You are on page 1of 6

A Profit Maximization Lot-sizing and Supplier

Selection Problem Under Price Discount

Sakada Srun Thanakorn Naenna


Department of Industrial Engineering, Department of Industrial Engineering,
Faculty of Engineering, Mahidol University Faculty of Engineering, Mahidol University
Nakhon Pathom, Thailand Nakhon Pathom, Thailand
sakada.sru@student.mahidol.edu thanakorn.nae@mahidol.ac.th

Abstract—This research addresses the development of multiple periods planning horizon. The order quantity
integrated inventory lot-sizing and supplier selection of each product type can be placed to a set of suppliers.
problem in supply chain management, in which a buyer The holding is product-dependent, and the ordering and
firm that has limited budget and warehouse space wants transportation cost are differed by various suppliers.
to procure multiple products from a group of potential The DMs need to decide how many units of what
suppliers and sell to their customer. In particular, the products to order from which suppliers in which
suppliers may provide different quantity discount periods.
scheme, ordering cost, and transportation cost. The
problem is modelled as mixed integer linear II. LITERATURE REVIEW
programming (MILP) by incorporating quantity
discount. To overcome the large scale problem, Genetic In 1913, the first lot sizing model is the renowned
Algorithm (GA) is developed in next phase. The Economic Order Quantity (EOQ), developed by Harris
comparison was conducted on both MILP and GA. The (1913). Afterward, this problem has become popular
results showed that GA has better performance in term research topic rapidly, and it has long attracted the
of solution feasibility and computational time. attention of researchers. It was one of the first
applications of mathematical modeling to guide one in
Keywords-optimization; inventory lot-sizing; supplier making business decisions, and it has spawned
selection; genetic algorithm; net revenue thousands of related studies over the past century that
have built on its major foundations and insights (Choi,
I. INTRODUCTION 2013).
Inventory represents a major financial investment
One of the major assumptions in EOQ model is
for any company. It represent 25 to 50 percent of total
deterministic and stable. When it came to deterministic
assets in manufacturing firms and 75 to 80 percent in
and time-varying demand, the version of dynamic lot-
wholesalers and retailers Shubham (2017). As a result,
sizing was introduced next by (Wagner & Whitin,
whether a company can effectively control its cost
1958). Subsequently, the model was extensively
greatly depends on the successfulness of its
developed by many researchers due to its prominence.
procurement policy.
The comprehensive literature review on development
In the world competitive market, multi-sourcing of lot-sizing problem and its solution approach could
has become a trend, and the alternative suppliers have be found in (Andriolo, Battini, Grubbström, Persona,
provided different selling strategy to engage the buyers. and Sgarbossa (2014); Brahimi, Absi, Dauzère-Pérès,
In this circumstance, it make the buyer forms to select & Nordli, 2017; Glock, Grosse, & Ries, 2014).
the right set of supplier in order to take advantage of
Lot-sizing, although a very important problem in
differentials in various factors such as product cost,
operations management, is not an independent
supplying capacity, transportation cost, discount
problem. Integrating lot-sizing and supplier selection
scheme, and ordering cost (Li, Ventura, Venegas,
implies making a joint decision on which products to
Kweon, & Hwang, 2018).
order in what quantities, and from which suppliers and
To fully assist the decision makers (DMs) in when (Rezaei, Davoodi, Tavasszy, & Davarynejad,
allocating the order, another research design is to take 2016). Basnet and Leung (2005) presented a multiple-
into account the inventory lot sizing and supplier period inventory lot-sizing scenario where multiple
selection synchronously. The problem become more products and multiple suppliers are taken into account.
complicated when the number of product and period A mixed integer programming is formulated, and
are multiple. DMs have to determine the optimal order enumerative algorithm and heuristic algorithm are
quantity of each product type in all periods in intention proposed to overcome large problem. Also, Cárdenas-
of maximizing the profit as to minimize the total cost Barrón, González-Velarde, and Treviño-Garza (2015)
(Salameh & Jaber, 2000). proposed a new algorithm based on reduce and
optimize approach (ROA) and new valid inequality to
This paper consist of a profit maximization lot solve the same problem.
sizing and supplier selection problem where the
demand of multiple products is known over the

