Professional Documents
Culture Documents
1
2
PanickerV.V. andSreeramK.Y.
distribution cost. Total weighted tardiness is summation of the product of tardiness and
weight of each order and total distribution cost is the summation of the distribution cost of the
vehicle to which orders are assigned. Initially a mathematical model is adopted from Cakici
et al. 2011 and solve that model using LINGO 11.0 an optimization modeling software for
linear, nonlinear and integer programming. Global optimum solution is obtained for number
of orders is quite less. Whenever the order size increases because of its solution method is
Branch and Bound algorithm in which effort requires grows exponentially with problem size
it is unable to give optimal solution in reasonable time. To solve the problems in which
number of orders is high, Clonal Selection Algorithm from the suite of artificial immune
systems algorithms that is inspired by the Clonal selection theory of acquired immunity is
used. The results obtained from the algorithm are compared with the results given by LINGO
11.0.
This paper focuses on the development of a solution for multi-objective productiondistribution system. The remaining part of this paper is organized as follows. Section 2
provides a literature review of the recent work on production-distribution systems. Section 3
presents the solution methodology using both LINGO and Clonal selection algorithm. Section
4 presents the results and discussion. The paper is concluded in Section 5 with limitations of
study and further scope.
2. LITERATURE REVIEW
Production distribution problem considered in this paper is an NP hard problem. This
problem is formulated as a mixed integer linear programming problem. But exact methods
are not used for their inefficient and expensive computations. Several researchers have
studied the decision problems in production-distribution systems. Cakici et al. 2011 consider
minimization of total distribution costs and total weighted tardiness as objective function. A
pareto optimal solution method using a non-dominated sorting genetic algorithm is applied to
get an optimal solution for the multi-objective optimization problem. Steinrcke, 2011
presents minimization of production cost, transportation cost, and bonus payments. A mixedinteger decision-making model by relaxing and/or fixing is used to formulate and solve the
multi objective optimization problem.
Pundoor and Chen, 2005 consider minimization of maximum delivery tardiness and
total distribution cost as the objective function. A fast heuristic is developed for solving the
multi objective problem. Hall and Potts, 2003 analyze minimization of total flow time and
total distribution cost. A dynamic programming algorithm is used to solve the multi objective
problem. Vanbuer et al. 1999 consider production distribution problem with minimization of
cost of owning trucks and operating costs. The problem is solved using heuristic search
algorithms.
This paper considers a production-distribution problem described in Cakici et al, 2011
and further extended to routing and heterogeneous fleet of vehicles. Exact methods are,
however, generally not very useful when a problem complexity is high. If heuristic methods
have been proposed the major disadvantage of heuristic methods is the possibility of
terminating at a local optimum that is far distant from the global optimum. A meta-heuristic
method clonal selection algorithm is proposed in this paper. Many authors used these
artificial immune systems algorithms for solving operational decision problems. Khalid et al,
2013 uses AIS to solve fixed charge transportation problem by considering distribution cost
as their objective. Agarwal et al, 2003 uses clonal selection algorithm for solving resource
constraint project scheduling problem by considering only one objective makespan. Engin
and Doyen, 2005 used clonal selection algorithm to solve hybrid flow shop scheduling
problem by considering makespan as objective. Bagheri et al, 2010 uses artificial immune
systems algorithm as a solution methodology for flexible job shop scheduling problem.
Collelo and Cortes, 2005 uses artificial immune systems algorithm to solve a multi objective
optimization problem. In the present work the ideas for solving multi-objective optimization
problem are adapted from the previous discussed papers.
3. SOLUTION METHODOLOGY
Initially a mathematical model is adopted from Cakici et al. (2011). The model is
solved using an optimization modeling software for linear, nonlinear and integer
programming, LINGO 11.0. Global optimum solution is obtained for small size problems
with the number of orders being less. Since LINGO software applies the solution method of
branch and bound algorithm, as the order size increases the computation time grows
exponentially with problem size. Hence, it is unable to provide an optimal solution in
reasonable time (Sreeram et al. 2013).
In this work, for medium and large size problems with more number of orders, a
Clonal Selection Algorithm from the suite of artificial immune systems algorithms is applied.
