of tumbling for ith bacterium. helps to reduce computation time, in particular. Therefore, a good
Step 4.6: Calculate J(i, j + 1, k) using θ i (j + 1, k, l) initial solution is produced by using the modified BFO given in
Step 4.7 (Swimming): Section 4.2.1. The solution used in SA is the second step of SABFO.
The SA method that is a strong local search strategy continues to
(i) Set the initial value of ms = 0 to be the counter of improve the solution until the initial solution meets the stop cri-
the swimming length. terion. The neighbors of the initial solution are discovered by this
(ii) as long as ms < Ns ; strategy because the optimal solution may be close to this solu-
• ms = ms + 1;
tion. When the algorithm stops, the final solution and the cost of
• If J(i, j + 1, k) < J
last , set Jlast = J(i, j + 1, k). the problem are given as output.
• Using the same direction vector θ i (j + 1, k, l) re-

produce and use this value to calculate the new 4.1. Encoding of a solution
J(i, j + 1, k) value.
• If ms = Ns , stop the swimming process for this
As in many heuristic algorithms, the first step in SABFO is to
bacterium. encode the problem in a way that is appropriate to the algorithm.
Step 4.8: If i = S, go to Step 4.2 with the next bacterium. Each solution for the DFLP, in other words facility layout, is ex-
Step 5: If j < Nc , it means that the life of the bacterium is not pressed as a matrix according to periods. In this matrix; each line
terminated. Go to Step 4. shows a period; the number of elements in a line shows the total
Step 6 (Reproduction): number of departments in the location and the location in which
each element is located shows which location that element (de-
Step 6.1: Calculate the Jhealth value as follows for the current partment) is assigned. Fig. 1 shows sample DFLP consisting of 3
k and l values and each i = 1,2,…,S. This function shows periods and 6 departments. Fig. 2 shows the coded solution of
how much nutrient a bacterium consumes during its life- this problem. Here, department 4 is assigned to location 6 in time
time and how successful it is to stay away from harmful period 1, location 4 in time period 2 and, finally, location 1 in
substances. In other words, it means the health status of time period 3. The solution obtained in this way provides the con-
ith bacterium. Sort bacteria from small to large according straints of assigning one department to only one location and a lo-
to Jhealth values. cation including only one department. Therefore, the total cost of

Nc+1 the solution, including the material handling costs and relocation
Jhealth = J (i, j, k, l ) cost, is easily calculated.
j=1

Step 6.2: Sr = S/2 bacteria, whose health status in the lower 4.2. Obtaining initial solution
half of the sequence is worse than the others, are con-
sidered dead. In this case, the rest of Sr bacteria, whose In heuristic algorithms, starting an algorithm from a good so-
health status in the upper part of the sequence are better, lution has an important effect both in terms of computation time
are reproduced by dividing. Each new generation bacte- and in the approaching the optimal solution or in approaching this
rial pair replaces the dead bacteria at the bottom of the solution more quickly. In SABFO approach developed for this rea-
list so that they are placed in the same positions as their son, the initial solution to be used in simulated annealing was ob-
parents. tained with BFO. For this reason, the modified BFO was used in-
stead of the classic BFO.
Step 7: If k < Nre , it means that the specified reproduction upper
limit has not been reached; go back to Step 3 and carry out
4.2.1. Structure of the modified BFO
the chemotactic steps with the next generation.
Bacteria start the chemotaxis loop with the values of the objec-
Step 8 (Elimination-Dispersal): Expose all bacteria in
tive function (J) obtained by the initial locations and the solution of
i = 1,2,…,S sequence to an elimination and dispersal process
these locations. The direction vector ((i)) determined in the range
depending on Ped probability. This process is carried out by
of [+ 1, −1] in the classical BFO is randomly generated according
random relocation of a bacterium from its current location
to the set of discrete values {−1, 0, +1} due to the problem struc-
and by keeping the population constant.
ture and has as many elements as the number of departments. For
Step 9: If l < Ned, go to Step 2; otherwise end the algorithm.
example, a direction vector may be created as (i) = { − 1, 0, 0, 1,
Some notations in the BFO, given the basic steps above, can be 0, −1} for DFLP with 6 departments. The department correspond-
expressed for the DFLP as follows: ing to the value “−1” is replaced by the department subsequent to
B. Turanoğlu, G. Akkaya / Expert Systems With Applications 98 (2018) 93–104 97

Fig. 1. A sample for DFLP.

