ISSN (Print) : 0974-6846

Indian Journal of Science and Technology, Vol 8(12), IPL010, June 2015 ISSN (Online) : 0974-5645

A Hybrid Approach for Cross-Docking Scheduling

Tokhmehchi Nafiseh1* and Makui Ahmad2
Department of Industrial Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran;
1

[email protected]
2
Department of Industrial Engineering, Iran University of Science and Technology, Tehran, Iran; [email protected]

Abstract
Background/Objectives: Cross-docking is an innovative logistical strategy, which means discharging the cargos from
in-bound trucks into the warehouse directly and saved automatically. This article investigates the scheduling of a distribution
system consisting of suppliers, cross-docks and the ultimate destinations. The aim of analyzing this problem is presenting a
scheduling model so that lateness, earliness and holding costs will be minimized. Methods/Statistical Analysis: Restrictions
are applied on the number of arrival and exit doors, departure time, release time, earliness, lateness, and the required space
for discharging. Since the offered model is NP-Hard and highly complicated, the hybrid form of firefly and genetic algorithms
is applied to solve the model. Results: Tuning the parameters performed and several numerical examples shows the appli-
ance of this model. Comparisons via ANOVA and interval plot show that hybrid form can act better than GA and FF in large
sizes. Conclusion/Application: The innovation of this article is thanks to its model type and solution method.

Keywords: Crossdocking, Hybrid Firefly and Genetic Algorithms, Logistics, Metaheuristic, Scheduling

1.Introduction this purpose, several scheduling procedures have been

introduced over the recent years. In general, scheduling
Cross-docking is a creative distribution strategy for the helps taking advantage of limited resources in a certain
goods that are time-sensitive. This method works as fol- time period so to optimize the objectives.
lows. The in-bound trucks are discharged in a cross-dock, The scheduling process for the trucks commonly divides
and directly reloaded to the out-bound trucks with mini- transportation procedure into two parts: 1) Discharging
mum standby time (standby time is commonly shorter in-bound trucks, 2) Loading out-bound trucks. Of course
than 24 hours). The following costs are reduced in this there is a time lag for conveying the cargos inside the cross-
strategy: Warehousing, maintenance, transportation, and dock terminal. This delay includes the required time for
human resources. This method also leads to shorter late- investigating, categorizing and carrying the materials.
ness in delivery, improving client satisfaction, needing less Each terminal has several doors. The doors are scheduled
warehousing room, smoother work flow, as well as lower for the trucks. The distance between each pair of doors is
risk and loss. The benefits of cross-docking system are already determined; hence, there is another delay for carry-
more conspicuous in cases with higher distribution costs. ing the materials between the pairs of doors as well1.
A good example holds for car manufacturers. In order to On the whole, scheduling is a new horizon in cross-
meet their needs of their assembly lines and distribution docking systems. It helps the managers transport and
centers, they have been increasingly attracted to cross- distributes their cargos faster. This improves the overall
docks. These docks are usually run by logistical companies efficiency of the organization.
in order to manage on-time delivery of goods. The remainder of the paper is organized in the following
Efficient operations of cross-docks require accurate manner. In Section 2 we briefly discuss the relevant
and simultaneous operations of in-bound and out-bound literature of cross-docking scheduling. Section 3 pres-
trucks by contriving proper scheduling approaches. For ents the problem and mathematical formulation of this.

