ICDeM 2012, 15 18 March, Kedah, Malaysia.

Solving Fixed Charge Transportation Problems using Artificial

Immune System

Khalid M. Altassan
King Saud University, CBA, Saudi Arabia, kaltassan@ksu.edu.sa

Mahmoud M. El-Sherbiny*
OR Dept. ISSR Cairo University, Egypt
Currently, King Saud University, CBA, Saudi Arabia, msherbiny@ksu.edu.sa

Ahmed D. Abid
King Saud University, CBA, Saudi Arabia, darwesh@ksu.edu.sa

Fixed-charge transportation problem (FCTP) is considered to be an NP-hard problem. Several genetic algorithms
based on spanning tree and Prfer number were presented. Most of such methods do not guarantee the feasibility of
all the generated chromosomes and need a repairing procedure for feasibility. Contrary to the findings in previous
works, this paper introduces an Artificial Immune System for solving Fixed Charge Transportation Problems
(AISFCTP). AISFCTP solves both balanced and unbalanced FCTP without introducing a dummy supplier or a
dummy customer. In AISFCTP a coding schema is designed and algorithms are developed for decoding such schema
and allocating the transported units. These are used instead of spanning tree and Prfer number. Therefore, a
repairing procedure for feasibility is not needed, i.e. all the generated antibodies are feasible. Besides, some mutation
functions are developed and used in AISFCTP. Due to the significant role of mutation function on the AISFCTPs
quality, its performances are compared to select the best one. For this purpose, various problem sizes are generated at
random and then a robust calibration is applied using the relative percentage deviation (RPD) method. In addition,
two problems with different sizes are solved to evaluate the performance of the AISFCTP and to compare its
performance with most recent methods.
Keywords: Fixed charge transportation, Convergence, Genetic algorithm, Artificial immune.

1. Introduction 1998), and genetic algorithms based on a matrix

The fixed charge transportation problem (FCTP) is permutation representation (Gottlieb et al., 2001),
considered to be an NP-hard problem (Murty, (Raidl and Julstrom, 2003), have improved the
1968). Usually, a FCTP is formulated and solved effective coding of the spanning tree method
as a mixed integer network programming problem. based on edge sets. The genetic algorithm creates
Theoretically, the FCTP can be solved by any a sorted set of edges to encode the spanning tree,
mixed integer programming; however, these which is efficient compared with evolution
methods are not employed because of their strategies (ES) at a certain level (Su and Zhan,
inefficient and expensive computation. Generally, 2006). Moreover, to improve solution quality,
solving methods can be classified as exact or Hajiaghaei (2010) addressed a nonlinear FCTP
heuristic methods. Exact methods for solving the using a spanning tree based genetic algorithm.
FCTP include the cutting planes method Nevertheless, the quality of solutions attained
(Rousseau, 1973), the vertex ranking method largely depends on the randomness.
(McKeown, 1975), and the branch-and-bound On the other hand, Some special AIS are
method (Palekar et al., 1990), amongst others. developed to solve complex optimization
Exact methods are, however, generally not very problems. One of them is aiNet (Timmis et. al,
useful when a problem reaches a certain level, 2004; de Castro and Timmis, 2002) that inspired
because they do not make the most use of the by biological immune system. Opt-aiNet (Timmis
special network structure of the FCTP. Therefore, et. al, 2004) is an application of aiNet in function
heuristic methods have been proposed, such as the optimization. Opt-aiNet considers the optimized
adjacent extreme point search method (Balinski, objective function as antigen, and the candidate
1961; Sun et al., 1993), and the Lagrangian solutions as antibodies. The candidate antibodies
relaxation method (Wright et al., 1989, 1991). evolve according to the matching degree between
Although these methods are usually antibodies and antigen that is fitness. The better
computationally efficient, the major disadvantage the matching between them, the less the mutation
of heuristic methods is the possibility of degree of candidate antibody, and vice verse. AIS
terminating at a local optimum that is far distant has used in many applications such as job shop
from the global optimum. Recently, some meta- scheduling problems (Orhan at el, 2004) and
heuristic methods employed in the FCTP, such as (Tsai, et al , 2007).
the Tabu search method for FCTPs (Sun et al., To improve the solution quality of the FCTP, this
1998), genetic algorithms (GA) based on a paper aims to introduce an artificial Immune
spanning tree with Prfer numbers (Gen et al., system for solving Fixed Charge Transportation
 ICDeM 2012, 15 18 March, Kedah, Malaysia.

