Sie sind auf Seite 1von 8

1

AN ANT-BASED ALGORITHM FOR THE CROSS DOCKING SCHEDULING


PROBLEM FOR DISTRIBUTION CENTERS

Kuancheng Huang*
Dept. of Transportation and Logistics Management, National Chiao Tung University
No. 1001, Ta Hsueh Road, Hsinchu City 30010, Taiwan
Phone: 886-3-5731928, Fax: 886-3-5720844, E-mail: kchuang@cc.nctu.edu.tw

Chi-Yu Hsieh
Dept. of Transportation Technology and Management, National Chiao Tung University
No. 1001, Ta Hsueh Road, Hsinchu City 30010, Taiwan

Yun-Cheng Chou
Dept. of Transportation Technology and Management, National Chiao Tung University
No. 1001, Ta Hsueh Road, Hsinchu City 30010, Taiwan


Abstract

With an aim to reduce inventory level, raise handling efficiency, and to improve customer service, more and
more logistics operators choose to adopt the cross-docking strategy. As no storage is involved at a distribution
center (DC), transportation coordination becomes critical for a successful implementation of cross docking.
The Cross-Docking Scheduling Problem (CDSP) determines the schedules of the inbound (supplier) and
outbound (customer) trucks for a DC with the objective of minimizing the total work time span. Owing to the
nature of the sequencing decision problem associated with the CDSP, this study designs the solution algorithm
based on Ant Colony Optimization (ACO), which has some inherited advantages over other meta-heuristics
due to its sequential framework for the searching process and the solution building procedure. An ant-based
algorithm is developed to generate the outbound sequence under the condition of one single truck. Given this
ant-generated initial sequence, a heuristic of several steps is used to make the scheduling decisions for both
the inbound and outbound vehicles under the multiple-truck situation. In the numerical experiment, the
developed algorithm is compared with a solution algorithm based on the classic J ohnsons Rule in the
literature. It is found that the developed ant-based algorithm is promising for providing a new approach to
solve the CDSP.

Keywords: Cross Docking, Distribution Center, Ant Colony Optimization, Heuristic



1. Introduction

In order to improve customer service and reduce operation cost, more and more logistics operators choose to
adopt the cross-docking strategy, for which only sorting and consolidating is involved at a distribution center
(DC), and storage and order-picking activities are minimized. The cost associated with inventory carrying and
material handling is significantly reduced. Meanwhile, customers can be served in a much more timely
fashion, and customer satisfaction is thus raised. The other important advantage of cross-docking operation is
the saving in transportation cost. As the inbound shipments from the suppliers and the outbound shipments to
the customers are both consolidated at a cross-docking DC, shipping volume is increased and the unit
transportation cost is reduced. The cross-docking operation at an DC is illustrated in Figure 1.

The benefit of cross-docking can never be realized if the shipping decisions of the inbound and outbound
shipments are not made in a integrated way. As the example in Figure 1, if the truck from Supplier A has not
arrived at the DC, none of the customer trucks is ready to leave. A poor scheduling plan for the cross-docking
2
operation can cause the congestion at the docks and the temporary storage and sorting/consolidating areas to
further affect the efficiency of the DC.

D C B A

c
b
a
CD Distribution Center
Supplier
Customer

D C B A

D C B A

c
b
a
CD Distribution Center
Supplier
Customer

c
b
a
CD Distribution Center
Supplier
Customer
CD Distribution Center
Supplier
Customer
CD Distribution Center
Supplier
CD Distribution Center
Supplier Supplier
Customer

Figure 1: Illustration of Cross-Docking Operation at Distribution Centers

For this study, the Cross-Docking Scheduling Problem (CDSP) determines the schedules of the inbound
(supplier) and outbound (customer) trucks for a DC with the objective of minimizing the total work time span.
Usually, an approximate solution algorithm is needed for the CDSP with a practical size, as the problem
complexity is increased in a factorial way. In particular, multiple trucks are considered in this study. Owing to
the nature of the sequencing problem associated with the CDSP, this study chooses the Ant Colony
Optimization (ACO), which has some inherited advantages over other meta-heuristics due to its sequential
framework for the searching process and the solution building procedure. An ant-based procedure is used to
generate an initial customer service sequence, by which the feasible solution is derived based on some simple
steps without too much computation effort.