978-1-7281-4466-5/20/$31.00 ©2020 IEEE 24


III. MATHEMATICAL MODELLING W Total storage capacity
M A large number
A. Model assumtions
The proposed inventory model in this study is Independent Decision Variable
formulated base on the following assumption. Number of product i ordered to supplier j
• Demand of product in each period are known over with price break k in period t
the finite planning horizon. Number of product i ordered to supplier j
• At most one order can be placed in each period, and with in period t
the entire ordered product are delivered at once in the Total purchasing quantity of product i in
beginning of the periods. period t
• Each supplier offers one of the quantity discounts (all Binary variable, set equal to 1 if purchase
unit discount/ incremental discount). quantity from supplier j in period t, 0
• Imperfect product rate are known. otherwise
• Imperfect product are sold in discounted price. Binary variable, set equal to 1 if purchase
• Ordering cost does not depend on the variety and quantity of produce i from supplier j with
quantity of product. price break k in period t, 0 otherwise
• Holding cost of product per period is product-
dependent. Dependent Decision Variable
• Transportation costs are known for each capacity of Expected beginning inventory level of
vehicle, and it is paid by the buyer. product i in period t
• The initial inventory level is zero. Expected ending inventory level of product
• Shortage is not allowed. i in period t
• The budget is used for covering ordering, purchasing Number of required vehicle from supplier j
and transportation costs, and it is limited. in period t
• The supplier capacity is limited.
• The warehouse has limited capacity.
C. Objective Function
B. Notation The net revenue will be maximize, and it is defined
Indices as the gross revenue minus all relevant cost. The
i index of product objective function, which aim to maximize the net
revenue, is formulated as follow:
j index of suppliers. Supplier j = 1,…j’
provide all-units quantity discount, and ( )=
supplier j = j’+1,…, J provide incremental
quantity discounts
t index of periods [(1 − )X
k index of price break
Parameters
+ X
Demand of product i in period t
Purchase cost of product i from supplier j
with price break k
The lower bound quantity of product i from −
supplier j with price break k
The upper bound quantity of product i from
supplier j with price break k
Imperfect rate of product i offered by −
supplier j in period t
Unit selling price per perfect unit of product
i − − −
( )
Unit selling price per imperfect unit of
product i
Holding cost of product i per period
Ordering cost from supplier j − (2 − ) − (1)
Transportation cost per vehicle form supplier 2
j
Vehicle capacity from supplier j The objective function as shown in (1) consists of
Capacity of supplier j in production of the gross revenue and the relevance cost which include
product i per period
Occupied space by product i in warehouse or purchasing cost, ordering cost, holding cost, and
vehicle transportation cost.
Maximum budget of buyer in period t

25
D. Constraints (8) is the buyer’s warehouse capacity constraint. In
equation (9), the total cost must not exceed the
X ≤ M × For all , , and (2) maximum budget. Equation (10) represent the supply
capacity of the supplier. The ending inventory must be
zero in equation (11). In constraint (12) & (16), the
≤X ≤ order quantity and number of vehicle must be integer.
For all , , , and Equation (13) & (15) are binary constraint. Constraint
(3)
(14) restrict that the ending inventory must be non-
negative.
U ≤ 1 For all , , and (4)
IV. DEVELOP GENETIC ALGORITHM
The Genetic Algorithm (GA) is developed to find
= + X For all i and t the near-optimal solution especially for large scale
(5)
problem. The general flow of Genetic Algorithm is
shown in Figure 1.
= − For all and (6)
∑ ∑ ∑ ∑ Initial population
≤ ≤ + 1
Evaluation
For all and (7)

Selection
≤ For all (8)

Crossover

P X
Mutation

No Yes
+ − ( ) Terminate? Stop

Figure 1. The general flow of Genetic Algorithm.