problem (AISFCTP) and study the effect of its tailored and implemented to fit the case at hand.
factors on the performance. In addition to that two For the FCTP, the problem of interest in this
problems with different sizes have been solved to research, we develop the following:
evaluate the performance of the AISFCTP and to
compare its performance with the recent methods. Start
The rest of the paper is organized as follows:
in section 2, FCTP is described. In section 3, the
g=1
proposed AISFCTP is described, and in section 4
the Parametric Analysis is carried out. Numerical Create initial population of l antibodies Ai with
experiments with proposed AISFCTP are algorithm1
presented in section 5. Finally, the conclusion and
future work are reported in section 6. i=1

2. Description of FCTP Clone ith Antibody Ai in the population CN times

Fixed Charge Transportation Problem (FCTP) can
be described as a distribution problem in which Mutate each of the CN clones
there are m suppliers (warehouses or factories) and
n customers (destinations or demand points). Each
Evaluate each of the CN clones
of the m suppliers can ship to any of the n
customers at a shipping cost per unit cij (unit cost Apply decoding Algorithm 2
for shipping from supplier i to customer j) plus a
fixed cost fij, assumed for opening this route. Each
Apply allocation Algorithm 3
supplier i =1, 2, . . ., m has si units of supply and
each customer j =1, 2, . . ., n demands dj units.
The objective is to determine which routes are to Calculate the fitness of each antibody Ai
be opened and the size of the shipment, so that the
total cost of meeting demand, given the supply Get the mutated clone with the best fitness BF
constraints, is minimized. The stander
mathematical model of FCTP can be represented
BF
as follows: BF fitness better Yes
replaces
m n than the fitness of Ai
Min z (cij xij fij yij ) (1) A ?
No i
i1 j1
i = i +1 :Repeat until i > n
m
s.t. x
i1
ij bj for j 1,...,n (2) Calculate the affinity between each two antibodies in the
population
n

x
j1
ij ai for i 1,...,m (3) Select the antibodies for the new mutation based on the
affinity

i, j xij 0 Create new antibodies to substitute the removed

antibodies
yij 0 if xij 0
g = g + 1 :Repeat until g > number of iterations
yij 1 if xij 0
End
Where xij is the unknown quantity to be
transported on the route (i, j) that from plant i to Figure 1. Main architecture of AISFCTP
customer j. The transportation cost for shipping
per unit from plant i to customer j is cij xij.
3.1 Coding Schema and initialization
One of the most important issues when designing
3. The Proposed Algorithm the AIS lies on its solution (antibody)
In this paper, a typical immune algorithm structure
representation. In order to construct a direct
is utilized. The algorithm preserves the essential
relationship between the problem domain and the
principles of natural immune systems including
AISFCTP, the proposed coding scheme (antibody
the cloning, mutation, and clonal selection. Figure
structure) consists of the set of all the integer
1 illustrates the flowchart of the main steps of the
numbers in the interval [1, m+n] with any
proposed method. The details of these steps are
sequence and without any repetition; where the
presented in the following subsections.
length of scheme is equal to m + n, where m is the
The implementation of the immune algorithm
number of suppliers and n is the number of
is often different for each problem handled. That
customers. Therefore, the length of each antibody
is, the representation and hence the creation of the
Ai equal to the sum of the problem dimensions and
solutions, the mutation, and the affinity should be
 ICDeM 2012, 15 18 March, Kedah, Malaysia.