The remainder of this paper is organized as follows. The second section provides the literature review. The
mathematical model of the CDSP and the solution algorithm are presented in the third and fourth section
respectively. The numerical experiment is described in the fifth section. Finally, conclusions are drawn in the
last section.

2. Literature Review

According to Agustina, et al.(2010), the mathematical models for cross docking can be classified as three
major levels: strategic, tactical, and operational. The focus of this study is the operational decision regarding
the scheduling of the trucks, and other operational issues such as dock door assignment and vehicle routing
are not considered. For a more general and complete introduction to CDSP, the recent survey papers of
Boysenand and Fliedner (2010) and Agustina et al. (2010) serves as excellent sources for more related
information.

This study focuses on the truck scheduling for cross-docking operation, unlike some integrated models such as
Shakeri et al. (2008) combining the truck scheduling problem with the dock door assignment. The type of the
CDSP in this study is similar to the problem of Chen and Lee (2009). The objective is to minimize the total
work time span, but their work is under the context of single-truck for both suppliers and customers. A
branch-and-bound algorithm is developed to derive the optimal solution. Lin and Lin (2010) deals with a
model similar to that in Chen and Lee (2009), but with the context under the manufacturing application with
two flow shop machines. The products of the second stage machine requires some product-specific semi-
finished products made by the first stage machine. A branch-and-bound exact algorithm is developed to find
the optimal solution for the problem with the size of 10 and 40 semi-finished and final products. In addition,
an iterative approximate heuristic is designed to deal with larger problem instances.

Song and Chen (2007) first extend the single-truck problem to the version with multiple inbound/supplier
trucks. Chen and Song (2009) later extend to the case of multiple trucks for both suppliers and customers.
They develop an integer programming (IP) model, which is solvable for the problem with very small size. For
larger problems, they develop a approximate heuristic based on the classic J ohn's Rule. For this study, the
3
same problem with some formulation modification is solved by a new solution algorithm, which provides a
better solution quality as described in the following sections.

3. Mathematical Model

The following IP formulation of (1) to (13) basically follows the one in Chen and Song (2009), though some
minor modifications have been made. Based on the numerical experiment of Chen and Song (2009), only
small instances of this IP formulation can be solved by current IP solvers, and heuristic algorithms are thus
required. The study develops an ant-based algorithm to tackle this problem, and the details of the algorithm
design are presented in the next section.

max
. C Min
(1)
s.t.

=
U u
iu
I i x 1
(2)

=
V v
jv
J j y 1
(3)
I i p C
i
a
i
(4)
j
j
a
i
b
j
S i J j q C C + ,
(5)
U u i I h I i p C w x x Q C
h
a
i ih hu iu
a
h
+ + }, { , ) 3 ( (6)
V v j J k J j q C z x y Q C
k
b
j jk kv jv
b
k
+ + }, { , ) 3 (
(7)
I i C C
a
i

max
(8)
J j C C
b
j

max
(9)
U u I i binary x
iu
, : (10)
V v J j binary y
jv
, :
(11)
I h i binary w
ih
, : (12)
J k j binary z
jk
, :
(13)
Notations
i, j:

indices of suppliers and customers respectively (I and J are the corresponding sets.)
u, v: indices of supplier and customer trucks respectively (U and V are the corresponding sets.)
p
i
, q
j
: processing (service/transportation) time for i and j respectively
Q: a big number less than the maximum possible completion time
C
max
: total work time span
C
i
a
, C
j
b
: completion time of i and j respectively
s
ij
: binary parameter for precedence requirement, if =1 indicating goods from supplier i is needed by customer j.
S: matrix of s
ij
(S
i
and S
j
are the ith row vector and the jth column vector of S respectively.)
x
iu
: binary variable, if =1 indicating supplier i is served by truck u.
y
jv
: binary variable, if =1 indicating customer j is served by truck v.
w
ih
: binary variable, if =1 supplier i is served before supplier h.
z
jk
: binary variable, if =1 customer j is served before customer k.