*Author for correspondence

 A Hybrid Approach for Cross-Docking Scheduling

Comprehensive explanations of solution methodology the trucks were available in the beginning of scheduling
based on hybridization of genetic algorithm and firefly horizon. The authors of this article proved that the prob-
algorithm is discussed in section 4. Section 5 is dedicated lem in question is NP Hard; hence, a heuristic approached
to tuning the parameters and numerical examples. Finally based on Johnsons law was suggested for solving it.
in section 6 conclusions is presented and some directions Vahdani and Zandieh6 considered the previous prob-
for future work are discussed. lem and used 5 meta-heuristic approaches for solving the
model. The answer of Problem3 was used as the initial
answer. It was proved that these meta-heuristic methods
2. Literature Review improve the results, with slightly longer run time.
There are not many researches about cross-docking. For Boysen & Fliedner1 were presented an optimizing
example up until 2005, no article has been published model. This model assigns fixed schedule for out-bound
about short-term scheduling for the trucks1. The current trucks, with the aim to minimize the number of delayed
part of this study has a brief shot about several researches transportations. This model considered the travel time
on cross-docking. McWilliams et al.2 were considered in- between arrival and exit doors as well. They also proved
bound truck scheduling for cross-docks of postal parcels that this model is NP Hard.
in postal industries. In this system, the parcels were con- In order to reduce the number of goods passing standby
veyed to out-bound trucks using fixed conveyor belt. The area, a sequence was presented for in-bound and out-bound
objective of this scheduling model was minimizing the trucks by Forouharfard and Zandieh7. Also Imperialist
time between discharging the first parcel till loading the Competitive Algorithm was used for solving the problem.
last postal parcel. Also, in order to detect the amount of Vahdani et al.8 were considered a similar problem, and for-
delays, a solution based on genetic algorithm was presented. mulated it as a mixed-integer programming model. They
In the end, numerical calculations showed that their sug- used genetic algorithm and electro-magnetism algorithm
gested method has lead to significant reduction of delays. for solving the problem. Two meta-heuristic algorithms,
Yu and Egbelu3 studied a cross-docking system with one including simulated annealing, and variable neighbor-
arrival and one exit door. The objective was minimizing hood algorithm were used for solving the similar problem
the total operation time. The cargos were considered to be by Soltani and Sajadi9. Arabani et al.10 assumed that all out-
interchangeable. Assigning the cargos to trucks and the bound trucks have due date. The target function aimed
travel-time between doors were supposed to be already to minimize the earliness and lateness for trucks. Three
determined. The only differences were the truck inter- meta-heuristic approaches were demonstrated for solv-
change time, and optional sequence for discharging the ing the problem. These approaches included genetic
in-bound trucks. In the end, the problem was formulated algorithm, evolutionary algorithm, and particle swarm
with a mixed-integer model, and solved using a heuristic algorithm. A similar problem was considered by Arabani
algorithm. In Wang and Regan4 a scheduling model was et al11. The time span was sectioned into several parts. This
contrived for in-bound trucks. Several dispatching rules, problem was formulated as an NP Hard model. It used
required for real dynamic environments, were presented break down approach, including two segments: 1. Finding
as follows: Among the few in-bound trucks that are wait- a fixed sequence for in-bound trucks; 2. Determining the
ing for discharging their cargos, only the first truck was sequence of out-bound trucks. Those subsets could find
considered. The next trucks were selected with First- a near-optimum answer with heuristic algorithms, and
com first-served policy. This method pays attention the accurate answer with Dynamic Planning Approach.
only to the standby time of in-bound trucks, which does Alpan et al.12 were considered a cross-dock with several
not influence the whole system significantly. In the end, arrival and exit doors. The aim of analyzing this problem
2 algorithms were introduced for solving the problems. was finding an optimum or near-optimum answer for the
Chen and Lee5 investigated the problem of cross-docking scheduling model, so to minimize the total costs of con-
flow shop scheduling aimed to find the optimum sequence veyance operations. Agustina et al.13 developed a model for
of in-bound and out-bound trucks that could minimize scheduling a cross-dock with delivery time window and
makespan. The discharge and loading times were con- temporary storage. Their target function aims to minimize
sidered to be different for each truck. In some cases, this distribution cost and holding cost. The resulted model was
included travel time as well. Also, it was assumed that all Mixed-Integer. They used Lingo for solving the model.

2 Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology
 Tokhmehchi Nafiseh and Makui Ahmad