the suppliers numbers represented by the integer Input:

numbers from 0 to m and the demands integer Ai 8 3 9 5 4 7 2 1 6
numbers from m+1 to m + n. Figure 2 depicts a i=1
sample antibody example, which is used to code a
Output:
4x5 FCTP such as problem in Table 1. As shown
D 4 5 1 3 2 The customers' order
in Figure 2, the cell values are between 1 and 4+5.
We can realize that any number can not be
repeated. S 3 4 2 1 The Suppliers' order
Figure 3. Illustrative example of decoding
8 3 9 5 4 7 2 1 6 algorithm
Figure 2. An example of proposed antibody
structure 3.3 Allocating algorithm
The allocating algorithm allocates the transported
The population is initialized randomly by calling units xij based on the order coming from decoding
algorithm1 l times to create l antibodies Ai where algorithm 2. In other words, this algorithm find a
l represents the population size. In this algorithm, feasible solution for FCTP based on the outputs of
the Rand(1, m+ n) is a function that returns a the decoding algorithm. This algorithm guarantees
random integer number in the interval[1, m + n], the validity of both the first (2) and the second (3)
Mod(x, y) is a function that returns the reminder of constraints. Also, this algorithm can be used to
x when it is divided by y and Q. Remove(k) is a solve both balanced and unbalanced transportation
function that eliminate kth element of queue Q. problems without introducing a dummy supplier
The Pseudocode of algorithm1 for creating or a dummy customer. The Pseudocode for the
individual antibody is presented as follows: allocating algorithm is presented as follows:
1. Create a collection list Q ={1, 2,, m +n }. 1. Set i equal to the first value in suppliers order
2. Set j = 1. S and set j equal to the first value in
3. Generate an integer number between 1 and m customers order D. i.e. i = S(1) and j =
+n and set it to variable c. Take the cell Ai (j) D(1).
4. Set k = Mod(c, Length(Q)); where Mod(c, 2. If ai = bj then { set xij equal to ai , remove
Length(Q)) is a function that returns the S(1) , and remove D(1)}
reminder of c when it is divided by length(Q). 3. If ai > bj then { set xij equal to bj , set ai
5. Add Q[k] to the antibody Ai in the position j. equal to ai - bj , and remove D(1)}
6. Remove the item k from the list Q 4. If ai < bj then { set xij equal to ai , set bj equal
7. j = j + 1. to bj - ai , and remove S(1)}
8. Repeat from step 3 to step 7 until j > n + m . 5. Repeat from step 1 to step 4 until (length of
9. Return the antibody Ai , where i =1,, l and l queue S = 0 or length of queue D = 0).
is the population size 6. Return xij i =1, 2,, m and j =1, 2,, n.
3.2 Decoding algorithm
The inputs of allocating algorithm are the
Decoding algorithm is used to decode the antibody
sequence of suppliers S and the sequence of
Ai into suppliers order S and customers order D. customers D (the output of decoding algorithm).
The inputs of this algorithm are the generated Based on these sequences the allocating algorithm
antibody Ai , the number of suppliers m ,and the allocates units Xij (feasible solution) of FCTP.
number of customers n while the results are the
sequence of suppliers S and the sequence of 3.4 Evaluating the Solutions
customers D. Figure 3. exhibits the results of Each antibody is evaluated to determine its fitness
applying Algorithm 1 on the antibody presented in as follows. As mentioned above each antibody is
Figure 2. The Pseudocode of the decoding decoded the result used as an input for allocation
algorithm is presented as follows: algorithm. The solution resulted from allocation
algorithm is evaluated using objective function
1. Set j = 1. (1). The value of objective (1) is assigned to the
2. Take the cell Ai (j) antibody as its fitness.
3. If Ai (j) n then add Ai (j) to the supplier
order S. 3.5 Cloning and Mutation
4. If Ai (j) > n then add Ai (j) to the customer Each antibody is cloned (copied) number of times,
order D. determined by the number of Cloning Number
5. j = j + 1. (CN). The clones are then mutated to get new
6. Repeat from step 2 to step 5 until j > n + m . antibodies that are different from their parent. In
7. Return the supplier order S and the customer the proposed AISFCTP, six different Mutation
order D. Functions (MFs) are developed and tested.
The first mutation method is a uniform random
number located in the range of 10% to 30% of the
sum of problem dimensions. The number of swaps
(NS) for this mutation is represented by Eq. (4),
 ICDeM 2012, 15 18 March, Kedah, Malaysia.