The objective (1) is to minimize the total work span, considering serving all suppliers and customers
Constraint (2) and (3) ensure suppliers and customers are served by one of their associated trucks respectively.
The completion time of each individual supplier is related to Constraint (4). The individual completion of each
customer is related to Constraint (5), which further considers its precedence requirement. Constraint (6)
regulates the relationship of the completion times between two suppliers that are assigned to the same truck.
Under such a case (x
iu
=x
hu
=1), if supplier i is scheduled to be served before supplier h (w
ih
=1), the
completion time of h would be that of i plus the processing time of h. Constraint (7), similar to (6), specifies
the completion time relationship among any pair of customers. Based on Constraint (8) and (9), C
max
is the
maximum of the completion times of all suppliers and customers and leads to the definition of the total work
time span. Finally, Constraint (10) - (13) define the binary variables.


4
4. Solution Algorithm

The whole solution algorithm designed in this study consists of two major parts, which are presented in the
following two sub-sections. The initial service sequences of customers under the condition of one single truck
are first generated by a group of ants, and the initial sequence generated by each ant is used to derive a
feasible solution, which includes the assignment and schedule of trucks for both suppliers and customers
under the condition of multiple trucks. Once the feasible solutions and their associated objective function
values for the ant-generated sequences in one iteration are derived, the pheromone is updated and used in the
next iteration.

4.1. Ant Sequence Generation Procedure

The first ACO algorithm was the ant system algorithm (ASA), which was proposed in the Ph.D. thesis of
Dorigo (1992) and formally published in Dorigo et al.(1996). The ASA together with several ACO variants
have been found to be effective for many combinatory optimization problems, including several sequencing
problems. For example, there were several successful applications to the traveling salesman problem (TSP),
such as Dorigo and Gambardella (1997) and Stutzle and Hoos (1997). For the introduction to the ACO and its
associated algorithms and applications, Dorigo and Stutzle (2004), Dorigo et al. (2006) and Mullen et al.
(2009) serve as excellent references.

For this study, the search space of ants is illustrated as the two-dimensional space consisting of the customers
and the stages. The latter in fact has the number equal to the number of customers as shown in Figure 2. For
each stage, the n nodes correspond to the choice of the n customers. The links are created only between the
nodes between two adjacent stages. Beginning with a dummy origin as the stage of zero, the ant chooses the
link based on (14), a typical choice probability defined in many ant-based algorithms. The probability is based
on the pheromone deposited by the ants in the previous iterations. In addition to pheromone, some problem
specific information, the heuristic value, is also incorporated. Once the ant reaches the dummy final
destination at the stage of (n+1), the initial assignment sequence is derived. As shown in Figure 2, the
illustrative sequence is 1-2-5-3-4-10-8-9-6-7, which is treated as the sequence of customers to be served, if
there is only one truck.


Figure 2: Illustration of Ant Search Space

otherwise
t t
t t
t B j if
t p
t B l
k
gil il
k
gij ij
k
g
k
gij
k
g
) (
)] ( [ )] ( [
)] ( [ )] ( [
) ( 0
) (




(14)
t: index of iteration

ij
(t): pheromone on link (i, j) for iteration t
: parameter to control the influence of pheromone

k
gij
(t): heuristic value of link (i, j) at stage g for ant k in iteration t
: parameter to control the influence of the heuristic value
B
k
g
(t): set of nodes already visited by ant k at stage g in iteration t
p
k
gij
(t): probability of choosing link (i, j) at stage g for ant k in iteration t
Stage
Customer
5