− +
In this study, the rank-based roulette wheel
selection method is adapted. The random value in the
range [0, 1] was generated and assigned for each pair
+ (2 − )+ ≤ For all (9) of parent chromosome, and the parent chromosome
2 pair was crossed if the assigned random value is less
than or equal to the crossover rate . In the proposed
GA, scattered crossover operator was employed. Some
X ≤ For all , , and (10) of the genes which have the randomized value less
than or equal to mutation rate was mutated. Since
the Bit flip was used in this study, the selected genes
X − = 0 For all (11) to be mutated was changed the value from 0 to 1 or 1
to 0. The solver time was not allowed to exceed 2
hours, or the algorithm is terminated when no further
X ℕ ∪ (0) For all , , , and (12) development could be obtained in the solution after 50
U = 0 1 For all , , , and successive generations. Moreover, to possibly obtain
(13) good quality solution, the termination condition of 600
≥ 0 For all and (14) generations was used in this study.
= 0 1 For all and (15) A. Chromosome Representation
ℕ ∪ (0) For all and (16) The chromosome is encoded as bianry value voctor
length of (IxT) which is marked by Y. The value of
In equation (2), buyer have to face the ordering cost gene equal to 1 if the order of product i is placed to
if the order is made. Equation (3) & (4) are used to any supplier in period t. The structure of such a
restrict that the ordered quantity must be in the interval chromosome is depicted in figure 2.
of price break quantity. The beginning inventory and
inventory balancing are shown in equation (5) & (6)
respectively. Equation (7) set the number of vehicle …
which is used by the supplier in each period. Equation Figure 2. Chromosome structure

26
B. Parameter seting Somehow, various products is not provided the
same feature by one supplier as well as many different
There are three basic parameters of GA - crossover
suppliers. All these aspects are presented in Table IV
probability, mutation probability, and population size.
and Table V.
The size IxT is taken to find population size. The
parameter setting as shown in Table 1 is used in this Table IV. Supplier related data by product type
study.
Supplier Product Defective Rate Capacity
Table I. GA parameter setting
1 0.06 2601
Level 1
Factor 2 0.04 1343
1 2 3 1 0.06 2043
2
Cross rate ( ) 0.75 0.85 0.95 2 0.06 1706
Mutation rate ( ) 0.1 0.15 0.2 1 0.04 2198
3
Population size Double Triple Quadruple 2 0.04 1286
1 0.05 2724
V. NUMERICAL EXAMPLE 4
2 0.05 1554
A few number of instance have been generated to
test the model and evaluate the proposed GA. The first
Table V. The parameter related to buyer
problem involves two kinds of product being
purchased from some of four suppliers. The horizontal Period 1 2 3 4 5
plan is five periods. The value of all the parameter was Budget 17470 15804 14202 16464 18685
generated by using random number generation tool Warehouse 7000 7000 7000 7000 7000
built in Microsoft Excel. The products related data and
supplier related data are shown in below table. The MILP is encoded in optimization software
LING version 12, and GA is coded in software
Table II. Product related data MATLAB. The computational engine is Dell Inspiron
3442 (Intel core i5 with 4 GB of RAM). To evaluate
Product

the performance of using LINGO and GA to solve the


proposed model, a set of large-scale problem instance
1 2 3 4 5 are tested. The base case instance are extend by
1 2383 2185 2454 2072 2867 0.2 1 5 2.5 increasing the number of product (I), the number of
supplier (J), and the number of period (T). Table 2
2 1679 1956 1653 1595 1721 0.25 2 6 3
show all the problem size which was used to evaluate
and compare the capability problem solving approach
Table III. Supplier related data for all product types in this study.

J K ( , ) Table II. Problem size scenario


1 2
1 1 - 350 3.9 4.2 Problem Scenario Problem size
1 2 351 - 900 3.7 3.9 No. ID
179 34 1949
(All-unit) 3 901 - 1780 3.6 3.4 1 PB-1 4 4 4 5
4 1781 - Infinite 3.4 3.3 2 PB-2 4 4 4 10
1 1 - 500 3.9 4.1
3 PB-3 4 4 4 20
2 2 501 - 880 3.8 3.9
(All-unit) 250 35 1921 4 PB-4 5 8 4 10
3 881 - 1730 3.5 3.6
5 PB-5 5 8 4 20
4 1731 - Infinite 3.4 3.2
6 PB-6 5 8 4 30
1 1 - 440 3.7 4
203 30 1808 7 PB-7 10 8 4 20
3 2 441 - 930 3.4 3.7
(Incremental) 3 931 - 1530 3.3 3.3 8 PB-8 10 8 4 30
4 1531 - Infinite 2.8 2.9 9 PB-9 15 12 4 20
1 1 - 500 3.8 4.1 10 PB-10 15 12 4 30
4 2 501 - 840 3.3 3.6
182 30 1786 For GA approach, full factorial design of
(Incremental) 3 841 - 1520 3.1 3.4
experiment is done for all the parameter in table I to
4 1521 - Infinite 2.8 3.1 find the best parameter set for each problem size.