where r is defined by a random number in the the original ones when Rand is close to zero. The
interval [0.1, 0.3]. closer the R to one is, the closer the number of
swaps to the max swaps no is. These two mutation
NS Rand (1, r ( m n )) (4) functions are suggested to allow the search to
escape from local optima by occasionally
The remaining five MFs are functions of two increasing the number of swaps. The number of
parameters. The first parameter is the non-uniform swaps for these two mutations are adapted with
factor based on which the number of swaps is Eq. (10) and Eq. (11), respectively.
determined. The second parameter is the degree of (1 NF ) u
non-uniformity (u). All the functions are designed NS MSN R (10)
to be directly related with u.
(T u )
The second MF is based on the fitness of the NS MSN R (11)
solution. As the FCTP is a minimization problem,
3.6 Affinity Function
the function is designed to be directly related with
the Normalized Fitness (NF) of the solution. That The selection of the antibodies from one
is, solutions with normalized fitness closer to one, generation to the next one depends on some
i.e. relatively bad solutions, will be subject to measurement of the affinity (similarity) among all
more number of swaps. This actually gives the the antibodies of the current generation. The
chance for low affinity solutions to mutate more in calculations of the affinity AF between each two
order to improve their affinities. The NS for this antibodies are applied to prevent similar solutions
mutation function is adapted with Eq. (5) and the with high evaluation from being copied to the next
normalized fitness of each antibody is calculated generation and hence dominating the search. This
using Eq.(6). is technically applied to reduce the chance of a
premature convergence to local optima.
(1 (1 NF ) u ) The technique used to check the similarity
NS MSN (5)
between every two antibodies in a population
LowestFitn ess - Fitness
NF (6) counts the number of similar variables in the two
LowestFitn ess - HighestFitness solutions. The affinity function of two antibodies
The third MF is designed to be inversely related Aj and Ak is represented as in Eq. (12).
with the ratio (T) of the current iteration number
AF(Aj, Ak)yi
(CIN) and the total number of iterations (TNI). i
That is, the more the search goes, the less the th th (12)
number of swaps is. This is really intuitive as in 1 if thei gene
ofAj thei gene
ofAk
yi =
where
contrast to the first stages of the search where a 0 Otherwise
real exploration of the search space through The basic idea is that the more the number of
significant changes in the solutions are required, at similar variables in the two antibodies is, the
the last stages of the search fine tuning with little higher the similarity between them. Based on a
changes of the supposed-to-be near-optimal specific parameter, the algorithm eliminates those
solutions is more reasonable. The number of solutions that have AF more than a specific
swaps (NS) for this mutation is represented in Eq. parameter -Number of Similarities (NS).
(7) where u is the degree of non-uniformity.
(1 (T u ) CIN 4. Parametric analysis
NS MNS , where T (7)
TNI In this section we try to discover the best MF from
The fourth MF is based on both the time and the the implemented six. Because the scale of the
normalized fitness of the solution. It basically uses objective functions in each problem is different,
the average of these two factors to decide the they could not be used directly. Therefore, the
number of swaps. Basically, the function is Relative Percentage Deviation (RPD) is used for
designed to be directly related with the fitness but each combinatioin (Taguchi, 1986). RPD is
inversely related with the time. The average of calculated by using Eq. (13).
Time and normalized Fitness (TF) is calculated as A lg sol Min sol
represented in Eq. (8) and the number of swaps for RPD 100 (13)
this mutation is adapted with (9). Min sol
Where Algsol and Minsol are the obtained
TF 1 ( NF (1 T ) (8)
2 objective values for each replication of trial in a
given combination and the obtained best solution,
(1 (1TF ) u ) respectively. After converting the objective values
NS MNS (9)
In the fifth and the sixth MFs, we include a to RPDs, the mean RPD is calculated for each
random factor (R) so that the number of swaps is trial. Five problems with different size are
based on the non-uniform factor, time and fitness generated and used to discover the best mutation
respectively, but with some randomization. The function from the implemented six. The
random factor R takes values between zero and characteristics of these problems are presented in
one. The functions behave almost the same way as Table 1.
 ICDeM 2012, 15 18 March, Kedah, Malaysia.