Li et al. (2012) introduced 3 different aspects of reloaded to outgoing trucks, or stay in the warehouse for
ptimum planning, including trucks-grouping, clustering,
o a period, usually less than 24 hours. Other conditions and
and in the end determining the sequence of trucks and assumptions in this problem are as follows:
assigning the trucks to the docks in cross-docking system. Unlike the time of exiting from warehouse that is
McWilliams and McBride13 used beam search algorithm variable, the arrival time to the warehouse is already
for solving the problem of scheduling postal parcel distri- determined.
bution centers, with the aim to minimize the time span of The warehouse space for discharging the incoming
the transfer operation in the terminal. They finally com- trucks is limited, but it is unlimited for outgoing trucks.
pared their suggested algorithm with other approaches. The number of incoming and outgoing trucks is limited.
Bellanger et al.14 were used a compound 3-step flow shop
model with aim to minimize completion time. They also 3.1 Notations, Parameters and Variables
used branch and bound algorithm. This article inves-
The following notations, parameters, and variables are
tigates the problem of truck scheduling in uncertainty
used for mathematical model of the problem:
conditions of truck arrival for in-bound trucks. Buijs et
al.15 focused on the problem of correlation among dif- i Index of items i = 1, 2, , I
ferent aspects of cross-docking problem, with the aim to n Index of inbound doors n = 1, 2, , N
support and develop the future practical decisions. They m Index of outbound doors m = 1, 2, , M
also presented an overall classification of decision trends zi Earliest delivery time for item i (hours)
in this area. Miao et al.16 used adaptive tabusearch to solve di Latest delivery time for item i (hours)
the problem of contriving a cross-docking management vi Arrival time for item i (hours)
system. Some numerical tests proved that the suggested a Earliness cost rate
solution, both with regards to effectiveness and efficiency, b Lateness cost rate
overcomes CPLEX solution.
d Inventory holding cost rate
In previous models of cross-dock scheduling, the
H Planning horizon h = 1, 2, , H (hours)
discharge area of in-bound trucks was considered to be
fi Required space for item i (m2)
unlimited, but in the real industrial world such condition
F Total space in in-bound yard (m2)
is completely impossible. In order to address this research
gap, a mathematical model was presented. The mathemati- Decision Variables:
cal model of this article is a development to13 model. Since ri Release timetime for item i (hours)
the previous model was NP Hard, the new one is obviously ai Departure time for item i (hours)
NP Hard as well. We can use metaheuristic algorithms for ei Earliness for item i (hours)
solving this problem. The difference of the mathematical ti Lateness for item i (hours)
model presented in this article, with the one suggested xinh If item i loaded at inbound dock door n at time h = 1;
by 13 was considering the assumption of having limited xinh = 0, otherwise
discharge space. This model minimized the earliness and ximh If item i unloaded at outbound dock door m at time
lateness costs, as well as holding costs. Additionally, it h =1; ximh = 0, otherwise
considers the condition of limited discharge space.
In the next part we will introduce the mathematical
3.2 Formation of the Model
model and its conditions.
The mathematical model of the problem becomes:
3.The Suggested Mathematical Minimize
Model I I I
TC = a ei + b ti + d (ai - ri )  (1)
In the beginning of this plans time-scope, the retailers i =1 i =1 i =1

(ultimate targets) order their requirements to the suppli- ST:

ers. The suppliers will also send the cargos from the docks
H N
using trucks. These trucks, after arriving in the warehouse
environment, either discharge their cargo to be directly
x
h =1 n =1
inh = 1 i = 1, 2, , I  (2)

Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology 3
 A Hybrid Approach for Cross-Docking Scheduling

H M
For larger scales, other solutions must be contrived.
x
h =1 m=1
imh = 1 i = 1, 2, , I  (3)
This article uses Hybrid form of Firefly Algorithm and
Genetic Algorithm as a solution method; therefore, in
N I
this section a brief description of each innovative solu-
x
n =1 i =1
inh N h = 1, 2, , H  (4)
tion will be raised. In each section, the solution structure
will be reviewed. In the end the hybrid solution will be
M I
suggested.
x
m =1 i =1
imh M , h = 1, 2, , H  (5)