27
VI. PERFORMANCE EVALUATION OF PROPOSED According to Table IV, the MILP which is encoded
GENETIC ALGORITHM in LINGO optimization software is capable to find the
optimal solution for the first two scenarios only, and
A. Genetic Parameter Selection
the solution time is increased exponentially when the
Each problem size was done the experiment on all problem size increase. The last 6 problem size was not
varies scenario of population size, crossover rate, and found the solution within 24 hours, so the lower bound
mutation rate. The best scenario which provides the of the solution was obtained.
highest value of objective in any problem size was
selected. The selected genetic parameters was shown C. Solution ending gap and solution gap
in Table III. The GA algorithm encoded in software MATLAB
could provide the near optimal solution within short
Table III. The selected genetic parameters period of time. The solution gap and computational
time improvement were presented in Table V.
Population Crossover Mutation
Scenario
size rate rate Table V. The performance of the proposed GA method
PB-1 20 0.85 0.15
PB-2 120 0.85 0.1 Ending Solution Computational
Scenario gap gap time improvement
PB-3 320 0.75 0.15 (%) (%) (%)
PB-4 240 0.85 0.2 PB-1 0 2.24 -114.00
PB-5 480 0.95 0.2 PB-2 0 2.21 88.48
PB-6 720 0.95 0.15 PB-3 2.38 6.93 99.58
PB-7 640 0.85 0.1 PB-4 1.11 2.73 99.86
PB-8 720 0.75 0.2 PB-5 100 9.98 99.80
PB-9 720 0.75 0.1 PB-6 100 8.09 99.75
PB-10 1080 0.85 0.15 PB-7 100 4.74 99.83
PB-8 100 3.05 99.50
B. Solution quality evaluation PB-9 100 3.78 99.42
The results obtained in the objective value (total net PB-10 100 6.87 99.06
revenue) and the solution time of the MILP and the
proposed GA method were presented in Table IV. Because the algorithm of the MILP approach
would test all choices of the decision, the solution time
Table IV. The results obtained of the MILP method is long for the large size problem, and it is different
and the proposed GA method from the GA method that randomly tests the
alternatives. In the other hand, if the problem size is
MILP results GA results large, the solution time of the MILP approach
increases exponentially based on the number of
Scenario

constraints while the proposed GA may be able finding


time (Sec)

time (Sec)
Objective

Objective
Solution

Solution
value

value

the feasible solution within short time. Figure 3 and


figure 4 respectively showed the solution gap and the
variation of LINGO computational time to GA
computational time.
PB-1 35465.45 (a) 48 34672.17 105
PB-2 71910.86 (a) 3080 70324.35 355
120
PB-3 144801.57 (b) 81732 134768.72 345
100
Error (%)

PB-4 90623.62 (b) 75902 88150.43 104


80
PB-5 187127.28 (c) 84672 168460.82 167 60
PB-6 285590.82 (c) 84672 262491.33 208 40
PB-7 374254.45 (c) 84672 356521.75 141 20
PB-8 551581.73 (c) 84672 534782.57 420 0
PB-9 545594.71 (c) 84672 524982.57 492
PB-10 866572.55 (c) 84672 807073.90 797
a The optimal solution. MILP(Ending gap)
b The best solution that found during solving 24 hours.
c The lower bound of the solution at 24 hours. GA (Relative error)