Table 1: Characteristics of FCT test problems (El-Sherbiny and Alhamali, 2011), (Altassan at el.,
2012) are solved, comparing with the solution
Problem Total Rang of Rang of fixed presented by them. The sizes of the problems are
size supply variable costs costs
45 and 510, respectively. The variable costs,
Lower Upper Lower Upper
Limit Limit Limit Limit and the fixed costs for first problem is given in
1418 310 1 10 90 250 Table 3. Also, the variable costs, and the fixed
510 1,500 1 8 100 400 costs for second problem is given in Table 4. The
1010 3,000 1 8 150 400 parameters used for the proposed method in these
1020 4,000 1 10 100 500 problems are optimally tuned parameters and
3030 7,000 1 10 10 100
operators from experimental results.
As illustrated in Table 2, the quality of the Table 3: Unit variable cost in 4 x 5 problem
results of using the first mutation function is very
close to the third one and both are superior to the Plants Costumers
others. But the third mutation method is most Shipping costs cij Fixed costs fij
superior. Therefore, in the next section, the third
1 2 3 4 5 1 2 3 4 5
mutation function will be used in our comparison
with the most recent algorithms in the literatures. 1 8 4 3 5 8 60 88 95 76 97
2 3 6 4 8 5 51 72 65 87 76
3 8 4 5 3 4 67 89 99 89 100
Table 2: The comparitive results of the RPD for
4 4 6 8 3 3 86 84 70 92 88
the MF
RPD of the test problems Mean
MF Concerning the first problem, the supplies and
14x18 5x10 10x10 10x20 30x30 RPD demands from each plant 1 to 4 for the each
1 0.2% 2.6% 0.3% 1.5% 0.3% 0.7% customer 1 to 5 are as follows:
2 0.3% 4.6% 1.6% 3.0% 2.9% 2.5%
b1 = 88, b2 = 57, b3 = 24, b4 = 73, b5 = 33, a1 =
3 0.2% 0.0% 0.0% 0.0% 0.0% 0.0%
4 0.1% 5.4% 1.4% 1.9% 1.2% 1.6% 57, a2 = 93, a3 = 50, a4 = 75. The obtained local
5 1.0% 8.3% 2.7% 6.1% 3.0% 3.9% optimal solution from our AISFCTP is the same as
6 0.0% 2.7% 0.1% 2.3% 0.8% 1.1% the solution found by (Khalid at el. 2012), (El-
Sherbiny and Alhamali, 2011), (Hajiaghaei et al.,
5. Numerical experiments 2010) and equal to 1484. The transportation
To evaluate the performance of the proposed allocation matrix for this solution is shown in
AISFCTP two problems with different sizes, Table 5.
previously addressed by (Hajiaghaei et al., 2010),
.
Table 4: Unit variable cost in 5 x 10 problem

Plants Costumers
Shipping costs cij Fixed costs fij
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
1 8 4 3 5 2 1 3 5 2 6 160 488 295 376 297 360 199 292 481 162
2 3 3 4 8 5 3 5 1 4 5 451 172 265 487 176 260 280 300 354 201
3 7 4 5 3 4 2 4 3 7 3 167 250 499 189 340 216 177 495 170 414
4 1 2 8 1 3 1 4 6 8 2 386 184 370 292 188 206 340 205 465 273
5 4 5 6 3 3 4 2 1 2 1 156 244 460 382 270 180 235 355 276 190

Table 5: Transportation allocation matrix of local The obtained local optimal solution for this
optimal solution 1484 for 4 x 5 problem problem found by the proposed algorithm is 6240
as illustrated in Table 6, while the solution found
D1 D2 D3 D4 D5 by (Altassan at el. 2012) is 6255 as illustrated in
S1 57 24 Table 7, the solution found by (El-Sherbiny and
S2 69 Alhamali, 2011) is 6296 as illustrated in Table 8,
S3 50 and the solution found by (Hajiaghaei et al., 2010)
S4 19 23 33 is 6305 as illustrated in Table 7. Therefore, we can
conclude that the proposed AISFCTP is superior
Concerning the second problem, the supplies and
than the other methods proposed by (Altassan at
demands from the each plant 1 to 5 for each
el. 2012), (El-Sherbiny and Alhamali, 2011), and
customer 110 are as follows: (Hajiaghaei et al., 2010).
b1 = 225, b2 = 150, b3 = 90, b4 = 215, b5 = 130, b6
= 88, b7 = 57, b8 = 124, b9 = 273, b10 = 133, and a1
= 157, a2 = 293, a3 = 150, a4 = 575, a5 = 310.
 ICDeM 2012, 15 18 March, Kedah, Malaysia.