such bacteria. In the swimming phase of the modified BFO, even if


a nutrient concentration does not increase in the specified direc-
tion, a nutrient scan is performed around that area. This process
can be seen as a search for neighbors of a solution of any kind.
While performing this search, the pairwise interchange heuristic
including the following steps is used (McKendall & Shang, 2006):
Fig. 2. A coded solution for DFLP.
Step 1: Calculate the cost (f(π )) of the facility layout (π ) in the
randomly determined direction.
Step 2: Select a random time period (t).
Step 3: Select two random departments again in the selected
period, and change the location of these departments to ob-
tain a new facility layout (π  ).
Step 4: Calculate the cost of the new facility layout (f(π ) ).
Step 5: Calculate the change in total cost ( f = f (π ) − f (π  )).
If f > 0, accept the change and do π = π  . Otherwise,
refuse the exchange.
Step 6: Repeat these steps as much as the swimming length
(Ns ).
The reproduction process in modified BFO is the same as that
used in the original BFO. In other words, the values of the objective
functions that each bacterium obtained at the end of each chemo-
Fig. 3. Application of direction vector to a bacteria. tactic step in the chemotaxis process are collected, and the related
value is expressed as Jhealth value. After performing the same pro-
cedure for all the bacteria that make up the population, the bac-
itself as much as the step length C, whereas the department corre- teria are sequenced according to Jhealth value, the bad half of the
sponding to the value “+1” is replaced by the department preced- sequence is eliminated from the population and the good half is
ing it as much as the step length C. If a department corresponds to copied to the location of the eliminated part. In the process of
a value of “0”, it maintains the same position. In this study, the di- elimination and dispersal, a random value is generated in the range
rection vector determined to a facility layout representing the bac- of (0–1) for each bacterium. If this value is less than the probabil-
teria was applied as follows: ity value Ped , the bacteria are eliminated. Instead of bacteria being
• A random period is selected. eliminated, a random new bacterium is reproduced and added to
• Considering a random department and the subsequent depart- the population.
ment of that department in the selected period, a part is
4.3. Simulated annealing
deemed to have been taken in such a period.
• In the specified direction vector, a location change is made ac-
Simulated annealing (SA) is a method inspired by the annealing
cording to the values corresponding to this part.
process, which refers to the process of heating and slowly cooling
• For all bacteria, the step length is assumed to be C = 1.
the solids. This method is based on the study of Metropolis, Rosen-
Fig. 3 gives an example of these steps. In Fig. 3a, department bluth, Rosenbluth, Teller, and Teller (1958). In the study, balanced
5 placed at location 2 in the randomly selected period was ran- dispersal of atoms and energy changes were studied at a cer-
domly selected. After department 3 subsequent to this department tain temperature level. Pincus (1970) emphasised that SA is linked
is taken, the direction vector will be applied to this binary com- to mathematical minimization. However, Kirkpatrick, Gerlatt, and
ponent. In the direction vector given in Fig. 3b, the values corre- Vecchi (1983) suggested for the first time that this approach can
sponding to the parts constituted by departments 5 and 3 are the be used as an optimization technique.
values “0” and “−1”. In this case, the location of the first depart- The most important advantage of SA is its ability to get rid
ment that is department 5 will not change, and department 3 will of the local optimum to find the global optimum. It obtains this
be replaced by the previous location. Finally, a location change, as property with the possibility of accepting worse neighborhood so-
shown in Fig. 3c, will occur. lutions than the current solution. The probability of acceptance is
In classical BFO, bacteria are moved in the specified direction. determined as a control parameter that decreases over the dura-
If there is improvement in the solution, it continues to move in tion of the SA period (Baykasoğlu & Gindy, 2001). SA is very simi-
the same direction and cost value J at each step gets the solution lar to other local search algorithms. However, these techniques are
value corresponding to the step taken. This process, called swim- usually stuck in the local optimum. Accepting bad solutions in a
ming in the algorithm, is repeated until the bacteria heal or reach controlled way or moving up the hill (known as “hill climbing”), is
the maximum number of swimming (Ns ). If one of these criteria the solution of SA for this problem. The SA algorithm is a kind of
is met, a chemotactic step is deemed to have been performed for threshold algorithm because it allows hill climbing.
98 B. Turanoğlu, G. Akkaya / Expert Systems With Applications 98 (2018) 93–104

Table 1
The parameters and their levels.

Parameters Levels

Symbol Description Level 1 Level 2 Level 3

A Number of bacteria 30 50 100


B Number of elimination-dispersal steps 2 3 4
C Number of reproduction steps 2 3 4
D Number of chemotaxis steps 2 3 5
E Number of swimming steps 100 200 500
F Elimination-dispersal probability 0.01 0.05 0.1
G Initial temperature 10 0 0 20 0 0 50 0 0
H Cooling rate 0.9 0.999 0.9999
J Maximum number of iterations 500 10 0 0 50 0 0

The basic steps of simulated annealing can be summarised as in the design of heuristic methods. The full factorial design leads
follows (McKendall, Shang, & Kuppusamy, 2006): to an increase in the total number of experiments depending on
the parameter, the parameter level, and the increase in experimen-
(i) Heuristic parameter values are selected.
tal repetition in each parameter combination. This also requires
(ii) An initial solution is generated and the cost is calculated.
higher time and cost investments. However, the Taguchi method
This solution is defined as the current solution.
saves time and money by minimizing the number of required ex-
(iii) A neighborhood solution is obtained using one of the local
periments, even if the parameters and their levels increase. In ad-
search techniques from the current solution.
dition, it provides a more efficient tuning procedure for the inves-
(iv) The cost of the neighborhood solution is calculated and
tigator.
compared to the cost of the current solution:
• If the cost is better, the neighborhood solution is consid-
The Taguchi method is a useful method to determine the best
combination among different levels of different parameters. In
ered the current solution.
• If the cost is worse, the neighborhood solution is consid-
cases where it is necessary to perform a significant number of ex-
periments for each combination of each parameter, including each
ered the current solution with the determined probabil-
level, it is possible to achieve results with far fewer experimen-
ity. Otherwise, the current solution will not change.
tal studies using orthogonal arrays in the Taguchi method. In this
(v) The counters and parameters are updated, and steps 2–4 are
method, factors affecting experimental work are classified into two
repeated until the stopping criterion is reached.
main groups: controllable factors and noise factors. With this ap-
Choosing the correct parameters for a successful SA algorithm proach, the optimal levels of controllable factors are determined,
is very important. The parameters are as follows: and the effect of noise factors is minimized.
• Initial temperature (Tb ): The initial temperature should be Taguchi has improved a transformation of the repetition data
chosen high enough so that the vast majority of the solutions to another value which is a measure of the variation present.
generated after the algorithm starts can be accepted. The value The transformation is the signal-to-noise (S/N) ratio. The ratio
of the initial temperature can be determined by means of ex- combines several repetitions into one value which represents the
periments. amount of variation present. In experimental designs, the maxi-
• Number of solutions to be generated at each temperature mizaiton of the S/N ratio is aimed. There are several S/N ratios
(CS): It is necessary to generate the solution enough to allow available depending on the type of the objective function. These
the system to reach the equilibrium state at any temperature are “lower is better (LB)”, “nominal is best (NB)” and “higher is
without decreasing the temperature value in SA. In this study, better (HB)”(Ross, 1988). Since objective function of DFLP is min-
the number of solutions is equal to multiplication of the num- imization of total cost, its corresponding S/N ratio is:
ber of periods and the number of departments (N∗ T) of the
problem.