N I
4.1 Genetic Algorithm
f x
n =1 i =1
i inh F h = 1, 2, , H  (6) Genetic Algorithm (GA) is a Search_algorithm heuris-
tic belong to the larger class of evolutionary algorithms
ai H, i = 1, 2, , I (7) (Gen, 1997). In this approach, a random solution will be
selected initially. Afterwards, the fitness function value
vi ri ai, i = 1, 2, , I  (8) will be calculated for each chromosome of the initial pop-
ulation. A suitable number of chromosome pairs shall be
ai - di ti - ei, i = 1, 2, , I (9) selected in this phase according to their fitness levels, so
to be used in later steps.
zi - ai ti - ei, i = 1, 2, , I (10)
In the next phase, crossover operator with probability
xinh, ximh = 0, 1 (11) of Pc will act upon the parent chromosomes, and com-
bines them to create new chromosomes (offspring). Then
ri, ai, ei, ti 0 (12) mutation with probability of Pm will happen on the chro-
mosomes created from crossover operation. Actually, by
The objective function of this problem is regarded changing the bits of these chromosomes, a new way for
as cost, and consists of 3 elements: Earliness penalty entering new data will be achieved. Then, in order to eval-
cost, lateness penalty cost, and cargo warehousing cost. uate the offspring, the fitness level of new chromosomes
Equation 2 indicates that every cargo uses only one arrival shall be calculated.
door for discharging. In the next level, a new population will be selected for
Constraint 3 guarantees that each cargo will use only the input of next phase. If the end conditions of algorithm
one exit door for reloading. Constraint 4 shows that the happen, the algorithm will end up; otherwise, the avail-
numbers of discharging procedures are fewer than the able population will be used as the input population of
total number of arrival doors. Constraint 5 makes sure the next phase.
that the number of reloading procedures will not exceed
the number of exit doors.
4.1.1 Initial Population
Next restrictions guarantee that the space required
for discharging in-bound cargos is limited. Constraint 7 In this paper, just like many genetic problems, the initial
demonstrates departure time is within the time scope of population is created randomly. The reason is creating a
the plan. Constraint 8 displays that release time is a num- large range of solutions. Of course the information from
ber between arrival time and departure time. Constraints systems users can be used as well. Their suggestions and
9 and 10 are respectively used for determining earliness comments might be applied to the initial population. An
and lateness. Equations 11 and 12 display the range of accurate solution shall also help.
decision variables.
4.1.2 Chromosome Structure
According to the decision variables of this model, the
4. Solution Method
following chromosome structure is suggested. Each row
Like mentioned before, the mathematical model of this indicates the variable values for one type of product. For
problem is NP Hard (Garey & Johnson, 1979). It has example the elements of second row are respectively:
complicated search space. In other words, accurate solu- Release time, departure, earliness, and lateness of product 2.
tion can be available only for small scales of this problem. The last 2 indexes show having or not having discharged

4 Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology
 Tokhmehchi Nafiseh and Makui Ahmad

r1 a1 e1 t1 x1nh x1mh This study generates a random chromosome using the

r a2 e2 t2 x2nh x2mh genes associated with numbers between 0 and 1. Then
2 the genes with mutation probability less than Pm will be
. . . . . .
selected. The parent chromosomes in similar conditions
. . . . . . will change randomly.

. . . . . .

rI aI eI tI x Inh x Im h 4.1.5 Investigation of Stop Condition
In this article, the algorithm stops when a certain number
Figure 1. The chromosomal structure.
of loops happen.

product 2 in door number n and in the time horizon h.