Table 9. Transportation allocation matrix found hybrid particle swarm method presented by (El-
by the AISFCTP for 5 x 10 problem Sherbiny and Alhamali, 2011), and the GA
D1 D2 D3 D4 D5 D6 D7 D8 D9 D10 presented by (Hajiaghaei et al., 2010) showed that
the proposed algorithm (AISFCTP) is superior to
S1 130 27
S2 79 90 124 the others. The performance of AISFCTP and the
S3 88 30 32 solution quality prove that AISFCTP is highly
S4 225 71 215 64 competitive and can be considered as a viable
S5 241 69 alternative to solve FCTPs.
Future work includes further experimentation
Table 8. Transportation allocation matrix found with parameters of AISFCTP, testing the proposed
by the AIAFCTP for 5 x 10 problem AISFCTP on other real life problems, and
D1 D2 D3 D4 D5 D6 D7 D8 D9 D10 investigating using other metaheuristic techniques
S1 27 130 combined with the proposed decoding and
S2 106 63 124 allocation algorithms for solving problems.
S3 88 57 5
S4 225 44 215 91 References
S5 268 42 Altassan, Khalid M., El-Sherbiny, Mahmoud M.,
Source: (Altassan at el. 2012) Ahmed D. Abid (2012), A Novel Artificial
Immune Algorithm for Solving Fixed Charge
Table 7. Transportation allocation matrix found in Transportation Problems, submitted to 2012
(El-Sherbiny and Alhamali, 2011) for 5 x 10 problem International Conference on Innovation and
D1 D2 D3 D4 D5 D6 D7 D8 D9 D10 Information Management.
S1 130 27 Balinski, M.L. (1961). Fixed cost transportation
S2 15 90 124 64 problem. Naval Research Logistics Quarterly,
S3 88 30 32 8, 41-54.
S4 225 135 215 De Castro, L.N. and J. Timmis, (2002). An
S5 241 69 artificial immune network for multimodal
Source: (El-Sherbiny and Alhamali, 2011)
function optimization. in Proc. of IEEE
congress on evolutionary computation, Vol. 1,
Table 6: Transportation allocation matrix found by
699-674.
M. Hajiaghaei et al. (2010) for 4 x 5 problem
El-sherbiny, M. M. , Alhamali, R. M, (2011), A
D1 D2 D3 D4 D5 D6 D7 D8 D9 D10 Hybrid Particle Swarm Method with Artificial
S1 90 67 Immune Learning for Solving the Fixed
S2 150 124 19 Charge Transportation Problem, submitted to
S3 5 88 57
S4 225 210 63 77 Computers & Industrial Engineering.
S5 273 37 Gen, M., K. Ida, and Y.Z. Li, (1998). Bicriteria
Source: (Hajiaghaei et al., 2010) transportation problem by hybrid genetic
algorithm. Computers and Industrial
6. Conclusion Engineering, 35(12) 363-366.
This paper has proposed an artificial immune Gottlieb, J., B.A.Julstrom, F.Rothlauf, and Raidl,
algorithm for solving fixed charge transportation (2001). G.R.: Prfer numbers: A poor
problem (AISFCTP). AISFCTP solves both representation of spanning trees for
balanced and unbalanced FCTP without evolutionary search. In: Spector L, et al., eds.
introducing a dummy supplier or a dummy Proc. of the 2001 Genetic and Evolutionary
customer. Although AISFCTP with population- Computation Conf. San Francisco: Morgan
based search is characterized as an evolutionary- Kaufmann Publishers, 343-350.
like algorithm, the major contributions are the Hajiaghaei, M. Keshteli, S. Molla-Alizadeh-
coding schema and the decoding algorithm that Zavardehi, R. Tavakkoli-Moghaddam, (2010).
avoid infeasibility of any candidate solutions. All Addressing a nonlinear fixed-charge
the generated antibodies are feasible solutions for transportation problem using a spanning tree-
FCTP. Therefore, in AISFCTP the presented based genetic algorithm, Computers &
schema, decoding and allocation algorithms are Industrial Engineering, 59, 259271.
used instead of spanning tree used with genetic Kannan, G., Sasi Kumar, P., & Vinay, V. P.
algorithm. While using the spanning tree and (2008) Comments on Nonlinear fixed charge
Prfer number with the GA may result non- transportation problem by spanning tree-based
feasible solutions, the AISFCTP guarantees the genetic algorithm by Jung-Bok Jo, Yinzhen
feasibility of all the generated solutions. Besides, Li, Mitsuo Gen, Computers & Industrial
some mutation functions are developed and used Engineering (2007). Computers & Industrial
in AISFCTP. Due to the significant role of Engineering, 55, 533534.
mutation function on the AISFCTPs quality, its McKeown, P. G., (1975). A vertex ranking
performances are compared to select the best one. procedure for solving the linear fixed charge
Also, the comparative study of the AISFCTP with problem. Operations Research, 1183-1191.
the method proposed by (Altassan at el. 2012), the
 ICDeM 2012, 15 18 March, Kedah, Malaysia.

