Computers & Operations Research 64 (2015) 225–232

Contents lists available at ScienceDirect

Computers & Operations Research

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

A new approach to solve the multi-product multi-period inventory lot

sizing with supplier selection problem
Leopoldo Eduardo Cárdenas-Barrón a,n, José Luis González-Velarde a,
Gerardo Treviño-Garza b
a
School of Engineering and Sciences, Tecnológico de Monterrey, E. Garza Sada 2501 Sur, Monterrey C.P. 64849, Nuevo León, Mexico
b
BNSF Railway Company, 2650 Lou Menk Drive, Fort Worth, TX 76131-2830, USA

art ic l e i nf o a b s t r a c t

Available online 23 June 2015 This research work deals with the multi-product multi-period inventory lot sizing with supplier
Keywords: selection problem. Formerly, this kind of problem was formulated and solved using an exhaustive
Inventory enumeration algorithm and a heuristic algorithm. In this paper, a new algorithm based on a reduce and
Lot sizing optimize approach and a new valid inequality is proposed to solve the multi-product multi-period
Multi-period inventory lot sizing with supplier selection problem. Numerical experiments ratify the success of the
Multi-products proposed heuristic algorithm. For the set of 150 benchmark instances, including 75 small-sized
Supplier selection instances, 30 medium-sized instances, and 45 large-sized instances, the algorithm always obtained
Mixed integer linear programming better solutions compared with those previously published. Furthermore, according to the computa-
Reduced costs
tional results, the developed heuristic algorithm outperforms the CPLEX MIP solver in both solution
Reduce and optimize approach (ROA)
quality and computational time.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction known heuristic rules, the lot sizing problem has been also solved
using other approaches, i.e. Hop and Tabucanon [10] have developed a
Supplier selection and lot sizing are essential activities in new and original approach to solve the lot sizing problem using an
supply chain management. Typically, in any echelon of a supply adaptive genetic algorithm. Later, Cárdenas-Barrón [4] discusses some
chain it is important to select the best suppliers from which to features of the adaptive genetic algorithm in Hop and Tabucanon [10].
acquire the products, the lot sizes, and the time to place the orders He concludes that it is convenient to solve the lot sizing problem with
during a finite planning horizon. Wagner and Whitin [23] algorithm because this always obtains the
The single-product multi-period inventory lot sizing problem has its optimal solution. In addition, Jans and Degraeve [11] have provided a
origins at the end of the 1950s and was first proposed by Wagner and comprehensive and complete review of metaheuristics for the lot
Whitin [23]. This problem can be represented as a mixed integer linear sizing problem. One can see from the reviews mentioned above that
programming (MILP) model. Moreover, Zangwill [30] has shown that there exist many approaches to solve the lot sizing problem that have
this problem is a fixed charge network problem. Wagner and Whitin been developed for distinct applications.
[23] have solved this type of problem optimally with a dynamic Another interesting research direction is on the inventory lot
programming algorithm. Furthermore, several dynamic lot size rules sizing with supplier selection problem which combines lot sizing
have been proposed to solve the single-product multi-period inventory and supplier choice decisions. The research works of Kasilingam
lot sizing problem. For instance, see Simpson [18] who conducted a and Lee [14] and Jayaraman et al. [12] address this problem. Both
scrutinized and comprehensive study of nine well-known published papers contain mixed integer programming models to select
rules. Since its introduction, it became one of the most studied and suppliers and determine the lot size of the products. Also, Dahel
extended problems in inventory management. A number of excellent [7] develops a multi objective mixed integer programming
reviews have been published in order to provide an introduction to the approach to select the number of suppliers to use and the lot size
lot sizing problem and its extensions. For example, De Bodt et al. [8], of each product to place the orders to suppliers for a multiproduct,
Bahl et al. [2], Kuik et al. [15] and Wolsey [29] are the first reviews on multi-supplier competitive sourcing environment.
the history of the single product lot sizing problem. In addition to well- Basnet and Leung [3] deal with the multi-product multi-period
inventory lot sizing with supplier selection problem. First they solved
the problem with an exhaustive enumerative search algorithm. Since
n
Corresponding author. Tel.: þ 52 81 83284235; fax: þ 52 81 83284153. this algorithm cannot obtain a solution within 2 h of computation for
E-mail address: [email protected] (L.E. Cárdenas-Barrón). instances with 60 or more binary variables, they proposed a heuristic

http://dx.doi.org/10.1016/j.cor.2015.06.008
0305-0548/& 2015 Elsevier Ltd. All rights reserved.
226 L.E. Cárdenas-Barrón et al. / Computers & Operations Research 64 (2015) 225–232