1  2
M
• Cooling rate (a): According to the annealing process, the suffi- S/NLB = −10log10 Yi (6)
ciently high initial temperature should be cooled slowly. Cool- M
i=1
ing rate can be determined experimentally, or this value can be
calculated with the methods proposed in the literature.
where Yi is performance characteristic value of the ith instance
• Stopping criteria: In an SA algorithm, criteria such as the num-
(i = 1,2,…,M). The best level of parameters or factors used in an
ber of maximum iterations and the ending temperature can be
experimental study in the Taguchi method is determined by or-
used as the stopping criterion. A function is used to reduce the
thogonal arrays. The selection of orthogonal arrays depends on
temperature at each iteration. One of the most used functions
the number of parameters and the number of levels of these pa-
is given below:
rameters. The proposed SABFO algorithm contains nine parame-
T(m+1) = Tm ∗ a ters, each with three levels. The levels of these parameters are
In this geometric function, Tm is the current temperature and a shown in Table 1. For an experiment with nine parameters of
is the cooling rate. The algorithm is stopped when it reaches the three levels each, the L27 orthogonal array was determined to
maximum number of iterations specified or the ending tempera- be the most suitable for use in this study. First, three different
ture. problems were randomly selected among the small-, medium- and
large-sized problems of the 48 problem data sets: P3, P17 and
5. Parameter tuning with Taguchi method P34. For these three problems, the SABFO algorithm was run three
times according to each parameter combination of the L27 array.
Parameter design greatly influences the performance of heuris- Thus, a total of 243 trials were made. To measure the performance
tic methods. For this reason, statistical experimental design meth- of SABFO in each problem, the objective function is transformed
ods, such as full factorial design and the Taguchi method, are used into its relative percentage deviation (RPD) described as follows
B. Turanoğlu, G. Akkaya / Expert Systems With Applications 98 (2018) 93–104 99

Table 2
L27 orthogonal array and the results of experiments.

Experiment No Parameters RPD3 RPD17 RPD34 S/N ratio

A B C D E F G H J

1 1 1 1 1 1 1 1 1 1 0,018852 0,0194477 0,026088 33,2662


2 1 1 1 1 2 2 2 2 2 0,007205 0,0206756 0,024764 34,3864
3 1 1 1 1 3 3 3 3 3 0,001655 0,0056423 0,006910 45,6159
4 1 2 2 2 1 1 1 2 2 0,008583 0,0208536 0,025466 34,1377
5 1 2 2 2 2 2 2 3 3 0,002467 0,0043937 0,0 0170 0 50,2563
6 1 2 2 2 3 3 3 1 1 0,028198 0,0198628 0,030988 31,4469
7 1 3 3 3 1 1 1 3 3 0,003917 0,0030155 0,003090 49,4581
8 1 3 3 3 2 2 2 1 1 0,012708 0,0232864 0,030699 32,6065
9 1 3 3 3 3 3 3 2 2 0,010950 0,0213719 0,028165 33,4043
10 2 1 2 3 1 2 3 1 2 0,009161 0,0208633 0,030563 33,1477
11 2 1 2 3 2 3 1 2 3 0,016044 0,0179947 0,024097 34,1197
12 2 1 2 3 3 1 2 3 1 0,067747 0,0630730 0,144903 20,0635
13 2 2 3 1 1 2 3 2 3 0,003579 0,0162895 0,027893 34,5340
14 2 2 3 1 2 3 1 3 1 0,019641 0,0187036 0,072727 26,9718
15 2 2 3 1 3 1 2 1 2 0,006068 0,0203693 0,026665 34,1164
16 2 3 1 2 1 2 3 3 1 0,101997 0,0959839 0,178487 17,6554
17 2 3 1 2 2 3 1 1 2 0,019322 0,0294415 0,031388 31,2972
18 2 3 1 2 3 1 2 2 3 0,014804 0,0232455 0,023856 33,5372
19 3 1 3 2 1 3 2 1 3 0,003476 0,0246023 0,030228 32,9212
20 3 1 3 2 2 1 3 2 1 0,006854 0,0163976 0,028536 34,2398
21 3 1 3 2 3 2 1 3 2 0,018680 0,0067918 0,015515 36,7382
22 3 2 1 3 1 3 2 2 1 0,005991 0,0197519 0,030173 33,5117
23 3 2 1 3 2 1 3 3 2 0,039094 0,0634707 0,142093 20,6639
24 3 2 1 3 3 2 1 1 3 0,004445 0,0273753 0,036293 31,5774
25 3 3 2 1 1 3 2 3 2 0,017961 0,0339890 0,104383 23,8462
26 3 3 2 1 2 1 3 1 3 0,013436 0,0267835 0,023936 33,0957
27 3 3 2 1 3 2 1 2 1 0,015216 0,0220829 0,028274 32,9568

Fig. 4. The mean S/N ratio plot for each level of the parameters.

(Pourvaziri and Naderi, 2014): Table 3


The best levels of the parameters.
total cos ti, j − LBi
RP Di, j = (7) Symbol Description Best levels
LBi
A Number of bacteria 30
where total costi,j denotes the total cost obtained for ith problem B Number of elimination-dispersal steps 2
in jth replication and LBi is the minimum total cost of ith problem. C Number of reproduction steps 4
RPDi,j symbolizes the RPD value of ith problem in jth replication. D Number of chemotaxis steps 3
E Number of swimming steps 500
The RPD value of the ith problem (RPDi ) was obtained by aver- F Elimination-dispersal probability 0.05
aging the RPD values of the three experiment replicates for each G Initial temperature 10 0 0
problem. In Formula 6, RPDi was used as the value of Yi . H Cooling rate 0.999
Table 2 shows the orthogonal array L27, RPDi value and S/N ra- J Maximum number of iterations 50 0 0
tio for each experimental trial. The mean S/N ratio plot for each
level of the parameters is shown in Fig. 4. Since the aim is max-
imization of the S/N ratio, the level with highest S/N ratio is se-
lected as best level. Table 3 shows the best levels of the param- using analysis of variance (ANOVA). As shown in Table 4, the most
eters. Also, the most significant parameter can be determined by important parameters are the maximum number of iterations, the
100 B. Turanoğlu, G. Akkaya / Expert Systems With Applications 98 (2018) 93–104

Table 4
The ANOVA of the S/N ratio.