One of the most important tasks for designing an ant-based algorithm is the setting of the heuristic value. For
the classic TSP, the heuristic value is usually set as the inverse of the link distance, which attracts the ants to
move to the neighboring nodes. Based on the characteristics of the CDSP, this study sets the heuristic value by
(15), in which a new set is defined to consider the number of suppliers required by the customer currently
under consideration, but not required by the customers already visited by the ant. A customer with fewer such
kind of suppliers is more likely to be selected and thus to be served earlier. This idea is similar to the concept
of LSFS (Less-Supplier First-Served) highlighted in Chen and Lee (2009).

{ } ) ( , | ) ( ,
) 1 ) ( (
1
) ( t B u S u u t U where
t U
t
k
g
j k
gj k
gj
k
gij
=
+
=
(15)
U
k
gj
(t): set of suppliers that is needed by customer j, but not needed by the customers already visited by ant k at stage g
in iteration t

Given the iterative structure of the ASA, the pheromone is updated from iteration to iteration based on (16), in
which the first part is about the evaporation, and the second part is the enhancement for the links used by ants.
In particular, the increase of pheromone for the used links is affected by the quality of the associated solution
as shown in (17).

=
+ = +
N
k
k
ij ij ij
t t t
1
) ( ) ( ) 1 ( ) 1 (
(16)

=
otherwise
t iteration for k ant by used is j i if
t f
M
t
k k
ij
0
) , (
) ( ) (
(17)
: parameter to control the evaporation of pheromone
N: Number of ants for each iteration
f
k
(t): objective function value achieved by ant k in iteration t
M: parameter to control the impact of solution quality on pheromone increase

Based on (14), the larger the pheromone and the heuristic value are, the more likely the link is chosen.
However, the relative importance of the pheromone versus the heuristic value is controlled by the two
parameters and . These two parameters together with the other two parameters and M build up a
mechanism, which makes the ants to follow the promising trials found earlier and, at the same time, search the
less-explored space so as to avoid being trapped at the local optimal solutions.

4.2. Feasible Solution Generation Procedure

The sequence generated by an ant as discussed in the previous sub-section is treated as the initial service
sequence of customers (i.e., the outbound shipments of a DC) under the condition of one truck. Based on this
sequence, the following 3-step procedure further transforms it into the assignment and scheduling of multiple
trucks for both suppliers and customers.
Step 1: Under the assumption of one supplier truck, determine the sequence of suppliers to be served based on
the ant-generated service sequence of customers. That is, if a customer is served earlier, its
corresponding suppliers are also served earlier. (Of course, if the service of a supplier is already
scheduled under the consideration of an earlier customer, it is not served for the second time.) For the
multiple suppliers of the same customer, the tie can be broken arbitrarily.
Step 2: Based on the supplier service sequence in Step 1, schedule the beginning of the service for the
supplier as soon as a truck is available.
Step 3: Given the service schedule of the multiple supplier trucks in Step 2, whenever a customer truck is
available, dispatch it to serve the customer that is ready to be served (i.e., the processes of all required
suppliers have already been finished) and has the largest processing time.

The design of Step 1 is motivated by Lin and Lin (2010), in which, under the condition of single truck for both
suppliers and customers, it is proved that the optimal supplier sequence can be derived according to the way in
6
Step 1 if the customer service sequence is fixed. Once the schedule of the supplier and customer trucks are
determined, the objective function (i.e., the total work time span) is computed, and the pheromone on the links
is updated for the next iteration.

5. Numerical Experiment

The design of the test problems basically follows that of Chen and Song (2009), but with some modifications,
as shown in Table 1. Regarding the problem scale, the number of suppliers (n
1
) is set at four levels (40 to 70),
and the number of customer are randomly generated based on a uniform distribution with the range of 0.8 to
1.2 times of the number of suppliers. The processing time for both suppliers and customers are assumed to be
uniformly distributed between 10 and 100.