Murty, K.G., (1968). Solving the fixed charge charge transportation problem, European
problem by ranking the extreme points, Journal of Operational Research,106 :441-456.
Operations Research, 268-279 Taguchi, G. (1986). Introduction to quality
Orhan Engin, Alper Dyen, (2004). A new engineering. White Plains: Asian Productivity
approach to solve hybrid flow shop scheduling Organization/UNIPUB.
problems by artificial immune system, Future Timmis, J., T Knight, L. N. Catro and Hart,
Generation Computer Systems, 20,10831095. (2004). An overview of artificial immune
Palekar, U.S., M.K. Karwan, and S.Zionts, systems. Computation in Cells and Tissues:
(1990). A branch-and bound method for the Perspectives and Tools Thought. Natural
fixed charge transportation problem , Computation Series, Springer-Verlag, 51-86.
Management Science, 36(9): 1092-1105. Tsai, J. T., Ho, W. H., Liu, T. K., & Chou, J. H.
Raidl, B. A., Julstrom, G.R., (2003). Edge-sets: (2007). Improved immune algorithm for global
An effective evolutionary coding of spanning numerical optimization and job shop
trees. IEEE Transactions on Evolutionary scheduling problems. Applied Mathematics
Computation, 7(3), 225-239. and Computation, 194, 406424.
Rousseau, J. M., (1973). A cutting plane method Wright, D., C. Haehling von Lanzenauer (1989).
for the fixed cost problem, Doctoral Solving the fixed charge problem with
dissertation, Massachusetts Institute of lagrangian relaxation and cost allocation
Technology. Cambridge, MA. heuristics. European Journal of Operational
Su, S., Zhan, D.C., (2006). New genetic Research, 42, 304-312.
algorithm for the fixed charge transportation Wright, D., C. Haehling von Lanzenauer (1991).
problem , Proceedings of the 6th World COLE: A new heuristic approach for solving
Congress on Intelligent Control and the fixed charge problem Computational
Automation, 7039-7043. results. European Journal of Operational
Sun, M., McKeown, P.G., (1993). Tabu search Research, 52, 235-246.
applied to the general fixed charge problem Zalinda Othman, Mohammad-Reza Rostamian
,Annals of Operations Research, 41(14):405- Delavar, Sarah Behnam, Sina Lessanibahri,
420. (2011). Adaptive genetic algorithm for fixed-
Sun, M., Aronson, J.E. Mckeown, P.G.,(1998). A charge transportation problem, International
tabu search heuristic procedure for the fixed Multi Conference of Engineers & Computer
Sciences (IMECS) March, 16-18.