Parameters Degrees of freedom Seq. sum of squares Adj. sum of squares Adj. mean squares F-value P-value

Number of bacteria 2 395.93 395.927 197.964 3.36 .087


Number of elimination-dispersal steps 2 15.46 15.465 7.732 0.13 .879
Number of reproduction steps 2 64.26 64.257 32.128 0.55 .599
Number of chemotaxis steps 2 11.25 11.248 5.624 0.10 .910
Number of swimming steps 2 2.91 2.912 1.456 0.02 .976
Elimination-dispersal probability 2 8.98 8.983 4.492 0.08 .927
Initial temperature 2 39.94 39.936 19.968 0.34 .722
Cooling rate 2 11.76 11.762 5.881 0.10 .906
Maximum number of iterations 2 413.62 413.618 206.809 3.52 .080
Error 8 470.65 470.646 58.831
Total 26 1434.75

number of bacteria and the number of reproduction steps since time. GADP (U) obtains the best results in 2 problems. DP_10S
their P-values are close to zero. and DP_10SI attain the best result in 1 problem. In these prob-
lems, it can be said that DP_10SI is better than these methods in
6. Computational study finding the best solution in the most effective time. The proposed
approach achieves the best result in 3 of the problems with 30
To measure the effectiveness of the proposed SABFO algorithm, departments and 10 periods. However, it seems to be better than
the test problems of Balakrishnan and Cheng (20 0 0) are used. the DP_10LI algorithm, which finds the best solution in 3 of these
These problems consist of 6, 15, and 30 departments, and 5 and problems, as CPU time (Table 10).
10 periods are considered for each problem. The proposed algo- SABFO finds better results than other methods at a consider-
rithm is coded in the MATLAB programming language. The dataset ably better computation time for all small problems (P1-P16). In
consists of 48 problems tests on a PC with the Intel (R) Core (TM) medium problems (P17-P32), it finds the 5 best solutions. Although
i7-4770 3.4 GHz. The results of SABFO are compared to the follow- the SA_EG_1 algorithm obtains the best solution in 11 of these
ing studies: problems, the CPU time spent finding the solution is several times
higher than that in SABFO. For the problems between Ps33 and
• Conway and Venkataramanan (1994) – Conway and Venkatara- P40, SABFO acquires the best solution in 5 of them and at a very
manan Genetic Algorithm (CVGA). reasonable computation time. For large problems (P41-P48), SABFO
• Balakrishnan and Cheng (20 0 0) – Nested Loop Genetic Algo- 3, GADP(U) 1, DP_10L 1, DP_10LI 3, DP_10SI and DP_10SI 1, respec-
rithm (NLGA). tively, obtain the best solution for the problem. Hovewer, it can be
• Balakrishnan et al. (2003) – Genetic Algorithms based on Dy- said that SABFO is better than GADP(U), DP_10L and DP_10LI as
namic Programming (GADP/GADP(R)/GADP(U)) CPU time.
• Erel et al. (2003) – Dynamic Programming and Simulated In general, SABFO achieves more effective results than other
Annealing algorithms (DP_10, DP_10I, DP_5, DP_5I, DP_10L, population-based approaches, such as CVGA, NLGA, GADP (R) and
DP_10LI, DP_5L, DP_5LI, DP_10S, DP_10SI, DP_5S, DP_5SI, SA_EG_1, GADP (U). It is also seen that SABFO is better both in solution value
SA_EG_2) and computation time than SA_EG_1, SA_EG_2 and SA_Robust ap-
• Baykasoğlu, Dereli, and Sabuncu (2006) – Ant Colony Optimiza- proaches where stimulated annealing are used. SABFO obtains the
tion (ACO) best solution value in 29 of 48 problems.
• Pillai, Hunagund, and Krishnan (2011) – Simulated Annealing
based on Robust Model (SA_Robust) 7. Conclusion and future work
The comparative results of the examined algorithms are given
Since the DFLP have NP-Hard structures, the solution time ex-
in the tables in Appendix A. In these tables, the result values in
ponentially increases as the problem size increases. This leads re-
bold are the best solution values. Tables 5 and 6 show the re-
searchers to use heuristic rather than analytical methods. One of
sults of the problems with 6 departments and 5 and 10 periods,
the most important reasons for this is the desire for a decision
respectively. SABFO has obtained the best results for all 8 prob-
without spending too much time on the problem being addressed.
lems, consisting of 6 departments and 5 periods, in reasonable CPU
Although heuristic methods do not guarantee the optimal solution,
time. Table 6 shows that SABFO is the only algorithm that finds the
they yield close results to the optimum in a reasonable computa-
best solutions for all problems with 6 departments and 10 periods.
tion time. This study presents a BFO-based new hybrid heuristic
These results were obtained in a highly efficient CPU time.
method for the DFLP. The BFO is a new heuristic method used for
In Table 7, the results obtained for the problems consisting of
the first time in the DFLP. In the proposed method, the initial solu-
15 departments and 5 periods are seen. SABFO finds the best solu-
tion obtained by BFO enters into the simulated annealing process.
tion in only one of these problems. SA_EG_1 algorithm obtaines
The so-called SABFO method, in general, yields promising results
the best results for other problems. However, SABFO is able to
at an acceptable CPU time. The difference between the best result
reach the best solution with an average deviation of 0.8% in a
values is close to 0%. Also, the Taguchi method is used to design
shorter time than SA_EG_1 as the computation time. Table 8 shows
the parameters of SABFO. The study contributes to both DFLP and
the results of problems with 15 departments and 10 periods, and
BFO literature.
that SABFO found the best solution in 4 problems under a fairly
The suggestions can be given for future studies as follows:
reasonable CPU time. In the remaining 4 problems, SA_EG_1 ob-
tains the best results. • New algorithms can be developed by hybridising BFO and other
As shown in Table 9, the proposed SABFO algorithm finds the heuristic methods for the DFLP.
best solution in 5 problems, consisting of 30 departments and • Different parameter values, direction vectors and local search
5 periods. For the remaining 3 problems, it approaches the best techniques of the proposed approach can be used in the BFO
solution with an average deviation of 0.2% at a reasonable CPU phase.
B. Turanoğlu, G. Akkaya / Expert Systems With Applications 98 (2018) 93–104 101

• In elimination and reproduction processes of the BFO, different • New heuristic approaches can be developed for the DFLP with
approaches similar to those used in the process of generating a budget constraint and multi-objective.
new individual and mutating the processes of the genetic algo- • Multiple floors, unequal areas and stochastic flows cases of the
rithm can be tried. DFLP can be tested by using BFO algorithm.
• The performance of the BFO can be evaluated by applying it to • It can be considered that rearrangement costs change during
other types of facility layout problems. planning periods.
• BFO can be applied to different combinatorial optimization
problems.
Appendix A

Table 5
Results for problems with N = 6 and T = 5.