Table 1: Design of Test Problems
Factors Values
Number of suppliers (|I|) 40, 50, 60, 70
Number of customers (|J|) Unif [0.8|I|, 1.2|I|]
Processing time (p
i
and q
j
) Unif[10, 100]
Number of trucks (|U| = |V|) 4, 10
Precedence requirement Unif[0, 1]

In addition to the issue of problem scale, four groups of test problems are designed to examine the
performance of the developed ant-based algorithm for the CDSP. The same number of trucks are used for both
suppliers and customers, and it is set as 4 or 10 to represent two levels of resource availability. Regarding the
precedence requirement, the entries of the |I|-by-|J| matrix S are generated by a uniform distribution between 0
and 1. If the value is higher than 0.75, the corresponding parameter s
ij
is set as 1, indicating customer j needs
the product from supplier i. Otherwise, it is set as 0. Unlike Chen and Song (2009), this study makes a setting
of s
ij
that does not lead to a dense precedence requirement matrix. It is believed that a dense precedence matrix
would make the scheduling decision less challenging. The reason is that the scheduling of customer trucks is
less relevant to the scheduling of supplier trucks, as not many customer trucks can be dispatched before all
suppliers are served. Under such a situation, the shipping decisions of the two stages (i.e., inbound and
outbound) at a cross-docking DC is almost de-coupled. Another threshold of 0.85 is used to further reduce the
density of the precedence requirement matrix. Thus, given the two levels of resource availability (4 and 10
trucks) and the two levels of precedence requirement (low and high, represented by the threshold of 0.85 and
0.75 respectively), there are four scenarios as shown in Table 2.

Table 2: Results of Numerical Experiment
Scenario Scale
Trucks Precedence Requirement 40 50 60 70
4 Low 8.7% 8.2% 7.6% 5.2%
4 High 7.3% 5.8% 6.3% 4.3%
10 Low 8.5% 8.9% 8.0% 8.5%
10 High 9.7% 10.4% 8.9% 6.4%

For each scenario and each problem scale, 10 test problems are generated. The solution of the ant-based
algorithm is compared with the heuristic of Chen and Song (2009), an approximate method based on the
classic J ohnson's Rule. The improvement in terms of the average of the 10 test problems is shown in Table 2.
For all problem scales and scenarios, the solution quality of the ant-based algorithm is consistently better than
that of the heuristic in Chen and Song (2009). In particular, it seems that the ant-based algorithm shows more
advantage when more trucks are available. However, its improvement is degraded, to some extent, when
problem scale is increased.

Regarding the environment of the numerical experiment, the operating system is Windows XP SP3, and the
hardware is a personal computer with Intel Core 2 Duo E8400 3.00GHz CPU and 2G RAM. Both algorithms
are coded by MATLAB. For the heuristic of Chen and Song (2009), the computation time is very short, only
about a couple of seconds. For the ant-based algorithm of this study, it takes 30 - 180 seconds for the various
scales of test problems in Table 2. Although the solution quality of the ant-based algorithm is significantly
7
better than that of the simple heuristic algorithm based on the classic J ohn's Rule, the extra computation time
is not nominal for today's personal computers. Therefore, there is an incentive to improve its computational
efficiency. Nonetheless, the developed ant-based algorithm should be able to handle most of the practical
cross-docking operations nowadays based on its current capability.

6. Conclusions

This study focuses on the Cross-Docking Scheduling Problem (CDSP), which determines the schedules of the
inbound (supplier) and outbound (customer) trucks for a DC with the objective of minimizing the total work
time span. This study designs an ant-based algorithm to take advantage of its sequential framework for the
searching process and the solution building procedure. Based on the results of the numerical experiment, the
solution quality of the developed algorithm is promising for providing a new approach to solve the CDSP.