The algorithm is inspired from the Clonal selection theory of acquired immunity. Few ideas
from Coello and Cortes (2002) on the Clonal Selection Algorithm have been extended to the
multi-objective optimization in this work. For validation, the solutions obtained from the
algorithm are compared with that given by LINGO 11.0.
3.1 Problem Environment
In this work, an integrated production distribution problem with one manufacturer and
more than one customer is considered. A customer is allowed to place more than one order.
The orders are received by a manufacturer and processed on a single production line, and
delivered to customers by capacitated homogenous fleet of vehicles based on number of
orders given by each customer. Whenever one customer places more than one order, the
manufacturer ships those orders in one capacitated vehicle. There exist an infinite number of
vehicles and a variable transportation cost incurs for each delivery based on transportation
time. Each order is associated with a customer, weight (penalty for tardy jobs), processing
time, due time, and size (volume or storage space required in the transportation unit). The
vehicle capacity is defined as the maximum total size of the jobs that can be delivered
together. Transportation times are also considered in addition to the processing times.
The objectives considered in this work are conflicting in nature i.e., optimizing one objective
leads to worsen the other. By delivering each order using one capacitated vehicle minimizes
the total weighted tardiness and at the same time number of vehicles used for distributing
goods to customer increases which leads to increase distribution cost. On other hand one
capacitated vehicle is used to distribute goods to more than one customer total distribution
cost objective decreases but tardiness of each order increases eventually total weighted
tardiness objective gets worsen. In order to make a better compromise between these two
objectives Pareto optimality is used and developed diverse set of solutions. The Selection of
solution is depends on high-level information available with decision maker and conditions
prevailing in decision makers environment.
3.1.1 Assumptions
These are assumptions considered at the time of solving the above problem. Out of
which first few scheduling assumptions are taken from Pinedo (2002). These are considered
only to subside the complexity of problem without loss of generality. Those are
1.
2.
3.
4.
5.
6.
7.
(3)
Ab{d}
Ab{n}
(2d)
Re-select
(2c)
Ab
(1)
(2a)
Select
Ab{n}
(2b)
C*
Clone
Maturate
Table 3.2 New antibodies possible from initially random generated ones
Tardiness representation
After generating all possible combinations of two halves of antibodies calculate both the
objectives.
3.2.3 Cloning or Proliferation:
It is the selection of high affinity antibodies from the pool generated randomly and
clone each antibody yi number of times depends on its rank. The value of y i can be obtained
from the relation given below
N
yi
0.5
i
Where,
multiplying factor,
N
number of antibodies,
i
rank of antibody,
yi
number of clones generated by ith rank antibody.
Total number of clones generated will be given by the following
N
N
Nc
0.5
i
i1
Select the first Nc non-dominated antibodies for the cloning. Before selecting them from the
pool of antibodies first delete the repeated antibodies and/or delete the antibodies pertaining
to the same objective function values if found. Otherwise we cannot get any non-dominated
antibodies because the non-dominated sorting of antibodies is done with respect to the
objective function values if objectives of the pairs of antibodies are same those are neither
dominate nor non-dominate each other.
If the number of non-dominated antibodies of rank 1 is less than Nc then consider the
non-dominated antibodies of lowest ranks till the number of antibodies is equal to Nc. All
these Nc number of antibodies are added to the memory. Rank the antibodies present in the
memory with respect to the objective function values then clone the each antibody y i number
of times. Thereafter affinity maturation is performed on all cloned antibodies.
3.2.4 Hypermutation:
It is the process in which one or more elements in the antibody are changed from its previous
form there by the objective function value will change. From the mutation the solutions,
which are present near to the variable space are traced. In the present work five types of
mutation processes are used because it is difficult to get minimum cost antibody from the
cloned one.
Five types of mutation processes used are
1. Inverted mutation
2. Pair wise exchange mutation
3. Insertion mutation
4. Displacement mutation
5. Shift mutation
In all the above mutation processes tardiness representation and distribution cost
representation are mutated separately. If both are mutated together in any mutation process
the elements from one representation are replaced by the elements of other representation
then antibody loses its feasibility. All these mutation process are done for each cloned
antibody. A common method of preforming mutation is by generating random numbers. Here
the random number generation with discrete uniform distribution between 1 and N in case of
tardiness representation and between 1 and mx in case of distribution cost representation.
Because of mutation if the feasibility of antibody is lost then making the antibody feasible by
take one order from the overloaded vehicle and assign it to a vehicle that is not assigned yet
and of least cost among the vehicles not assigned. This procedure is continued till all the
Attribute
Penalty of orders (weight)
Volume of orders (size)
Transportation time from manufacturer to customer
Processing time of order
Vehicle capacity
Input generation
DU [1,10]
DU [1,25]
DU [1,100]
DU [1,10]
DU [50,150]
10
Weight
10
Processing times
Transportation time
between manufacturer
and customer
45
68
64
31
63
63
43
80
62
75
Due date
50
39
18
38
94
48
72
37
19
Volume of order
10
18
19
21
23
13
15
10
Vehicle capacity
80
89
77
101
81
58
147
148
107
57
Solution for the above-considered example by considering both vehicle routing and
heterogeneous fleet of vehicles are for 20 problem instances. The problem without
considering routing is used so as to validate the clonal selection algorithm developed for this
problem and verify whether it is working correctly or not, in comparison with LINGO 11.0.
Later the same problem with routing only and both heterogeneous fleet of vehicles and
routing are considered. The objective values by using scalarization method is compared with
without routing problem. The objective function values for the production distribution
problem solved by using LINGO 11.0 without considering routing and heterogeneous fleet of
vehicles and by considering the routing and heterogeneous fleet of vehicles and solved by
Clonal selection algorithm are tabulated in table 4.4
Table 4.4: Comparison of objective function value between Lingo and Algorithm
Sl.
no.
Number
of
orders
1
2
3
4
5
6
7
8
9
10
5
6
7
8
9
10
11
12
13
14
Objective
function
value using
LINGO 11.0
1715
1403
2480
3550
2618
3361
3025
3164
3803
4442
Objective
function
value
using
Algorithm
1072
1039
1824
2815
1786
2032
1685
2482
2374
3166
Sl.
no.
Number
of
orders
11
12
13
14
15
16
17
18
19
20
15
16
17
18
19
20
21
22
23
24
Objective
function
value using
LINGO 11.0
4026
3337
4676
6233
4097
5653
4246
6850
8707
7887
Objective
function
value
using
Algorith
m
2950
2198
3193
4568
1832
4126
3152
4075
5140
5340
The production distribution problem with routing and heterogeneous fleet of vehicles solved
by using CLONALG is compared with the same problem with homogeneous fleet that is
solved by using LINGO tabulated in shown in table 4.4. The problem is tested for objective
function value and computational times taken to reach convergence are plotted.
In the Figure 4.1 the objective function values of production distribution problem by
homogeneous and heterogeneous fleet of vehicles by considering routing is plotted for 20
problem instances. From the figure it is concluded there is significant difference in objective
function value between both the cases.
Figure 4.1: Comparing the objective function values for homogeneous (LINGO) and heterogeneous fleet of
vehicles
From the objective function values given in Table 6.12 for 20 problem instances generated
using Table 4.1.
Figure 4.2 shows how the computational time required to reach the convergence is
varying with the increase in number of orders. From the figure it is concluded that the amount
of computational required reaching convergence shows an increasing trend. The maximum
amount of time required is more than the maximum amount of time required in last two
cases.
1
2
3
4
5
6
7
8
9
10
11
12
13
1 2 5 7 9 8 4 6 3 10 3 0 3 0 1 2 0 0 0 1
1 2 5 7 9 8 4 6 3 10 0 0 6 0 0 1 0 0 0 3
1 2 5 7 9 8 4 6 3 10 3 0 3 0 0 2 0 0 0 2
1 2 5 7 9 8 4 6 3 10 0 0 3 0 0 4 0 0 0 3
5 4 3 8 7 9 1 10 2 6 1 0 3 0 0 4 0 0 0 2
4 8 7 10 9 6 1 2 5 30 0 2 0 0 4 0 0 0 4
7 9 8 4 5 10 1 3 6 20 0 3 0 0 3 0 0 0 4
5 7 4 9 8 10 2 6 3 1 1 1 1 2 1 0 1 2 1 0
5 4 8 7 9 1 2 10 6 31 2 2 0 2 1 0 0 0 2
5 4 8 7 9 1 2 10 6 3 1 1 1 1 1 1 1 1 1 1
5 4 8 7 9 1 2 10 6 3 1 2 0 1 2 1 0 1 2 0
5 4 8 7 9 1 2 10 6 3 1 2 0 2 1 0 1 1 2 0
4 3 1 5 2 6 7 10 9 8 0 0 2 0 0 4 0 0 0 4
Affinity values
TWT
TDC
3093.0
731.0
4739.0
520.0
3408.0
665.0
4291.0
530.0
3158.0
699.0
3500.0
597.0
3752.0
556.0
2431.0
1312.0
2562.0
871.0
2382.0
1539.0
2512.0
1154.0
2462.0
1212.0
3900.0
534.0
The curve generated using the above Pareto optimal solutions is called Pareto optimal front.
Pareto optimal solutions obtained from CLONALG is shown in Figure 4.3 and Pareto optimal
front is shown in Figure 4.4.
One manufacturer receives multiple orders but there can be multiple customers
with multiple manufacturers.
In this study, all jobs manufactured by the manufacturer is assumed to be of same
size but there is a possibility of considering an effective unit size.
One vehicle is used only once, if the use of that vehicle is advantage
References:
Agarwal, R., Tiwari, M.K., and Mukherjee, S.K., 2007. Artificial immune system
based approach for solving resource constraint project scheduling problem.
International Journal of Advanced Manufacturing Technology, 34, 584-593.
Bagheri, A., Zandieh, M., Iraj M., and Yazdani, M., 2010. An artificial immune
algorithm for the flexible job-shop scheduling problem. Future Generation
Computer Systems, 26, 533-541.
Cakici, E., Mason, S. J., and Kurz, M. E., 2011. Multi-objective analysis of an
integratedsupply chain scheduling problem. International Journal of Production
Research, 50 (10), 26242638.
Choi, E., Tcha, D.W., 2007. A column generation approach to the heterogeneous fleet
vehicle routing. Computers & Operations Research, 34, 20802095.
Coello, C.A.C., Rivera, D.C., and Cortes, N.C., 2003. Use of an artificial immune
system for job shop scheduling. Lecture Notes in Computer Science, 2787, 110.
Coello Coello, C. A., and Cortes, N. C., 2005. Solving multi-objective
optimization problems using an artificial immune system. Genetic Programming
and Evolvable Machines, 6, 163190.
Dasgupta, D., and Nino, L. F., 2009, Immunological computation: Theory and
applications, CRC press.
Deb, K., 2001. Multiobjective optimization using evolutionary algorithms,
Chichester,UK,Wiley.
De castro, L. N. and Timmis, J., 2002. An Introduction to Artificial Immune Systems:
A New Computational Intelligence Paradigm, Springer-Verlag.
De castro, L.N., Von Zuben, F., 2000, The clonal selection algorithm with engineering
applications. In: Proceedings of Workshop on Articial Immune Systems and their
Applications, GECCO, 3637.
Deb, K., 2001. Multi-objective optimization using evolutionary algorithms,
Chichester, UK, Wiley.
Hall, N.G. and Potts, C.N., 2005. The coordination of scheduling and batch
deliveries. Annals of Operations Research, 135, 4164.
Kuby, J., R. A. Goldsby, T. J. Kindt and B. A. Osborne. Immunology. Sixth
Edition, W. H. Freeman, San Francisco, CA, 2006.
Pinedo, M., 2002. Scheduling: Theory, algorithms, and systems. Engelwood
Cliffs, NJ: Prentice Hall.
Pundoor, G. and Chen, Z.L., 2005. Scheduling a production-distribution system to
optimize the tradeoff between delivery tardiness and distribution cost. Naval
Research Logistics, 52, 571-589.
Steinrucke, M., 2011. An approach to integrate production-transportation planning
and scheduling in aluminum supply chain network. International Journal of
Production Research, 49 (21), 65596583.
Van Buer, M. G., Woodruff, D.L., and Olson, R. T., 1999. Solving the medium
newspaper production/distribution problem. European Journal of Operational
Research, 115(2), 237-253.
Wassan,N.A.andOsman,I.H.,2002.Tabusearchvariantsforthemixfleetvehicle
routingproblem,JournaloftheOperationalResearchSociety,53,76882.
16
17
18