4.2 Firefly Algorithm
The next member means having or not having reloaded at
door number n and in time horizon h. Firefly algorithm is an evolutionary model, based on
Swarm intelligence algorithms, originated from nature.
4.1.3 Survivability of Each Chromosome This algorithm was introduced by Yang in 2007. Its main
idea arose from the optical relation between fireflies. In
Since the Fitness Function is a negative aspect, it must be
this algorithm, cooperation (and probably competition)
minimized. In order to calculate survivability, number 1 is
of simple and low-intelligence members lead to higher
divided by Fitness Function Tc. The result will be divided
levels of intelligence, that could not be achieved by any of
by the sum of each and every quotient. The quotient is
the members alone.
obtained from dividing 1 by every Tc.
This algorithm is based on getting to optimum limit
1 when investigating behaviors and characteristics of fire-
flies. It consists of 3 main attributes: 1) Firefly becomes
TC j
Pi = pop - size j = 1, , pop - size  (13) lighter and more attractive when it moves randomly. 2)
1
TC j
The beauty and attractiveness of fireflies are proportional
j =1 to their brightness and light level. This attraction decreases
by increasing the distances (of course this comparison is
4.1.4 Producing the Next Generation from the view point of other fireflies). 3) The light inten-
sity decreases with the rise in light absorptive factor. It is
Three operators are reviewed here as follows:
controlled with a certain scale.
1) Copy: A% of the best chromosomes from previous Fireflies produce rhythmic and short flashes of light.
generation with the highest survivability will be cop- The flashing pattern of each firefly is different from the
ied without any change. others. The search module of firefly algorithm is as fol-
2) Cross-over: B% of chromosomes in each generation lows. Each firefly is compared with each and every other
will be selected from the best of previous generation, firefly. If the sample firefly has lower light than the com-
and crossed over each other in two-by-two couples. pared firefly, it goes toward the one with stronger light
This way, B offspring will be produced. Among B (the problem of finding maximum point). This action
offspring and B parents, B chromosomes with the results in concentration of the particles around the par-
optimum Fitness Function will be selected. ticles with stronger light. In the next loop of algorithm,
The cross-over method in this article is as follows: if a particle with stronger light appears, again the other
A chromosome consisting of 0 and1 will be generated particles will move toward it. There is a failing in this
randomly. Then the locations from each of the two method. The movements of the particles are adjusted
parents, which are associated with number 1, will be only according to the light level of one firefly, which is
substituted with each other. Other elements will remain in fact the local optimum. The global optimum has no
fixed. In result, two new offspring will be obtained. effect on the search module of this algorithm; therefore,
3) Mutation: C% of next generation chromosomes will be the problem space will not be searched optimally, and
generated with mutation. There are numerous meth- more loops will be required for finding the optimum
ods for mutating. point.

Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology 5
 A Hybrid Approach for Cross-Docking Scheduling

4.2.1 Steps of Firefly Algorithm Among different hybrids between algorithms,

The details of applied algorithm are as follows: c ombining via genetic algorithm can improve the search
A- Producing the initial solution and appraising performance and also it is a fortune to hybrid optimiza-
them (Note that the structure of each firefly is similar to tion to gain their objectives18.
Figure 1.) The range of hybrid algorithm methods is enormous.
B- For every other firefly (like a) and for every firefly This article applies Hybrid Iterative Sequentialization to
(like b), we calculate light intensity according to (14). its problem, with the help of Matlab software scripting.
According to the operations and attributes of the two

2
I = I 0 e - dr  (14) algorithms, firefly and genetic, considering the results of
numerous tests, we found out that firefly algorithm dem-
In (14) I0is Intensity Coefficient Base Value, d is Light onstrates more intention to exploit space. It has similar
Absorption Coefficient and r is the distance to light characteristics to crossover in genetic algorithm; hence, the
source. In addition light absorption coefficient typically results from hybrid method of Iterative Sequentialization
changes from 0.01 to 100 (Yang, 2010). seem to have higher quality.
C-IfIa is less than Ibthe new answer generate according The hybrid form of Iterative Sequentialization algo-
to (15). rithm acts as follows. First of all, the output of genetic
2
algorithm will be used as the input of firefly algorithm.
x = xa + b0 e - dr (xb - xa ) + ae  (15) Then the output of firefly algorithm will be used as the
input of genetic algorithm. This cycle continues until
In (15) xa,xbshow the position of firefly a,b.e is a
predetermined number of loops for calculating the tar-
random vector and b0 Attraction coefficient base value. In
get function will be passed. The following diagram shows
addition a is mutation coefficient and a [0,1]
a schematic view of Hybrid Iterative Sequentialization
Note that attraction coefficient base value usually
method.
considered 1 and random vector is equal to [rand-1/2]
(Yang, 2010).
D- Appraising of new population and selecting the 4.3.1 The Step of HGF
best solution. The steps of Iterative sequentialization Hybrid of Genetic
E- Loop to step 2, if the termination conditions are and Firefly algorithms (HGF) are shown below.
not met.
Produce an initial population (p) randomly.
While (not stop condition)
4.3The Hybrid Form of Genetic Algorithm
For pg=1: P
and Firefly Algorithm Mate pc percent of population and evaluate
The hybridization process of the two algorithms is like them.
the process of humans bearing child, with the difference Mutate pm percent of population and evaluate them.
that a hybrid algorithm might have more than 2 parents. Combine population in a single set and rank them.
These algorithms inherit part of the parents charac- Shorten best members and produce new population of
teristics. The target of algorithms hybridization is to size P.
create or improve new properties that are not existing in End for pg
a singlealgorithm17. Set ultimate population as fireflies.
For pf=1: P
For pl=1: P
If (If<Il)
Generate firefly and update it according to (15).
End if
End for pl
End for pf
Sort the solution and select the best.
Figure 2. Iterative sequentialization hybrid. End while.