Description Problem 1 Problem 2 Problem 3 Problem 4 Problem 5 Problem 6 Problem 7 Problem 8 Mean CPU Time(in second)

CVGA 108,976 105,170 104,520 106,719 105,628 105,606 106,439 104,485 NA


NGLA 106,419 104,834 104,320 106,515 105,628 104,053 106,978 103,771 NA
GADP 106,419 104,834 104,529 106,583 105,628 104,315 106,447 103,771 NA
DP_10 106,419 104,834 104,320 106,509 105,628 103,985 106,447 103,771 <1
DP_10I 106,419 104,834 104,320 106,509 105,628 103,985 106,447 103,771 <1
DP_5 106,419 104,834 104,320 106,885 105,737 104,053 106,447 104,185 <1
DP_5I 106,419 104,834 104,320 106,515 105,737 104,053 106,447 104,185 <1
SA_EG_1 106,419 104,834 104,520 106,399 105,737 103,985 106,439 103,771 55
SA_EG_2 106,419 104,834 104,320 106,399 105,628 103,985 106,439 103,771 52
ACO 106,419 104,834 104,320 106,509 105,628 104,053 106,439 103,771 70
SA (Robust) 106,419 105,731 107,650 108,260 108,188 107,765 108,114 107,248 NA
SABFO 106,419 104,834 104,320 106,399 105,628 103,985 106,439 103,771 4.61
Best solution 106,419 104,834 104,320 106,399 105,628 103,985 106,439 103,771
% Deviation of SABFO value 0 0 0 0 0 0 0 0
from best solution

Table 6
Results for problems with N = 6 and T = 10.

Description Problem 9 Problem 10 Problem 11 Problem 12 Problem 13 Problem 14 Problem 15 Problem 16 Mean CPU Time (in second)

CVGA 218,407 215,623 211,028 217,493 215,363 215,564 220,529 216,291 NA


NGLA 214,397 212,138 208,453 212,953 211,575 210,801 215,685 214,657 NA
GADP 214,313 212,134 207,987 212,741 210,944 210,0 0 0 215,452 212,588 NA
DP_10 214,313 212,134 207,987 212,741 211,022 209,932 214,252 212,588 <1
DP_10I 214,313 212,134 207,987 212,741 211,022 209,932 214,252 212,588 <1
DP_5 214,313 212,138 208,246 213,117 211,022 210,0 0 0 214,252 213,002 <1
DP_5I 214,313 212,138 208,060 212,747 211,022 210,0 0 0 214,252 213,002 <1
SA_EG_1 214,313 212,134 207,987 212,747 211,076 210,0 0 0 214,823 212,588 215
SA_EG_2 214,313 213,015 208,351 212,747 211,072 209,932 214,438 212,588 206
ACO 217,251 216,055 208,185 212,951 211,076 210,277 215,504 214,621 1326
SA (Robust) 220,776 217,412 219,024 217,350 217,142 217,397 219,788 220,144 NA
SABFO 214,313 212,134 207,987 212,530 210,906 209,932 214,252 212,588 14.27
Best solution 214,313 212,134 207,987 212,530 210,906 209,932 214,252 212,588
% Deviation of SABFO value 0 0 0 0 0 0 0 0
from best solution
102 B. Turanoğlu, G. Akkaya / Expert Systems With Applications 98 (2018) 93–104

Table 7
Results for problems with N = 15 and T = 5.

Description Problem 17 Problem 18 Problem 19 Problem 20 Problem 21 Problem 22 Problem 23 Problem 24 Mean CPU Time (in second)

CVGA 504,759 514,718 516,063 508,532 515,599 509,384 512,508 514,839 NA


NGLA 511,854 507,694 518,461 514,242 512,834 513,763 512,722 521,116 NA
GADP (R) 493,707 494,476 506,684 500,826 502,409 497,382 494,316 500,779 NA
GADP (U) 484,090 485,352 489,898 484,625 489,885 488,640 489,378 500,779 NA
DP_10L 484,054 489,322 491,310 487,884 491,617 490,205 490,544 494,994 111
DP_10LI 483,568 489,322 491,310 487,275 491,346 489,847 490,051 493,577 119
DP_5L 484,972 491,102 493,632 489,929 494,040 490,782 491,984 496,841 20
DP_5LI 482,123 488,840 493,632 489,480 494,040 490,782 490,251 496,672 28
DP_10S 484,369 487,274 491,790 487,956 491,178 490,305 490,161 494,954 14
DP_10SI 483,708 485,702 491,790 486,851 491,178 489,947 489,583 494,534 22
DP_5S 484,369 489,819 493,224 489,698 493,097 492,275 492,430 496,990 2
DP_5SI 483,708 488,382 492,597 489,698 491,738 492,202 489,155 496,473 10
SA_EG_1 481,378 478,816 487,886 481,628 484,177 482,321 485,384 489,072 1635
SA_EG_2 481,792 488,592 492,536 485,862 489,946 488,452 487,576 493,030 946
ACO 501,447 506,236 512,886 504,956 509,636 508,215 508,848 512,320 167
SA (Robust) 506,847 500,284 508,011 503,699 502,622 499,891 502,919 507,970 NA
SABFO 480,453 484,853 489,981 486,006 488,556 488,196 487,476 491,789 145.9
Best solution 480,453 478,816 487,886 481,628 484,177 482,321 485,384 489,072
% Deviation of SABFO value 0 1.26 0.43 0.90 0.90 1.22 0.43 0.56
from best solution

Table 8
Results for problems with N = 15 and T = 10.

Description Problem 25 Problem 26 Problem 27 Problem 28 Problem 29 Problem 30 Problem 31 Problem 32 Mean CPU Time (in second)

CVGA 1,055,563 1,061,940 1,073,603 1,060,034 1,064,692 1,066,370 1,066,617 1,068,216 NA


NGLA 1,047,596 1,037,580 1,056,185 1,026,789 1,033,591 1,028,606 1,043,823 1,048,853 NA
GADP (R) 1,004,806 1,006,790 1,012,482 1,001,795 1,005,988 1,002,871 1,019,645 1,010,772 NA
GADP (U) 987,887 980,638 985,886 976,025 982,778 973,912 982,872 987,789 NA
DP_10L 986,811 985,154 989,081 979,139 986,029 976,917 985,535 990,844 712
DP_10LI 984,344 984,779 988,635 976,456 983,846 974,436 982,790 990,372 724
DP_5L 991,093 987,453 993,799 983,208 989,680 979,297 992,897 992,962 55
DP_5LI 988,322 985,147 993,318 982,632 985,966 978,683 989,272 988,959 67
DP_10S 986,592 984,601 990,218 978,726 984,975 976,610 987,019 990,247 206
DP_10SI 983,070 983,826 990,153 977,548 983,053 975,290 986,325 988,584 218
DP_5S 995,319 988,396 992,824 982,270 987,963 981,406 992,807 993,902 7
DP_5SI 991,801 985,360 990,794 982,112 982,893 979,731 988,870 990,376 19
SA_EG_1 982,298 973,179 985,364 974,994 975,498 968,323 977,410 985,041 6470
SA_EG_2 984,013 983,550 988,465 980,045 982,191 973,199 985,270 989,520 3867
ACO 1,017,741 1,016,567 1,021,075 1,007,713 1,010,822 1,007,210 1,013,315 1,019,092 2407
SA (Robust) 1,059,100 1,022,447 1,068,402 1,054,997 1,051,395 1,057,543 1,037,066 1,040,450 NA
SABFO 982,087 979,095 982,914 974,144 979,376 970,247 983,527 984,664 306.05
Best solution 982,087 973,179 982,914 974,144 975,498 968,323 977,410 984,664
% Deviation of SABFO value 0 0.60 0 0 0.40 0.20 0.63 0
from best solution

Table 9
Results for problems with N = 30 and T = 5.

Description Problem 33 Problem 34 Problem 35 Problem 36 Problem 37 Problem 38 Problem 39 Problem 40 Mean CPU Time (in second)

CVGA 632,737 647,585 642,295 634,626 639,693 637,620 640,482 635,776 NA


NGLA 611,794 611,873 611,664 611,766 604,564 606,010 607,134 620,183 NA
GADP (R) 603,339 589,834 592,475 586,064 580,624 587,797 588,347 590,451 NA
GADP (U) 578,689 572,232 578,527 572,057 559,777 566,792 567,873 575,720 NA
DP_10L 581,805 574,657 581,030 571,730 561,079 567,202 572,262 575,445 1324
DP_10LI 579,741 570,915 581,030 569,874 561,079 567,154 568,196 575,445 1499
DP_5L 583,082 576,592 581,691 575,024 561,424 570,435 573,878 576,091 222
DP_5LI 581,942 571,563 580,549 574,070 561,424 570,435 571,254 576,091 397
DP_10S 581,805 575,004 581,170 571,749 561,078 568,554 572,706 575,273 131
DP_10SI 579,741 570,906 577,402 569,596 561,078 568,554 571,580 575,273 306
DP_5S 582,858 576,106 581,262 574,110 562,857 570,356 572,797 576,149 23
DP_5SI 581,369 572,511 580,186 573,001 562,857 570,356 569,145 576,149 182
SA_EG_1 583,081 573,965 580,102 572,139 563,503 574,805 573,361 581,614 21,710
SA_EG_2 583,227 574,116 577,787 573,446 565,735 570,905 571,499 581,966 10,691
ACO 604,408 604,370 603,867 596,901 591,988 599,862 600,670 610,474 2061
SA (Robust) 579,704 576,350 586,831 584,318 570,492 572,782 571,703 596,835 NA
SABFO 578,415 570,630 577,390 568,289 558,345 572,536 569,993 577,873 758
Best solution 578,415 570,630 577,390 568,289 558,345 566,792 567,873 575,273
% Deviation of SABFO value 0 0 0 0 0 1.01 0.37 0.45
from best solution
B. Turanoğlu, G. Akkaya / Expert Systems With Applications 98 (2018) 93–104 103

Table 10
Results for problems with N = 30 and T = 10.

Description Problem 41 Problem 42 Problem 43 Problem 44 Problem 45 Problem 46 Problem 47 Problem 48 Mean CPU Time (in second)

CVGA 1,362,513 1,379,640 1,365,024 1,367,130 1,356,860 1,372,513 1,382,799 1,383,610 NA


NGLA 1,228,411 1,231,978 1,231,829 1,227,413 1,215,256 1,221,356 1,212,273 1,245,423 NA
GADP (R) 1,194,084 1,199,001 1,197,253 1,184,422 1,179,673 1,178,091 1,186,145 1,208,436 NA
GADP (U) 1,169,747 1,168,878 1,166,366 1,154,192 1,133,561 1,145,0 0 0 1,145,927 1,168,657 NA
DP_10L 1,174,773 1,175,323 1,174,023 1,155,879 1,128,136 1,144,030 1,143,814 1,168,142 7008
DP_10LI 1,171,853 1,169,138 1,174,023 1,152,684 1,128,136 1,143,824 1,142,494 1,167,900 7358
DP_5L 1,180,120 1,179,022 1,175,920 1,157,918 1,131,518 1,147,517 1,147,016 1,170,929 595
DP_5LI 1,171,413 1,174,421 1,170,019 1,156,016 1,131,518 1,147,517 1,145,934 1,170,929 945
DP_10S 1,172,434 1,175,551 1,175,240 1,155,998 1,129,143 1,144,539 1,143,788 1,167,163 1477
DP_10SI 1,171,178 1,170,747 1,165,525 1,153,981 1,128,784 1,144,092 1143,183 1,167,163 1827
DP_5S 1,181,743 1,177,212 1,176,997 1,158,507 1,132,926 1,149,893 1,147,041 1,171,658 63
DP_5SI 1,180,087 1,170,810 1,173,529 1,156,517 1,132,926 1,149,893 1,146,987 1,171,428 413
SA_EG_1 1,175,756 1,173,015 1,166,295 1,154,196 1,141,738 1,158,322 1,157,505 1179,888 87,200
SA_EG_2 1,174,815 1,177,743 1,171,932 1,154,945 1,140,116 1,158,227 1,163,761 1,177,565 46,152
ACO 1,223,124 1,231,151 1,230,520 1,200,613 1,210,892 1,239,255 1,248,309 1,231,408 8663
SA (Robust) 1,172,691 1,182,286 1,188,620 1,198,487 1,198,674 1,202,033 1210,573 1,209,088 NA
SABFO 1,168,453 1,170,042 1,160,204 1,149,944 1,132,136 1,144,677 1,160,830 1,172,857 3806.24
Best solution 1,168,453 1,168,878 1,160,204 1,149,944 1,128,136 1,143,824 1,142,494 1,167,163
% Deviation of SABFO value 0 0.1 0 0 0.35 0.07 1.6 0.49
from best solution

References Mazinani, M., Abedzadeh, M., & Mohebali, N. (2013). Dynamic facility layout prob-
lem basedon flexible bay structure and solving by genetic algorithm. Interna-
Balakrishnan, J., Jacobs, F. R., & Venkataramanan, M. A. (1992). Solutions for the tional Journal of Advanced Manufacturing Technology, 65, 929–943.
constrained dynamic facility layout problem. European Journal of Operational Re- McKendall, A. R., Jr., Shang, J., & Kuppusamy, S. (2006). Simulated annealing heuris-
search, 15, 280–286. tics for the dynamic facility layout problem. Computers & Operations Research,
Balakrishnan, J., & Cheng, C. H. (20 0 0). Genetic search and the dynamic layout prob- 33, 2431–2444.
lem. Computers and Operations Research, 27(6), 587–593. McKendall, A. R., Jr., & Shang, J. (2006). Hybrid ant systems for the dynamic facility
Balakrishnan, J., Cheng, C. H., Conway, D. G., & Lau, C. M. (2003). A hybrid genetic layout problem. Computers & Operations Research, 33, 790–803.
algorithm for the dynamic plant layout problem. International Journal of Produc- Metropolis, N., Rosenbluth, A. W., Rosenbluth, M. N., Teller, A. H., & Teller, E. (1958).
tion Research, 86, 107–120. Equations of state calculations by fast computing machines. J. Chem. Phys., 21,
Baykasoğlu, A., & Gindy, N. N. Z. (2001). A simulated annealing algortihm for the 1087–1092.
dynamic layout problem. Computers and Opeartions Research, 28, 1403–1426. Majhi, R., Panda, G., Majhi, B., & Sahoo, G. (2009). Efficient prediction of stock mar-
Baykasoğlu, A., Dereli, T., & Sabuncu, İ. (2006). An ant colony algorithm for solving ket indices using adaptive bacterial foraging optimization (ABFO) and BFO based
budget constrained and unconstrained dynamic facility layout problems. Omega, techniques. Expert Systems with Applications, 36, 10097–10104.
34, 385–396. Mandal, P., Sarkar, B. K., Saha, R., Chatterjee, A., Mookherjee, S., & Sanyal, D. (2015).
Bermejo, E., Cordon, O., Damas, S., & Santamaria, J. (2015). A comparative study on Real-time fuzzy-feedforward controller design by bacterial foraging optimiza-
the application of advanced bacterial foraging models to image registration. In- tion for an electrohydraulic system. Engineering Applications of Artificial Intelli-
formation Sciences, 295, 160–181. gence, 45, 168–179.
Chatzis, S. P., & Koukas, S. (2011). Numerical optimization using synergetic Mishra, S. (2005). A hybrid least square-fuzzy bacteria foraging strategy for har-
swarms of foraging bacterial populations. Expert Systems with Applications, 38, monic estimation. IEEE Trans. Evol. Comput., 9(1), 61–73.
15332–15343. Mohammadi, M. (2015). Bacterial foraging optimization and adaptive version for
Chen, H., Zhu, Y., Hu, K., & Ma, L. (2014). Bacterial colony foraging algorithm: Com- economicallyoptimum sitting, sizing and harmonic tuning orders setting of
bining chemotaxis, cell-to-cell communication, and self-adaptive strategy. Infor- LCharmonic passive power filters in radial distribution systems withlinear and
mation Sciences, 273, 73–100. nonlinear loads. Applied Soft Computing, 29, 345–356.
Chen, Y., Jiang, Y., Wahab, M. I. M., & Long, X. (2015). The facility layout problem in Moslemipour, G., Lee, T. S., & Rilling, D. (2012). A review of intelligent approaches
non-rectangular logistics parks with split lines. Expert Systems with Applications, for design-ing dynamic and robust layouts in flexible manufacturing systems.
42, 7768–7780. International Journal of Advanced Manufacturing Technology, 60, 11–27.
Conway, D. G., & Venkataramanan, M. A. (1994). Genetic search and the dynamic Panda, R., & Naik, M. K. (2015). A novel adaptive crossover bacterial foraging opti-
facility layout problem. Computers and Operations Research, 21(8), 955–960. mization algorithm for linear discriminant analysis based face recognition. Ap-
Devi, S., & Geethanjali, M. (2014). Application of modified bacterial foraging opti- plied Soft Computing, 30, 722–736.
mization algorithm for optimal placement and sizing of distributed generation. Pandit, N., Tripathi, A., Tapaswi, S., & Pandit, M. (2012). An improved bacterial for-
Expert Systems with Applications, 41, 2772–2781. aging algorithm for combined static/dynamic environmental economic dispatch.
Erel, E., Ghosh, J. B., & Simon, J. T. (2003). New heuristic for the layout problem. Applied Soft Computing, 12, 3500–3513.
Journal of the Operational Research Society, 54, 1275–1282. Passino, K. M. (2002). Biomimicry of bacterial foraging for distributed optimization
García-Hernández, L., Palomo-Romero, J. M., Salas-Morera, L., & Arauzo-A- and control. IEEE Control Systems Magazine, 22(3), 52–67.
zofra, A. (2015). A novel hybrid evolutionary approach for capturing decision Pillai, V. M., Hunagund, I. B., & Krishnan, K. K. (2011). Design of robust layout for Dy-
maker knowledge into the unequal area facility layout problem. Expert Systems namic Plant Layout Problems. Computers & Industrial Engineering, 61, 813–823.
with Applications, 42, 4697–4708. Pincus, M. (1970). A Monte Carlo method for the approximate solution of certain
Guo, P., Cheng, W., & Wang, Y. (2017). Hybrid evolutionary algorithm with extreme types of constrained optimization problems. Operation Research, 18, 1225–1228.
machine learning fitness function evaluation for two-stage capacitated facility Palomo-Romero, J. M., Salas-Morera, L., & García-Hernández, L. (2017). An island
location problems. Expert Systems with Applications, 71, 57–68. model genetic algorithm for unequal area facility layout problems. Expert Sys-
Guzman, M. A., Delgado, A., & Carvalho, J. D. (2010). A novel multiobjective opti- tems with Applications, 68, 151–162.
mization algorithm based on bacterial chemotaxis. Engineering Applications of Pourvaziri, H., & Naderi, B. (2014). A hybrid multi-population genetic algorithm for
Artificial Intelligence, 23, 292–301. the dynamic facility layout problem. Applied Soft Computing, 24, 457–469.
Hezer, S., & Kara, Y. (2013). Eşzamanlı Dağıtımlı Ve Toplamalı Araç Rotalama Prob- Pourvaziri, H., & Pierreval, H. (2017). Dynamic facility layout problem based on open
lemlerinin Çözümü İçin Bakteriyel Besin Arama Optimizasyonu Tabanlı Bir Al- queuing network theory. European Journal of Operational Research, 259, 538–553.
goritma. Journal of the Faculty of Engineering and Architecture of Gazi University, Rosenblatt, M. J. (1986). The Dynamics of plant layout. Management Science, 32(1),
28(2), 373–382. 76–86.
Kang, S., & Chae, J. (2017). Harmony search for the layout design of an unequal area Ross, P. J. (1988). Taguchi techniques for quality engineering. The United States of
facility. Expert Systems with Applications, 79, 269–281. America: McGraw-Hill, Inc. (Chapter 8).
Kim, D. H., Abraham, A., & Cho, J. H. (2007). A hybrid genetic algorithm and bacterial Sanyal, N., Chatterjee, A., & Munshi, S. (2011). An adaptive bacterial foraging algo-
foraging approach for global optimization. Information Sciences, 177, 3918–3937. rithm for fuzzy entropy based image segmentation. Expert Systems with Applica-
Kirkpatrick, S., Gerlatt, C. D., Jr., & Vecchi, M. P. (1983). Optimization by simulated tions, 38, 15489–15498.
annealing. Science, 220, 671–680. Sathya, P. D., & Kayalvizhi, R. (2011a). Modified bacterial foraging algorithm based
Lacksonen, T. A., & Enscore, E. E., Jr. (1993). Quadratic assignment algorithms for the multi level thresholding for image segmentation. Engineering Applications of Ar-
dynamic layout. International Journal of Production Research, 31(3), 503–517. tificial Intelligence, 24, 595–615.
104 B. Turanoğlu, G. Akkaya / Expert Systems With Applications 98 (2018) 93–104

Sathya, P. D., & Kayalvizhi, R. (2011b). Optimal multilevel thresholding using bacte- Ulutas, B. H., & Islier, A. A. (2009). A clonal selection algorithm for dynamic facility
rial foraging algorithm. Expert Systems with Applications, 38, 15549–15564. layout problems. Journal of Manufacturing Systems, 28, 123–131.
Sengupta, A., & Bhadauria, S. (2015). Bacterial foraging driven exploration of multi Ulutas, B., & Islier, A. A. (2015). Dynamic facility layout problem in footwear indus-
cycle fault tolerant datapath based on power-performance tradeoff in high level try. Journal of Manufacturing Systems, 36, 55–61.
synthesis. Expert Systems with Applications, 42, 4719–4732. Urban, T. L. (1992). Computational performance and efficiency of lower procedures
Supriyono, H., & Tokhi, M. O. (2012). Parametric modelling approach using bacterial fort he dynamic facility layout problem. European Journal of Operational Re-
foraging algorithms for modelling of flexible manipulator systems. Engineering search, 57(2), 271–279.
Applications of Artificial Intelligence, 25, 898–916. Venkaiah, C., & Kumar, D. M. V. (2011). Fuzzy adaptive bacterial foraging conges-
Şahin, R. (2008). Dinamik tesis düzenleme problemi için bir tavlama benzetimi tion management using sensitivity based optimal active power re-scheduling of
sezgiseli. Journal of the Faculty of Engineering and Architecture of Gazi University, generators. Applied Soft Computing, 11, 4921–4930.
23(4), 863–870. Xiao, Y., Xie, Y., Kulturel-Konak, S., & Konak, A. (2017). A problem evolution algo-
Şahin, R., & Türkbey, O. (2009). A new hybrid tabu-simulated annealing heuristic for rithm with linear programming for the dynamic facility layout problem—A gen-
the dynamic facility layout problem. International Journal of Production Research, eral layout formulation. Computers and Operations Research, 88, 187–207.
47(24), 6855–6873. Zhao, W., & Wang, L. (2016). An effective bacterial foraging optimizer for globalop-
timization. Information Sciences, 329, 719–735.

You might also like