algorithm to solve the problem in a reasonable computational time. optimization problem by solving it on a small set of variables. Within
On the other hand, Moghadam et al. [16] propose a new mathema- this context, the ROA approach always solves the problem over a small
tical formulation for the multi-period inventory lot sizing problem feasible space that contains a near optimal solution. Treviño-Garza [19]
with supplier selection and solve it using a hybrid intelligent algo- has proved that the ROA is capable of solving other types of binary
rithm based on fuzzy neural networks and genetic algorithms. integer problems (i.e. single machine total weighted tardiness problem,
Aissaoui et al. [1] present an excellent and comprehensive review set covering problem and set partitioning problem) in a reasonable
on supplier selection and lot sizing. They mainly focus on papers that time. The noteworthy results achieved by Treviño-Garza [19] have
deal with the determination of the best combination of suppliers and encouraged us to apply ROA to multi-product multi-period inventory
allocation of the orders in order to cover different purchasing lot sizing with supplier selection problem in the present research
requirements. In the same year, Wadhwa and Ravindran [22] model article.
the supplier selection problem as a multi-objective optimization The rest of this paper is organized as follows. Section 2 presents
problem in which one or more purchasers order multiple products the formal definition of the multi-product multi-period inventory
from different suppliers in a multiple sourcing network. Later, Ustun lot sizing with supplier selection problem and its mathematical
and Demirtas [21] propose a model that integrates the well-known formulation. Section 3 proposes a heuristic algorithm to solve the
analytic network process (ANP) and achievement scalarizing func- multi-product multi-period inventory lot sizing with the supplier
tions to select the best suppliers and establish the optimal lot sizes selection problem and also provides the results of the extensive
between the chosen suppliers by considering tangible–intangible computational experiments. Finally, Section 4 gives some conclu-
criteria and time horizon. Afterwards, Ho et al. [9] review the sions and future research directions.
literature related to the multi-criteria decision making approaches
for supplier selection and evaluation. Tsai et al. [20] develop an
attribute-based ant colony system (AACS) which is a framework to 2. The multi-product multi-period inventory lot sizing with
study the critical factors to select the best suppliers for purchasing supplier selection problem
decisions. Ware et al. [24] present an exhaustive and well documen-
ted state of the art literature review. Basically, Ware et al. [24] critique 2.1. Problem definition
the papers that deal with the supplier selection problem. Afterwards,
Ruiz-Torres et al. [17] propose a mathematical optimization model to Basnet and Leung [3] formally define the multi-product multi-
determine the optimal demand allocation over a set of suppliers period inventory lot sizing with supplier selection problem as follows.
considering the risk of supplier failures. Ware et al. [25] study the Consider the situation where the deterministic demand of multiple
interrelationship of supplier selection criteria using the interpretive discrete products is known over a given finite planning horizon. Each
structural modeling (ISM) approach. Choudhary and Shankar [5] product can be supplied from a set of suppliers, for example one or
propose an integer linear programming optimization model to jointly more suppliers could be chosen in each period for the procurement of
determine the timing of procurement, lot-sizes, and the selection of a product. There is a supplier ordering cost that is incurred in each
suppliers and carriers in order to minimize the total cost thru a finite period when an order is placed to the supplier. Also, there is a product
planning horizon. In a subsequent paper, Choudhary and Shankar [6] holding cost per period that applies to each product in inventory
develop a multi-objective integer linear programming model to make when it is carried through a period in the finite planning horizon.
the best decisions on inventory lot-sizing, supplier selection, and There are no capacity constraints and shortages are not allowed. The
carrier selection. Ware et al. [26] propose a methodology for the decision maker (manager) seeks to decide what products to buy in
flexible supplier selection problem in which both qualitative and what lot sizes from which suppliers and in which periods to place the
quantitative factors are considered jointly. They use the analytical orders. Basically, Basnet and Leung [3] propose a mixed integer linear
hierarchy process (AHP) and interpretive ranking process (IRP) for the programming (MILP) formulation which is presented in Section 2.2.
qualitative model, and a mixed non-integer linear programming
problem for the quantitative model. In the same year, Ware et al. 2.2. Mixed integer linear programing model (MILP) for the multi-
[27] analyze the impact of demand variation on a multi-product, product multi-period inventory lot sizing with supplier selection
multi-source, multi-period model for the supplier selection problem. problem
Ware et al. [28] develop a mixed-integer non-linear program to
address the dynamic supplier selection problem. More recently, Basnet and Leung [3] propose the following MILP formulation for
Karsak and Dursun [13] present a fuzzy multi-criteria group decision the multi-product multi-period inventory lot sizing with supplier
making methodology which uses the well-known quality function selection problem (the notation and the mathematical model are
deployment (QFD), and a fusion of fuzzy information and a 2-tuple placed here just for self-completeness of this paper).
linguistic representation model for supplier selection. Indices
This research paper mainly improves the work of Basnet and
Leung [3]. It is of fundamental importance to mention that the  i ¼1,2,3,…, I index for products.
inventory lot sizing with supplier selection problem in Basnet and  j ¼1,2,3,…, J index for suppliers.
Leung [3] is a complex combinatorial NP hard optimization problem.  t ¼1,2,3,…, T index for periods.
Heuristics or approximation algorithms play a vital role in solving NP
hard problems with the hope that a near optimal solution in a short Parameters
amount of time can be found. Consequently, this research presents a
heuristic algorithm that provides a quality solution for the multi-  Dit ¼demand for product i in period t.
product multi-period inventory lot sizing with supplier selection  Pij ¼purchase price for product i from supplier j.
problem in a reasonable amount of computational time.  Hi ¼holding cost for product i per period.
Basnet and Leung [3] have left as a fertile area for future research  Oj ¼ ordering cost for supplier j.
the improvement of their solutions. Therefore the main goal of this
research paper is to propose a new approach to solve the multi-product Decision variables
multi-period inventory lot sizing with supplier selection problem
heuristically. This new approach is based on the reduce and optimize  Xijt ¼ lot size for product i ordered from supplier j in period t.
approach (ROA). It is possible to accelerate the solution process of an  Yjt ¼1 if an order is placed to supplier j in period t, 0 otherwise.
 L.E. Cárdenas-Barrón et al. / Computers & Operations Research 64 (2015) 225–232 227

Mixed integer linear programming (MILP) formulation feasible region based on the original problem and optimizing it.
Then, the problem can be formulated as follows: Although, in most of the cases, ROA does not ensure optimality, it
Model: can be used because it obtains near optimal solutions in a
0 ! reasonable time. Nonetheless, if the optimal solution of the
XXX XX XX X
t X X
t
Min P ij X ijt þ Oj Y jt þ Hi @ X ijk  Dik problem at hand is feasible for the reduced problem then ROA
t j i j t i t k¼1 j k¼1 guarantees the optimal solution. Fundamentally, ROA is laid out in
ð1Þ the following statement:
Assume that a mathematical formulation has n binary variables
Subject to :
(i.e. y1,y2,…,yn) and BC is the set of all binary variables of the whole
t X
X X
t mathematical formulation. Let BO be the set that has the binary
X ijk  Dik Z 0 8i ; t ð2Þ
variables that are equal to 1 at the unknown optimal solution. Let
k¼1 j k¼1
BR be a reduced set that contains a portion of the binary variables.
! Then it is required that the before mentioned sets satisfy the
X
T
Dik Y jt X ijt Z 0 8 i; j ; t ð3Þ following condition: BO D BR DBC.
k¼t Clearly, the challenge of ROA is to obtain a set BR such that BR is
a subset of BC and further BR also covers the set BO. Then ROA
Y jt ¼ f1; 0g 8j ; t ð4Þ repeatedly optimizes the problem over a reduced feasible space BR
using a commercial MIP solver (i.e. CPLEX). In other words, the
X ijt Z0 8 i; j ; t ð5Þ
problem is solved after fixing the binary variables not in set BR
The objective function (1) is the total cost that is incurred by equal to 0 (i.e. yi ¼0 if and only if yi A BC-BR).
the buyer and comprises of the total purchase cost of the products, It is crucial to mention that there are some situations in which
the total ordering cost, and the total holding cost for carrying ROA does not find good solutions. These situations are: (1) when
inventory in each period. Constraints (2) ensure that all demand is the set BR contains only a portion of set BO and (2) when the sets
satisfied in the period in which it occurs; these constraints also BR and BO are disjoint. Furthermore, it is worth mentioning that
guarantee avoiding shortages. Constraints (3) establish that is not the set BO with the optimal solution is always unidentified.
possible to place an order without charging a corresponding Conversely, in ROA it is required that the set BR be of small size
ordering cost. Constraints (4) define the binary variables. Con- such that the optimization problem can be handled by any
straints (5) impose non-negativity conditions on the remaining commercial optimization software (i.e. CPLEX) within a reasonable
decision variables. This mathematical formulation (1)–(5) is very amount of computational time. Likewise, finding a practical size for
difficult to solve since it is an NP hard problem. Furthermore, the the set BR that includes the binary variables that are equal to 1 in
problem becomes unsolvable when the instances are of large size. the optimal solution is the most challenging task in ROA. Therefore,
It is important to remark that the mathematical formulation (1)– the first set BR where ROA initiates is sometimes not as good as
(5) can only be solved to optimality by commercial integer linear desired. Therefore the set BR needs to be improved. Basically, ROA
programming solvers for very small instance sizes. For instances of consists of four phases. The first phase of ROA constructs a feasible
large size that are commonly found in the real world, the problem reduced set of variables (initial set BR). The second phase optimizes
cannot be solved optimally. the mathematical model within the reduced set of variables. The
third phase selects new variables to be included in the reduced set
of variables. The fourth phase refines the reduced mathematical
3. Solving the multi-product multi-period inventory lot sizing model and then goes to the second phase.
with supplier selection problem based on the reduce and Here, there are some ways to construct the initial set BR:
optimize approach (ROA)
I. create the initial set BR randomly,
This section presents a heuristic algorithm for solving the II. solve the full complete model of the problem for a short time
mixed integer linear programming (MILP) for the multi-product and then use its incumbent solution
multi-period inventory lot sizing with supplier selection problem, (i.e. the variables equal to 1 in the incumbent solution form the
along with, the computational results for the 150 instances solved set BR),
in Basnet and Leung [3]. III. solve an LP-relaxation of the problem and use its optimal
One way to strengthen the mathematical formulation is to add solution
valid inequalities to the model. For that reason, the mathematical (i.e. the variables equal to 1 in the optimal LP-relaxation
formulation (1)–(5) just presented above is further reinforced by solution form the set BR),
P
including the following valid inequality (6): j Y jt r J ; 8 t A T. IV. solve the problem with a heuristic to construct the initial set
This inequality imposes a boundary on the total of number of BR,
suppliers used in any period t. Even though the valid inequality V. or with a combination of the above ways.
(6) is redundant to the mathematical formulation, this is valuable
in aiding CPLEX to create some new cuts. In this paper option IV is preferred using the Wagner–Whitin
It is important to mention that Basnet and Leung [3] do not [23] algorithm. The third phase of ROA is to enhance the set BR
provide any valid inequality because their solution approaches do iteratively. This task can be performed by taking information from
not require this. It was detected that the valid inequality (6) has a the solution of the reduced problem previously solved. Here, it is
positive effect on the optimization process, particularly on the proposed to obtain the reduced costs of the variables that are not
following two issues: the solution quality of the instances and the in set BR in order to enhance BR. The problem is then re-optimized
computational time. over BR until a stop condition is met or when no more negative
reduced costs (in a minimization problem) are found.
3.1. Reduce and optimize approach (ROA) It is important to mention that the reduced costs of binary
variables cannot be determined. Nevertheless, an LP-relaxation
Here, the Reduce and Optimize Approach (ROA) is explained in can be solved easily and then the reduced costs of the variables
a brief manner as follows. ROA is based on establishing a reduced that are not in the set BR can be obtained. Afterwards, the variables
228 L.E. Cárdenas-Barrón et al. / Computers & Operations Research 64 (2015) 225–232

with negative reduced costs need to be included in the set BR. In It is important to point out that the formulation models always
order to prevent that set BR reach an unmanageable size giving its contain constraints (2)–(5) and the proposed valid inequality (6).
continuous growth in each improvement iteration, the variables The solution procedure is described by the pseudo-code shown in
within set BR with a reduced cost greater than or equal to zero are the following algorithm:
dropped from the set. Additionally, a maximum of number of
variables to be added to set BR is also established.

It is worth mentioning that several popular algorithms exist that The above heuristic algorithm works as follows.
are used to solve problems similar to the multi-product multi-period
inventory lot sizing with supplier selection problem. Some of them – Step 1. The initial set is constructed applying the Wagner–
are based on genetic algorithms (GA), particle swarm optimization Whitin [23] algorithm. In this step the objective is to obtain
(PSO), ant colony optimization (ACO), Scatter Search (SS), Differential from which supplier j and in which period t the product must
Evolution (DE), among others. These algorithms have a population of be purchased, and there is no need to determine the lot sizes. In
candidate solutions and they improve the set of solutions iteratively. other words, the Wagner–Whitin [23] algorithm is used simply
In contrast, the reduce and optimize approach (ROA) always considers to determine which binary variables (i.e. Yjt) must be included
a small set of binary variables and optimizes the problem over this in the initial set BR.
set. This is the main difference between other popular algorithms (GA, – Step 2. Starting with an initial feasible set BR for the optimiza-
PSO, ACO, SS, DE) and ROA. tion problem the RedMILP model is solved for β time units. If
To the best of our knowledge this paper is the first to apply the clock o τ then go to Step 3. Otherwise the algorithm stops and
ROA to the inventory lot sizing with supplier selection problem. returns the incumbent solution.
– Step 3. Once the RedMILP is solved the value of each binary
3.2. Proposed heuristic algorithm based on ROA variable (Ysjt) of the incumbent solution is determined and will
be used in Step 4.
In this subsection, a heuristic algorithm is proposed. The following
– Step 4. With the values of all binary variables the continuous
abbreviations: RedMILP, RelaxMILP, max-nva, nvnrc, LBvar and UBvar
variables set is constructed setting the lower and upper bounds
are used in the heuristic algorithm. RedMILP is the MILP model but
to the corresponding values obtained in Step 3.
when the optimization problem is solved over a reduced set of binary
– Step 5. Here, a surrogate problem called RelaxMILP is solved.
variables (Yjt). RelaxMILP is a relaxed MILP model in which all binary
This action refers to a tactic used to determine the reduced
variables are relaxed. The abbreviation max-nva refers to the max-
costs for each variable defined in Step 4.
imum number of variables to add to the set BR. The abbreviation
nvnrc represents the number of variables with negative reduced cost. – Step 6. In this step, if there are no more variables with negative
LBvar and UBvar denote lower bound and upper bound for the decision reduced costs to add to BR then the algorithm stops and returns
variable Yjt. Also, YWW represents the solution found by the Wagner– the incumbent solution. Otherwise, the set BR can be enhanced
jt
Whitin [23] algorithm. The superscript WW corresponds to the initials in Step 7.
of Wagner–Whitin. The superscript S in Ysjt refers to “solution” and the – Step 7. Now, the set BR is improved as follows: If nvnrcomax-
superscript C in Ycjt corresponds to “continuous”; this is when the nva then add the nvnrc variables to set BR. Else add the max-
binary variable is relaxed to take continuous values. Finally, RCYjt nva variables with most negative reduced cost to set BR. Return
represents the reduced cost of variable Ycjt. to Step 2.
 L.E. Cárdenas-Barrón et al. / Computers & Operations Research 64 (2015) 225–232 229

Table 1
Characteristics of the instances tested.

Category Size of the instance l,m,n Binary variables Yjt Continuous variables Xijt Constraints including proposed valid inequality

Small 3,3,10 15 45 65
3,3,15 45 135 195
4,4,10 40 160 210
4,4,15 60 240 315
5,5,20 100 500 620
Medium 10,10,50 500 5,000 5,550
15,15,100 1,500 22,500 24,100
Large 20,20,100 2,000 40,000 42,100
20,20,200 4,000 80,000 84,200
50,50,200 10,000 500,000 510,200

Table 2
Results of the computational experimentation.

Category Size of the Total average cost Basnet and Total average cost proposed Difference Percent of Average No. of optimal Average GAP
instance Leung [3] Algorithm heuristic algorithm improvement time (min) solutions %

Small 3,3,10 102,584 101,953.6 630.4 0.614520783 0.009757778 14/15 0.199984542

3,3,15 147,887 147,162.9 724.0667 0.489608057 0.089163333 15/15 0
4,4,10 126,345 124,526 1,819 1.439708734 0.016915556 15/15 0
4,4,15 187,320 185,699.3 1,620.667 0.865186134 3.440541111 14/15 0.179920572
5,5,20 303,529 300,900.2 2,628.8 0.866078694 4.40799 12/15 0.071243571
*
Medium 10,10,50 1,357,190 1,337,662 19,527.87 1.438845458 30 0.854599987
*
15,15,100 3,856,800 3,810,899 45,900.87 1.190128258 30 2.068323929
*
Large 20,20,100 5,048,826 4,975,149 73,677.47 1.459298987 30 3.861037554
*
20,20,200 10,026,074 9,915,138 110,935.5 1.106470323 30 9.180299397
*
50,50,200 25,373,121 23,534,629 1838,492 7.245823904 30 6.993766063

n
In these instances CPLEX cannot prove optimality.

Table 3 reported by Basnet and Leung [3]. The computational results are
Results of computational results comparing CPLEX and the proposed heuristic presented and discussed in an effort to evaluate the efficiency of
algorithm solutions. the proposed heuristic algorithm.
Category Size of CPLEX 1 h Total average Difference Percent of
the cost proposed improvement 3.3.1. Hardware and software
instance heuristic
Here, the technical details related to the computational experi-
algorithm
30 min ments that were utilized to evaluate the performance of the
proposed heuristic algorithm are specified below. The heuristic
Small 3,3,10 101,939.8 101,953.6  13.8  0.01354 algorithm just explained in Section 3.2 was coded in Cþ þ using
3,3,15 147,162.9 147,162.9 0 0 Microsoft Visual Studio 2008. The experiments were executed on
4,4,10 124,526 124,526 0 0
4,4,15 185,681.3 185,699.3  18.0667  0.00973
an Hp Elite 8300 with the following technical details: Intel(R) Core
5,5,20 300,866.4 300,900.2  33.8  0.01123 (TM) i5-3470 M CPU at 3.20 GHz 3.30 GHz, 8.00 GB RAM and 64-
Medium 10,10,50 1,336,996 1,337,662  666.467  0.04985 bit Operating system with Windows 7. The IBM(R) ILOG(R) CPLEX
15,15,100 3,816,286 3,810,899 5,387.067 0.14116 (R) Interactive Optimizer 12.5.1.0 was used to solve the instances.
Large 20,20,100 5,084,377 4,975,149 109,228.1 2.148309
In order to identify if the algorithm is in fact capable of finding
20,20,200 10,195,845 9,915,138 280,706.3 2.753144
50,50,200 27,632,948 23,534,629 4,098,318 14.83127 good solutions it is required to disable some special features of
CPLEX. In this case the following parameters were disabled:
cliques, covers, disjoint cuts, lift and project cuts, flow covers,
flow paths, fractional cuts, generalized upper bound cover cuts,
As is typical for the majority of heuristic algorithms, one has to
implied bound cuts, zero half cuts, and mixed integer
define some parameters. In this algorithm it is required to
rounding cuts.
determine the initial set BR and the value for the following
parameters:β,τ and max-nva. The values used for β , τ and max-
nva are 8 min, 30 min, and 400, respectively. It is clear that for 3.3.2. Benchmark of instances
very large instances the number of variables with negative To evaluate the performance of this new approach, the same
reduced costs is also a large number. It is therefore crucial that instances for the multi-product multi-period inventory lot sizing
the maximum number of variables (max-nva) be established with supplier selection problem proposed by Basnet and Leung [3]
considering the size of the largest instances which have 10,000 were used. The benchmark set contains 150 instances. These can
binary variables. The value of max-nva was set to 4% of the total be classified as small, medium, and large. The 75 small instances
number of binary variables in the largest instances. contain between 30 and 100 binary variables. The 30 medium
instances have between 500 and 1500 binary variables. The 45
3.3. Computational results large instances have between 2000 and 10,000 binary variables.
The medium and large instances are considerably hard since
This section presents the comparison of results of the proposed not all of them could be solved to optimality in a reasonable time
heuristic algorithm with the existing results in the literature with the exhaustive enumerative algorithm proposed by Basnet
230 L.E. Cárdenas-Barrón et al. / Computers & Operations Research 64 (2015) 225–232

Table 4
Results for each of the large instances comparing CPLEX and proposed heuristic algorithm.

Category Instance Size of the Total cost of CPLEX % GAP Total cost of proposed heuristic algorithm % GAP % GAP
number instance 1h 30 min improvement

Large 106 20,20,100 5,101,378 6.8205 4,949,737 3.645207 3.175293

107 20,20,100 5,112,809 6.5253 4,973,406 3.620841 2.904459
108 20,20,100 4,936,862 4.3332 4,881,671 3.166821 1.166379
109 20,20,100 5,170,381 6.5458 5,035,871 3.773959 2.771841
110 20,20,100 5,045,316 6.8011 4,914,496 4.031854 2.769246
111 20,20,100 4,954,312 5.4348 4,893,951 4.150232 1.284568
112 20,20,100 5,143,246 4.9235 5,100,410 4.049635 0.873865
113 20,20,100 5,141,382 5.8799 5,027,762 3.540048 2.339852
114 20,20,100 5,270,327 7.4353 5,137,684 4.73138 2.70392
115 20,20,100 5,244,504 7.3144 5,114,953 4.663494 2.650906
116 20,20,100 4,953,208 6.4777 4,837,091 3.981566 2.496134
117 20,20,100 5,130,331 5.7772 5,058,205 4.290106 1.487094
118 20,20,100 5,150,935 6.5836 5,030,712 4.095935 2.487665
119 20,20,100 4,905,368 4.8005 4,786,268 2.255994 2.544506
120 20,20,100 5,005,291 6.4772 4,885,011 3.918492 2.558708
121 20,20,200 10,617,676 13.2956 10,352,356 10.46451 2.831089
122 20,20,200 9,793,130 9.2344 9,706,282 8.265681 0.968719
123 20,20,200 10,066,510 11.7528 9,847,077 9.316777 2.436023
124 20,20,200 10,374,157 13.0547 10,161,970 10.74234 2.312355
125 20,20,200 10,323,373 12.257 10,166,683 10.55314 1.703857
126 20,20,200 10,915,254 18.1395 9,950,331 7.695811 10.44369
127 20,20,200 9,972,145 12.6676 9,692,877 9.512366 3.155234
128 20,20,200 10,063,046 10.9896 9,897,932 9.168488 1.821112
129 20,20,200 9,866,199 10.8207 9,699,241 8.945368 1.875332
130 20,20,200 10,145,424 11.4844 9,901,540 8.804447 2.679953
131 20,20,200 10,101,287 12.2981 9,642,607 7.198859 5.099241
132 20,20,200 9,930,787 11.9457 9,766,838 10.09757 1.84813
133 20,20,200 10,391,155 13.4413 10,130,116 10.59151 2.849789
134 20,20,200 10,156,716 11.1047 9,962,665 8.981969 2.122731
135 20,20,200 10,220,813 11.4238 9,848,562 7.365647 4.058153
136 50,50,200 27,770,956 26.0174 23,678,697 7.447789 18.56961
137 50,50,200 26,396,570 22.3583 23,352,424 8.247507 14.11079
138 50,50,200 27,479,244 25.9933 23,102,756 5.926949 20.06635
139 50,50,200 27,444,473 24.7807 23,746,045 7.965204 16.8155
140 50,50,200 27,931,490 26.8685 23,436,186 6.450238 20.41826
141 50,50,200 27,960,998 26.5155 23,764,888 7.529305 18.98619
142 50,50,200 28,174,415 26.3748 23,592,379 5.822328 20.55247
143 50,50,200 27,047,419 25.3755 22,918,244 6.235138 19.14036
144 50,50,200 27,697,870 25.9144 23,819,901 8.285169 17.62923
145 50,50,200 27,677,086 25.7188 23,288,453 5.784126 19.93467
146 50,50,200 27,627,311 26.2596 22,996,596 5.096765 21.16283
147 50,50,200 27,699,334 25.2242 23,984,561 8.430313 16.79389
148 50,50,200 28,261,279 26.4413 23,834,520 6.635927 19.80537
149 50,50,200 27,356,126 24.8628 23,493,593 7.232866 17.62993
150 50,50,200 27,969,642 25.5966 24,010,197 7.816865 17.77974

Table 5
Results of computational results comparing CPLEX and the proposed heuristic algorithm solutions for the case when Xijt is discrete.

Category Size of the CPLEX 1 h Total average cost proposed heuristic algorithm Difference Percent of No. of optimal
instance 30 min improvement solutions

Small 3,3,10 101,939.8 101,953.6  13.8  0.01354 14/15

3,3,15 147,162.933 147,162.9 0 0 15/15
4,4,10 124,526 124,526 0 0 15/15
4,4,15 185,681.267 185,699.3  18.0667  0.00973 14/15
5,5,20 300,866.4 300,910.4  44  0.01462 9/15
*
Medium 10,10,50 1,337,295.93 1,340,292  2996.33  0.22406
*
15,15,100 3,838,747.6 3,827,811 10,936.13 0.284888
*
Large 20,20,100 5,134,674.67 5,000,178 134,496.9 2.619384
*
20,20,200 11,280,595.6 9,930,991 1349,605 11.96395
*
50,50,200 100,020,161 23,457,449 76,562,712 76.54728

n
In these instances CPLEX cannot prove optimality.

and Leung [3] or CPLEX. Those instances were solved with CPLEX instance size (l, m, n). In the optimization problem, the number of
during 1 h. Conversely, all instances were solved by the proposed the binary variables is given by the product lnm and the number of
algorithm with a time limit of 30 min. constraints including the valid inequality is given by
The instance size can be coded as follows (l, m, n), where l mnn þ lnmnn þn. For example, for any instance of size (50, 50,
indicates number of suppliers, m denotes the number of products, 200), the optimization problem has 10,000 binary variables and
and n means the number of periods. There are 15 instances of each 510,200 constraints (see Table 1).
 L.E. Cárdenas-Barrón et al. / Computers & Operations Research 64 (2015) 225–232 231

Table 6
Results for each of the large instances comparing CPLEX and proposed heuristic algorithm for the case when Xijt is discrete.

Category Instance Size of the Total cost of CPLEX % Total cost of proposed heuristic algorithm % GAP % GAP
number instance 1h GAP 30 min improvement

Large 106 20,20,100 5,123,888 6.7 4,975,005 3.599656 3.100344133

107 20,20,100 5,100,036 5.9 5,002,062 3.865613 2.034386934
108 20,20,100 5,156,031 8.07 4,904,076 2.789043 5.280956777
109 20,20,100 5,210,239 7.08 5,060,789 4.008527 3.071472537
110 20,20,100 5,050,624 6.79 4,934,160 4.327494 2.462505734
111 20,20,100 5,025,172 6.55 4,917,616 4.269463 2.280537223
112 20,20,100 5,257,103 6.45 5,117,854 3.630376 2.81962443
113 20,20,100 5,256,075 8 5,054,967 3.867703 4.132297199
114 20,20,100 5,283,281 7.03 5,169,690 4.728846 2.301154288
115 20,20,100 5,213,989 6.33 5,136,432 4.748363 1.581636595
116 20,20,100 4,965,828 6.45 4,866,706 4.325171 2.124829313
117 20,20,100 5,213,949 6.8 5,081,610 4.089232 2.710767827
118 20,20,100 5,166,390 6.74 5,049,583 4.326714 2.413286488
119 20,20,100 4,952,720 6.37 4,825,053 3.628085 2.741915309
120 20,20,100 5,044,795 7.01 4,907,064 4.088455 2.921544743
121 20,20,200 11,828,390 21.85 1,036,4667 6.771477 15.07852274
122 20,20,200 10,462,578 14.59 9,706,702 6.311368 8.27863179
123 20,20,200 11,806,726 24.09 9,871,906 3.754827 20.33517283
124 20,20,200 11,981,344 24.24 10,174,116 5.500032 18.7399683
125 20,20,200 11,716,363 22.2 10,205,389 6.440756 15.75924396
126 20,20,200 11,336,472 20.42 9,976,380 5.972624 14.44737645
127 20,20,200 10,881,768 19.43 9,711,031 6.580882 12.84911789
128 20,20,200 11,214,475 20.1 9,889,069 5.905732 14.19426773
129 20,20,200 10,498,725 15.61 9,729,708 7.141728 8.468271659
130 20,20,200 11,151,468 20.11 9,911,019 6.749398 13.36060233
131 20,20,200 11,236,852 21.28 9,666,902 4.335438 16.94456205
132 20,20,200 11,026,251 20.54 9,774,891 6.860016 13.67998374
133 20,20,200 11,202,863 19.06 10,139,403 7.757929 11.30207052
134 20,20,200 11,833,552 23.85 9,986,048 4.514016 19.33598385
135 20,20,200 11,031,107 18.78 9,857,631 6.144325 12.63567467
136 50,50,200 27,404,834 25.06 23,561,601 7.521681 17.53832
137 50,50,200 249,310,062 91.78 23,109,566 7.121767 84.65823
138 50,50,200 26,674,181 23.87 23,036,138 6.975596 16.8944
139 50,50,200 257,458,338 92 23,692,418 7.72138 84.27862
140 50,50,200 27,630,837 26.11 23,413,680 6.862459 19.24754
141 50,50,200 27,184,137 24.32 23,628,153 8.057577 16.26242
142 50,50,200 27,456,091 24.58 23,580,648 6.995462 17.58454
143 50,50,200 26,390,723 23.46 22,878,502 7.029271 16.43073
144 50,50,200 27,113,797 24.32 23,727,141 8.791777 15.52822
145 50,50,200 222,679,663 90.77 23,242,631 5.575987 85.19401
146 50,50,200 239,997,786 91.51 22,912,704 4.71337 86.79663
147 50,50,200 27,240,694 24.01 23,905,620 8.827475 15.18253
148 50,50,200 27,741,416 25.19 23,801,140 7.40853 17.78147
149 50,50,200 26,678,982 23.11 23,389,453 7.930488 15.17951
150 50,50,200 259,340,873 91.99 23,982,346 7.691801 84.2982

3.3.3. Computational results method because it obtains 93.34% (70/75) of the optimal solutions
In Basnet and Leung [3] there are two algorithms: enumerative (see the penultimate column of Table 2).
and heuristic. The results obtained by the proposed heuristic It can be observed from Table 3 that when the results of the
algorithm are compared with the best results obtained with the proposed heuristic algorithm are compared with CPLEX with a run
heuristic algorithm of Basnet and Leung [3]. The computational time of 1 h, a more impressive improvement is observed for the
results are shown in Table 2. In detail, Table 2 shows Basnet and large instances of (50,50,200) which contain 10,000 binary vari-
Leung’s [3] solutions, the solutions of the proposed algorithm, the ables. In this case the average percent improvement is 14.831%.
difference, the percent of improvement, the computational time, In order to demonstrate the power of the proposed heuristic
the number of optimal solutions achieved, and the gap. Each value algorithm, Table 4 presents the results for each one of the large
in each line of Tables 2 and 3 is an average over 15 instances. It is instances. From results of Table 4, it was calculated that the average %
important to note that CPLEX is able to prove optimality before the gap of improvement for the large instances 106–121, 121–136 and
time limit for only the small instances. 136–150 are 2.280%, 3.080%, and 17.921%, respectively. Furthermore,
In line with the results of Table 2 one can conclude that the the maximum percent improvements are 3.175%, 10.44%, and
proposed heuristic algorithm was able to find on average better 21.162% for the large instances 106–121, 121–136, and 136–150,
solutions than those reported by Basnet and Leung [3] for all respectively. These results demonstrate that the heuristic algorithm
instance sizes. It is observed that for the very large instances (50, performs very well. Moreover, the heuristic algorithm is noteworthy
50, 200) with 10,000 binary variables, a very significant percent of from both business and computational perspectives.
improvement of 7.245% was obtained. This indicates that the
performance of the Basnet and Leung [3] algorithm deteriorates
for large instances. 3.3.4. Integer linear programming model (ILP)
With regard to small instances, the proposed heuristic algo- Since constraints (3) do not satisfy the integrality property, the
rithm based on ROA is shown to be an effective and efficient optimization problem can be formulated as an integer linear
232 L.E. Cárdenas-Barrón et al. / Computers & Operations Research 64 (2015) 225–232

programming model (ILP). Thus, if we consider the optimization Acknowledgements

problem (1)–(4) and the lot sizes as discrete variables then the
constraints (5) are restated as follows: This research was supported by the Tecnológico de Monterrey
X ijt Z 0 and Integer 8 i; j ; t Research Group in Industrial Engineering and Numerical Methods
ð6Þ
0822B01006 and the Mexican National Council for Science and
Then the problem becomes an ILP model. Obviously, con- Technology (Grant SEP-CONACYT CB-2011-01-166397). The
straints (6) significantly increase the complexity of the optimiza- authors thank Dr. Chuda Basnet for kindly sharing with us the
tion problem. The number of integer variables in the ILP model for set of benchmark instances that make possible this research paper.
the instances are given in column fourth of Table 1. For example, A special gratitude is given to Dr. Neale R. Smith for his valuable
the large instances (50,50,200) have 500,000 integer variables and feedback that greatly enhanced this research work. Finally, the
10,000 binary variables. authors thank the anonymous reviewers for their valuable sugges-
Each line of Table 5 shows an average over 15 instances. From tions that improved this paper.
Table 5 one can conclude that the proposed heuristic algorithm has an
impressive performance for the large instances of (20,20,200) and References
(50,50,200) which contain 4000 and 10,000 binary variables. In these
cases, the average percent improvements are 11.963% and 76.547% [1] Aissaoui N, Haouari M, Hassini E. Supplier selection and order lot sizing
for large instances of (20,20,200) and (50,50,200), respectively. modeling: a review. Comput Oper Res 2007;34(12):3516–40.
[2] Bahl HC, Ritzman LP, Gupta JND. Determining lot sizes and resource require-
Furthermore, with regard to small instances, the proposed heuristic
ments: a review. Oper Res 1987;35(3):329–45.
algorithm based on ROA also has an effective and efficient perfor- [3] Basnet C, Leung JMY. Inventory lot-sizing with supplier selection. Comput
mance because it obtains 89.333% (67/75) of the optimal solutions Oper Res 2005;32(1):1–14.
(see the last column of Table 5). [4] Cárdenas-Barrón LE. Adaptive genetic algorithm for lot-sizing problem with
self-adjustment operation rate: a discussion. Int J Prod Econ 2010;123
Table 6 shows the results for the large instances for the integer (1):243–5.
linear programming model. From results of Table 6, it was computed [5] Choudhary D, Shankar R. Joint decision of procurement lot-size, supplier
that the average % gap of improvement for the large instances 106– selection, and carrier selection. J Purchasing Supply Manage 2013;19(1):16–26.
[6] Choudhary D, Shankar R. A goal programming model for joint decision making
121, 121–136, and 136–150 are 2.798%, 14.360%, and 39.523%, of inventory lot-size, supplier selection and carrier selection. Comput Ind Eng
respectively. Additionally, the minimum and the maximum percent 2014;71:1–9.
improvements are (1.581%, 5.280%), (8.278%, 20.335%), and (15.179%, [7] Dahel NE. Vendor selection and order quantity allocation in volume discount
environments. Supply Chain Manage 2003;8(4):335–42.
86.796%) for the large instances 106–121, 121–136, and 136–150, [8] De Bodt MA, Gelders LF, Van Wassenhove LN. Lot sizing under dynamic
respectively. These results demonstrate that the heuristic algorithm demand conditions: a review. Eng Costs Prod Econ 1984;8(3):165–87.
also performs very well for the integer linear programming model. [9] Ho W, Xu X, Dey PK. Multi-criteria decision making approaches for supplier
evaluation and selection: a literature review. Eur J Oper Res 2010;202
(1):16–24.
[10] Hop NV, Tabucanon MT. Adaptive genetic algorithm for lot-sizing problem
4. Conclusions with self-adjustment operation rate. Int J Prod Econ 2005;98(2):129–35.
[11] Jans R, Degraeve Z. Meta-heuristics for dynamic lot sizing: a review and
In this paper, for the first time, the simple and powerful reduce and comparison of solution approaches. Eur J Oper Res 2007;177(3):1855–75.
[12] Jayaraman V, Srivastava R, Benton WC. Supplier selection and order quantity
optimize approach (ROA) was used to solve the multi-product multi- allocation: a comprehensive model. J Supply Chain Manage 1999;35(2):50–8.
period inventory lot sizing with supplier selection problem. This problem [13] Karsak EE, Dursun M. An integrated fuzzy MCDM approach for supplier
is a new variant of the lot sizing inventory problem and is very difficult evaluation and selection. Comput Ind Eng 2015;82:82–93.
[14] Kasilingam RC, Lee CP. Selection of vendors—a mixed-integer programming
to solve to optimality. A new valid inequality also was added to reinforce approach. Comput Ind Eng 1996;31(1-2):347–50.
the mathematical model. It was observed that the new valid inequality [15] Kuik R, Salomon M, Van Wassenhove LN. Batching decisions: structure and
has a beneficial impact on both solution and time issues. Solving the 150 models. Eur J Oper Res 1994;75(2):243–63.
[16] Moghadam MRS, Afsar A, Sohrabi B. Inventory lot-sizing with supplier selection
benchmark instances including 75 small-sized instances, 30 medium- using hybrid intelligent algorithm. Appl Soft Comput 2008;8(4):1523–9.
sized instances, and 45 large-sized instances revealed that the proposed [17] Ruiz-Torres AJ, Mahmoodi F, Zeng AZ. Supplier selection model with con-
heuristic algorithm yields high quality solutions within a reasonable CPU tingency planning for supplier failures. Comput Ind Eng 2013;66(2):374–82.
[18] Simpson NC. Questioning the relative virtues of the dynamic lot sizing rules.
time. The performance of the heuristic algorithm is very satisfactory
Comput Oper Res 2001;28(9):899–914.
because it found better solutions than those reported by Basnet and [19] Treviño-Garza G. A heuristic approach to general 0-1 integer programming.
Leung [3]. Additionally, the results from detailed computational experi- PhD dissertation. Arizona State University; 2009 UMI-Number: 3354471.
ment confirm the superiority of the proposed heuristic algorithm over [20] Tsai YL, Yang YJ, Lin C-H. A dynamic decision approach for supplier selection
using ant colony system. Expert Syst Appl 2010;37(12):8313–21.
the CPLEX MIP solver. For example, compared with CPLEX with time [21] Ustun O, Demirtas EA. Multi-period lot-sizing with supplier selection using
limit of 1 h, the heuristic algorithm improved the solutions of all large achievement scalarizing functions. Comput Ind Eng 2008;54(4):918–31.
instances in 30 min. These results ratify the success of the proposed [22] Wadhwa V, Ravindran AR. Vendor selection in outsourcing. Comput Oper Res
2007;34(12):3725–37.
heuristic algorithm. Moreover, one could conclude that ROA has become [23] Wagner HM, Whitin TM. Dynamic version of the economic lot size model.
a prominent methodology for tackling complex optimization problems. Manage Sci 1958;5(1):89–96.
There are several future research directions that may be pursued [24] Ware NR, Singh SP, Banwet DK. Supplier selection problem: a state-of-the-art
review. Manage Sci Lett 2012;2(5):1465–90.
after this paper. Immediate extensions are incorporating important [25] Ware, N.R., Singh, S.P., Banwet D.K. . A study of inter-relationship of supplier
issues like allowing shortages and capacity constraints such as space selection criteria using ISM approach. In: Proceedings of the seventh national
and budget. However, this will significantly increase the complexity of conference, INDIACom-2013 computing for nation development; 2013.
[26] Ware NR, Singh SP, Banwet DK. Modeling flexible supplier selection frame-
the optimization problem. Another future direction could be to apply work. Global J Flexible Syst Manage 2014;15(3):261–74.
the proposed algorithm to other NP hard problems such as distinct [27] Ware NR, Singh SP, Banwet DK. Analyzing the effect of demand variation on
variants of the lot sizing inventory problem or any optimization multi-product, multi-source, multi-period model for supplier selection pro-
blem. Ind Eng J 2014;7(2):13–8.
problem with binary variables such as the vehicle routing problem
[28] Ware NR, Singh SP, Banwet DK. A mixed-integer non-linear program to model
(VRP) and the inventory routing problem (IRP), just to name two dynamic supplier selection problem. Expert Syst Appl 2014;41(2):671–8.
possibilities. Also, the development of metaheuristics to solve the [29] Wolsey LA. Progress with single-item lot-sizing. Eur J Oper Res 1995;86
multi-product multi-period inventory lot sizing with supplier selec- (3):395–401.
[30] Zangwill WI. A backlogging model and a multi-echelon model of a dynamic
tion problem and comparing their results with our solutions possibly economic lot size production system—a network approach. Manage Sci
could be another interesting research opportunity. 1969;15(9):506–27.