6 Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology
 Tokhmehchi Nafiseh and Makui Ahmad

5. Computational Result Table 3. Obtained values in parameter tuning

CGA MGA IGA S LCFF MFF IFF ISHGF
This section divided in 3 parts. First general inputs are
1 0.3 0.15 50 50 0.01 0.1 50 0.4736
given. Next parts tunning the parameters via taguchi
methods implemented and finally several examples in 3 2 0.3 0.15 50 50 50 0.5 100 0.5469
sizes are given to illustrate the application of the proposed 3 0.3 0.15 50 50 100 0.9 150 0.8978
methodology in real-world environments. 4 0.3 0.2 100 80 0.01 0.1 50 0.3472
5 0.3 0.2 100 80 50 0.5 100 0.2571
5.1 General Inputs 6 0.3 0.2 100 80 100 0.9 150 0.3904
Table 1 shows the general data to solve the problem. 7 0.3 0.25 150 100 0.01 0.1 50 1.0334
8 0.3 0.25 150 100 50 0.5 100 0.9002
5.2 Tunning the Parameters 9 0.3 0.25 150 100 100 0.9 150 0.1787
In order to apply proposed algorithms in best mode, a 10 0.5 0.15 100 100 0.01 0.5 150 0.9064
parameter-tuning method named taguchi design was 11 0.5 0.15 100 100 50 0.9 50 0.5974
used (Taguchi et al., 2005). In these way parameters of 12 0.5 0.15 100 100 100 0.1 100 0.3904
3 types of hybrid identified. Proposed hybrids are based 13 0.5 0.2 150 50 0.01 0.5 150 0.0000
on Genetic and Firefly Algorithms, so the vital factors
14 0.5 0.2 150 50 50 0.9 50 0.2825
derived from them. Because of considering 7 param-
15 0.5 0.2 150 50 100 0.1 100 0.2067
eters with 3 levels, taguchi present 27 tests to be solved.
Following Tables show the level of parameters and the 16 0.5 0.25 50 80 0.01 0.5 150 0.0930
17 0.5 0.25 50 80 50 0.9 50 0.6035

Table 1. General data 18 0.5 0.25 50 80 100 0.1 100 0.9789

19 0.8 0.15 150 80 0.01 0.9 100 0.7103
parameters Value
20 0.8 0.15 150 80 50 0.1 150 0.9098
H 20
21 0.8 0.15 150 80 100 0.5 50 1.0844
vi Uniform(2,5)
Zi Uniform(5,8) 22 0.8 0.2 50 100 0.01 0.9 100 0.4278

di Uniform (8,12) 23 0.8 0.2 50 100 50 0.1 150 0.6776

a 3000 24 0.8 0.2 50 100 100 0.5 50 0.9637

b 12000 25 0.8 0.25 100 50 0.01 0.9 100 0.7795

d 2400 26 0.8 0.25 100 50 50 0.1 150 0.2101

fi Uniform(1,4) 27 0.8 0.25 100 50 100 0.5 50 0.4736

F 1000

Table 2. Levels of parameters

parameter 1 2 3
CGA: Crossover Percentage 0.3 0.5 0.8
MGA: Mutation Percentage 0.15 0.2 0.25
IGA: Number of GA Iterations 50 100 150
Hybrid of
S: Population size 50 80 100
GA and
FF LCFF:Light Absorption 0.01 50 100
Coefficient
MFF: Mutation Coefficient 0.1 0.5 0.9
IFF: Number of FF Iterations 50 100 150
Figure 3. Main effects plot for means ISHGF.

Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology 7
 A Hybrid Approach for Cross-Docking Scheduling

results of 27 runs in related percentage deviation (RPD) 5.4 Comparisons

form in (16).
In order to compare between 3 solutions, ANOVA
| A lg orihm Sol - Best Sol | Statistical Test was devised with SPSS Software in addi-
RPD = 100  (16) tion we consider each size separately to apply more focus
Best Sol
on it. Following tables and diagrams show the compari-
Finally, number achieved from 27 runs with different sons between algorithms.
parameter values create following mean of means plot From the results obtained by ANOVA in Table 6
with aim to MINITAB 17 Software. (sig>0.05) and interval plot in Figure 4 we can claim that
Lowest levels of every parameter in mean of means there is not significant difference between 3 proposed
plot has the best effects on final solution. Considering this solution in small size.
fact the best value reported in Table 4. The results obtained by ANOVA in Table 7 shows there
is significant difference between algorithms (sig<0.05) but
because of overlapping between algorithms in Figure 5 we
5.3 Computational Results
cannot certainly claim that ISHGF is the best in medium size.
In this section 18 different problems in 3 sizes (small, Because of value of sig (<0.05) in Table 8 and not exis-
medium and large) solved with HGF algorithm to show tence of overlapping in interval plot of large size we can
the appliance of proposed algorithm. In this way every report that ISHGF can act better among their parents.
test runs for 3 times and we report the average of them. Following diagram shows the convergance path of
problem number 5.
Table 4. The best parameter values
Algorithm CGA MGA IGA S LCFF MFF IFF
ISHGF 0.5 0.2 100 50 0.01 0.9 150

Table 5. Final solution

Tests I N M HGF GA FF
Small size
1 3 4 3 2.88 3.8743 3.1358
2 4 4 5 3.84 4.4844 4.2905
3 5 4 6 4.8 5.4919 5.4056
4 6 6 7 5.76 6.2629 6.0991
Figure 4. Interval plot for small size.
5 7 5 8 6.72 7.6728 7.6109
Medium size
6 20 13 17 19.2001 26.2401 25.3363 Table 6. ANOVA of small size
7 22 13 12 21.1034 27.7803 27.0147 Sum of Mean
8 24 15 13 23.0301 30.0847 30.2837 Squares df Square F Sig.
9 26 15 15 24.961 33.7748 32.6843 Between Groups 1.490 2 .745 .299 .747
10 28 16 14 26.7347 36.0363 33.6844 Within Groups 29.870 12 2.489
11 30 16 14 28.7928 43.1811 38.4040 Total 31.359 14
12 32 18 16 30.7302 54.8773 53.3551
Large size
Table 7. ANOVA of medium size
13 45 20 21 43.2009 101.7391 98.6454
14 50 21 22 48.0006 121.3595 119.3855 Sum of Mean
15 60 25 24 57.6004 141.8014 134.4125 Squares df Square F Sig.
16 70 26 30 67.3142 162.8291 151.0233 Between Groups 500.177 2 250.088 3.613 .048
17 80 30 35 76.8001 176.3561 172.1651 Within Groups 1246.009 18 69.223
18 100 50 60 96.0318 221.7715 203.8707 Total 1746.186 20

8 Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology
 Tokhmehchi Nafiseh and Makui Ahmad

scheduling model, which minimizes lateness and earliness

times, in common with holding cost. Restrictions have
been applied to the number of arrival and exit doors,
departure time, release time, earliness, lateness, and the
space required for discharging.
Since the problem model is NP Hard, which is highly
complicated, two metaheuristic algorithms were used:
Genetic and Firefly. These algorithms were hybridized
with Iterative Sequentialization to solve the problem. In
addition to apply proposed algorithms in best mode, a
Figure 5. Interval plot for medium size. parameter-tuning based on taguchi design was applied.
Finally 18 test problems in 3 sizes solved via ISHGF, GA
Table 8. ANOVA of large size and FF. Comparisons via ANOVA and interval plot show
Sum of Mean that HGF can act better than GA and FF in large sizes.
Squares df Square F Sig. Future studies might investigate the scheduling func-
Between Groups 14204.266 2 7102.133 3.888 .044 tion becoming multi-objective. For example decreasing
Within Groups 27402.139 15 1826.809 the number of delayed transportations can be another
Total 41606.404 17
objective. In addition the model can extend with con-
sidering social or environmental issues. Changes in the
solution approach can help develop the problem further.