Figure 3. Ending gap and relative error

28
100000 slightly decreased in GA, but the computational time
Computational time (Second)
is must faster than the exact approach. Compared to
90000
the commercial solver LINGO, the meta-heuristic
80000 approach is more effective and efficient.
70000
60000
REFERENCES
50000
40000
[1] Andriolo, A., Battini, D., Grubbström, R. W., Persona, A., &
30000 Sgarbossa, F. (2014). A century of evolution from Harris‫׳‬s
20000 basic lot size model: Survey and research agenda.
International Journal of Production Economics, 155, 16-38.
10000 doi: https://doi.org/10.1016/j.ijpe.2014.01.013
0 [2] Basnet, C., & Leung, J. M. Y. (2005). Inventory lot-sizing with
1 2 3 4 5 6 7 8 9 10 supplier selection. Computers & Operations Research, 32(1),
1-14. doi: https://doi.org/10.1016/S0305-0548(03)00199-0
MILP Computational Time [3] Brahimi, N., Absi, N., Dauzère-Pérès, S., & Nordli, A. (2017).
Single-item dynamic lot-sizing problems: An updated survey.
GA Computational Time European Journal of Operational Research, 263(3), 838-863.
doi: https://doi.org/10.1016/j.ejor.2017.05.008
[4] Cárdenas-Barrón, L. E., González-Velarde, J. L., & Treviño-
Figure 4. The solution time of the MILP and Garza, G. (2015). A new approach to solve the multi-product
proposed GA multi-period inventory lot sizing with supplier selection
problem. Computers & Operations Research, 64, 225-232.
VII. CONCLUSION doi: https://doi.org/10.1016/j.cor.2015.06.008
[5] Choi, T. M. (2013). Handbook of EOQ Inventory Problems:
This research addresses the problem of Stochastic and Deterministic Models and Applications:
coordinating lot-sizing and supplier selection Springer US.
problems with its objective of maximal net profit. This [6] Glock, C. H., Grosse, E. H., & Ries, J. M. (2014). The lot
problem constantly faces three decision problems that sizing problem: A tertiary study. International Journal of
are usually treated separately in the literature and are Production Economics, 155, 39-51. doi:
https://doi.org/10.1016/j.ijpe.2013.12.009
mostly led to heuristic solutions. Therefore, this
[7] Harris, F. W. (1913). How Many Parts to Make at Once.
research aims to answer the following questions: (1) Operations Research, 38(6), 947-950. doi:
When is the right time to purchase the product? (2) 10.1287/opre.38.6.947
How many unit of product should be purchased in any [8] Li, X., Ventura, J. A., Venegas, B. B., Kweon, S. J., & Hwang,
specific period? (3) From which supplier should the S. W. (2018). An integrated acquisition policy for supplier
firm place the order? To answer these questions, the selection and lot sizing considering total quantity discounts
and a quality constraint. Transportation Research Part E:
research is broken down to two phases. The first phase Logistics and Transportation Review, 119, 19-40. doi:
focuses on analytical definition on the research https://doi.org/10.1016/j.tre.2018.09.003
problem and mathematical modeling. The second [9] Rezaei, J., Davoodi, M., Tavasszy, L., & Davarynejad, M.
phase is to develop the meta-heuristics solution (2016). A multi-objective model for lot-sizing with supplier
selection for an assembly system. International Journal of
approach that can solve problem in practical size. Logistics Research and Applications, 19(2), 125-142. doi:
Experiments on large-scale problems show that 10.1080/13675567.2015.1059411
directly solving the mathematical model takes a [10] Salameh, M. K., & Jaber, M. Y. (2000). Economic production
prohibitively long computation time. Therefore, the quantity model for items with imperfect quality. International
second phase is focused on developing an efficient Journal of Production Economics, 64(1), 59-64. doi:
https://doi.org/10.1016/S0925-5273(99)00044-4
meta-heuristic solution approach namely Genetic
[11] Shubham, V. U., Pawan Kumar. (2017). To Study the
Algorithm. The algorithm is coded in software Inventory Management System at Organization Level.
MATLAB. In this dissertation, the parameter setting is International Journal of Engineering Science and Computing,
adapted from the previous well known publication. 7(No.8).
The effort of GA do achieve solution feasibility for all [12] Wagner, H. M., & Whitin, T. M. (1958). Dynamic Version of
the Economic Lot Size Model. Management Science, 5(1), 89-
problem size in this research. With experiments on the 96. doi: 10.1287/mnsc.5.1.89
same simulated problem instances, the net revenue is

29

You might also like