For future research directions, the immediate one is to boost the computational efficiency of the ant-based
algorithm, even though its current capability in general can meet the practical needs. A reduction of the
computation time is expected, if more effort is paid to the coding languages and skills. In addition, since the
ant system algorithm is a very basic version of the ant-based algorithm family, search efficiency,
computational performance, and solution quality can be improved by employing some advanced ant-based
algorithms to generate superior initial sequence.

Meanwhile, the current solution algorithm can be improved if a more sophisticated procedure is developed to
replace the 3-step procedure in Sub-section 4.2 to derive the feasible solution based on the ant-generated
initial sequence. For example, Step 1 can generate the best supplier truck sequence based on the ant-generated
customers sequence under the assumption of one single truck. If the multiple supplier trucks are considered at
the same time, there is a chance to derive a better schedule.

Finally, more practical considerations of the cross-docking operation at distribution centers can be included in
the model. For example, the customer service time window can be a critical issue for some industries or types
of business. However, the challenge is that a much more complicated procedure is needed, without too much
computation effort, to derive a good feasible solution based on the ant-generated initial sequence.

7. References

Agustina, D., Lee, C.K.M. and Piplani, R. (2010), A review: mathematical models for cross docking planning,
International J ournal of Engineering Business Management, Vol. 2, No. 2, pp. 47-54.

Boysen, N. and Fliedner, M. (2010), Cross dock scheduling: Classification, literature review and research
agenda, Omega, Vol. 38, pp. 413-422.

Chen, F. and Lee, C.Y. (2009), Minimizing the makespan in a two-machine cross-docking flow shop problem,
European J ournal of Operational Research, Vol. 193, pp. 59-72.

Chen, F., and Song, K. (2009), Minimizing makespan in two-stage hybrid crossdocking scheduling problem.
Computers and Operations Research, Vol. 36, pp, 2066-2073.

Dorigo, M. and Gambardella, L. M. (1997), Ant colony system - a cooperative learning approach to the
traveling salesman problem, IEEE Transactions on Evolutionary Computation, Vol. 1, pp. 53-66.

Dorigo, M. and Stutzle, T. (2004), Ant colony optimization, MIT Press, Cambridge, MA.

Dorigo, M. (1992), Optimization, Learning and Natural Algorithms (in Italian), Ph.D. Dissertation,
Dipartimento di Elettronica, Politecnico di Milano, Italy.

8
Dorigo, M., Birattari, M. and Stutzle, T. (2006), Ant colony optimization, IEEE Computational Intelligence
Magazine, Vol. 1, pp. 28-39.

Dorigo, M., Maniezzo, V. and Colorni, A. (1996), Ant system: Optimization by a colony of cooperating agents,
IEEE Transactions on Systems, Man, and Cybernetics - Part B, Vol. 26, pp. 29-41.

Lin, B.M.T. and Lin, Y.M. (2010), Two-machine flowshop scheduling with supportive constraints,
Proceedings of the 40th International Conference on Computers and Industrial Engineering, Awaji Island,
J apan.

Shakeri, M., Yoke, M., Hean Low, M.Y., and Li, Z. J uly13-16, (2008), A generic model for crossdock truck
scheduling and Truck-to-Door assignment problems, IEEE International Conference on Industrial Informatics,
pp. 857-864.

Song, K. and Chen, F., August 18-21, (2007), Scheduling cross docking logistics optimization problem with
multiple inbound vehicles and one outbound vehicle, Proceeding of the IEEE International Conference and
Automation and Logistics, J inan, China, pp. 3089-3094.

Stutzle, T. and Hoos, H. H. (1997), The MAX-MIN ant system and local search for the traveling salesman
problem, in Back, T. et al., (eds.), Proceedings of the 1997 IEEE International Conference on Evolutionary
Computation (ICEC97), IEEE Press, Piscataway, NJ , 309-314.

Mullen, R. J ., Monekosso, D., Barman, S. and Remagnino, P. (2009), A review of ant algorithms, Expert
Systems with Applications, Expert Systems with Applications, Vol. 36, pp. 9608-9617.

Das könnte Ihnen auch gefallen