7. References
1. Boysen N, Fliedner M. Cross-dock scheduling: Classification,
literature review and research agenda. Omega. 2010;
38(6):41322.
2. McWilliams DL, Stanfield PM, Geiger CD. The parcel
hub scheduling problem: A simulation-based solution
Figure 6. Interval plot for large size. approach. Computers & Industrial Engineering. 2005;
49(3):393412.
10
3. Yu W, Egbelu PJ. Scheduling of inbound and outbound
9. 5 trucks in cross docking systems with temporary stor-
9 age. European Journal of Operational Research. 2008;
8. 5
184(1):37796.
4. Wang JF, Regan A. Real-time trailer scheduling for cross
Best Cost

8
dock operations.Transportation Journal. 2008; 8:520.
7. 5
5. Chen F, Lee CY. Minimizing the makespan in a two-ma-
7
chine cross-docking flow shop problem.European Journal
6. 5
0 10 20 30 40 50 60 70 80 90 100
of Operational Research. 2009; 193(1):5972.
Iteration
6. Vahdani B, Zandieh M. Scheduling trucks in cross-docking
Figure 7. Convergancepath of ISHGF. systems: Robust meta-heuristics. Computers & Industrial
Engineering.2010; 58(1):1224.
7. Forouharfard S, Zandieh M. An imperialist competitive
6.Conclusion and Suggestions for algorithm to schedule of receiving and shipping trucks in
Future cross-docking systems. The International Journal of Advanced
Manufacturing Technology. 2010; 51(9-12):117993.
This article reviewed a cross-docking system, in which the 8. Vahdani B, Soltani R, Zandieh M. Scheduling the truck
cargos are discharged from in-bound trucks and reloaded holdover recurrent dock cross-dock problem using robust
on out-bound trucks with minimum standby interval. meta-heuristics. The International Journal of Advanced
This problem has been analyzed with the aim to develop a Manufacturing Technology.2010; 46(5-8):76983.

Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology 9
 A Hybrid Approach for Cross-Docking Scheduling

9. Soltani R, Sadjadi SJ. Scheduling trucks in cross-docking IEEE International Conference on Industrial Engineering
systems: A robust meta-heuristics approach.Transportation and Engineering Management (IEEM);2011 Dec. p. 1315.
Research Part E: Logistics and Transportation Review.2010; 14. Bellanger A, Hanafi S, Wilbaut C. Three-stage hybrid-flow-
46(5):65066. shop model for cross-docking. Computers & Operations
10. Arabani AB, Ghomi SF, Zandieh M. A multi-criteria Research. 2013;40(4):110921.
cross-docking scheduling with just-in-time approach.The 15. Buijs P, Vis IF, Carlo HJ. Synchronization in cross-docking
International Journal of Advanced Manufacturing networks: A research classification and framework.
Technology.2010; 49(5-8):74156. European Journal of Operational Research. 2014;
11. Arabani AB, Ghomi SF, Zandieh M. Meta-heuristics 239(3):593608.
implementation for scheduling of trucks in a cross-dock- 16. Miao Z, Cai S, Xu D. Applying an adaptive tabu search algo-
ing system with temporary storage. Expert systems with rithm to optimize truck-dock assignment in the cross-dock
Applications. 2011;38(3):196479. management system. Expert Systems with Applications.
12. Alpan G, Larbi R, Penz B. A bounded dynamic pro- 2014;41(1):1622.
gramming approach to schedule operations in a cross 17. Eshghi K, Kariminasab M. Combinatorial Optimization &
docking platform. Computers & Industrial Engineering. Metaheuristics. Tehran: MehrAzinPress; 2012. p. 12538.
2011;60(3):38596. 18. Lobo FG, Goldberg DE. Decision making in a hybrid genetic
13. Agustina D, Lee CKM, Piplani R. Cross-docking schedul- algorithm. IEEE International Conference onEvolutionary
ing with delivery time window and temporary storage. 2011 Computation. 1997; 8:1215.

10 Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology