Available online at www.sciencedirect.

com

ScienceDirect
Procedia CIRP 56 (2016) 199 202

9th International Conference on Digital Enterprise Technology - DET 2016 Intelligent Manufacturing in
the Knowledge Economy Era

General model for single-item lot-sizing with multiple suppliers, quantity

discounts, and backordering
Hesham K. Alfaresa, Rio Turnadib *
a
King Fahd University of Petroleum and Minerals (KFUPM), Dhahran, 31261, Saudi Arabia
b
Universitas Pertamina, Jakarta, 12220, Indonesia
* Corresponding author. E-mail address: alfares@kfupm.edu.sa, rio.turnadi@universitaspertamina.ac.id

Abstract

Lot-sizing is a very important problem in production and inventory planning under variable demand. In this work, a general model is proposed
for a single-item lot-sizing problem with multiple suppliers, quantity discounts, and backordering of shortages. Mixed integer programming
(MIP) is used to formulate the problem and obtain the optimum solution for small problems. Due to the large number of variables and
constraints in practical problems, the model is too difficult for an optimal solution. Therefore, an effective heuristic method is developed by
modifying the well-known silver-meal heuristic. This heuristic method is shown to be effective for this problem, producing near-optimal
solutions much faster than MIP.
2016 The Authors. Published by Elsevier B.V.
2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
Peer-review under responsibility of the Scientific Committee of the 9th International Conference on Digital Enterprise Technology - DET
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
2016.
Peer-review under responsibility of the scientific committee of the 5th CIRP Global Web Conference Research and Innovation for Future Production

Keywords: Lot sizing; inventory control; quantity discounts; mixed integer programming; Silver-Meal heuristic

1. Introduction and incremental discounts. Lee et al. [4] developed a model

Classical lot-sizing literature approaches are focused on the
single-item lot-sizing problem. Okhrin and Richter [1]
developed a lot-sizing model for a single item by considering
restrictions on the minimum order quantity. Wagner [2]
developed a lot-sizing model considering two product types (a
perishable product, and a durable product). The perishable
product is strictly used to fully satisfy the demand, i.e. no
shortage is allowed. Any unmet demand for the perishable
product results in losing the sale. The second product type is
more durable, and can be used for backlogging unmet
demands. For this product, shortages in the current period can
be satisfied in future periods. A dynamic model was
developed for solving this lot-sizing problem. Chu et al. [3]
proposed a capacitated single-item lot-sizing model,
considering holding, backlogging and outsourcing in the cost
function. In their model, they assumed that holding,
backlogging and outsourcing costs are linear.
Quantity discounts, providing lower unit purchase costs for
larger customer orders, fall into two types: all-units discounts
for a joint replenishment problem with both incremental and
all-units quantity discounts. Their single-item lot-sizing model
considered four cost-components: ordering cost, purchasing
cost, transportation cost, and holding cost. Jung and Klein [5]
included quantity discounts in the cost function for an
economic order quantity (EOQ) model assuming price-
dependent demand. Ramasesh [6] used price discounts as an
incentive in lot sizing, assuming the vendor offers a limited-
time price reduction.
Maiti et al. [7] introduced a multi-item lot-sizing problem
with two-storage types (owned storage and rented storage),
considering an all-units discount policy and assuming
shortages are not allowed. Because of the complexity of the
model, genetic algorithms (GA) were used for the solution.
Drake and Pentico [8] and San-Jose and Garcia-Laguna [9]
considered backordering in a lot-sizing problem under
quantity discounts. San-Jose et al. [10] developed a model for
lot-sizing with partial backlogging, in which customers will
accept backordering if their waiting time is short, but a longer
waiting time will result in more lost sales. In a related model,

2212-8271 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the scientific committee of the 5th CIRP Global Web Conference Research and Innovation for Future Production
doi:10.1016/j.procir.2016.10.054
200 Hesham K. Alfares and Rio Turnadi / Procedia CIRP 56 (2016) 199 202

San-Jose et al. [11] allowed both backordering and lost sales, SBt Beginning shortage level in period t.
assuming an exponential partial backlogging rate. Toews et al. SEt Ending shortage level in period t.
[12] considered a backordering cost that is linearly dependent
1 if there is inventory at the beginning of period t, 0 if
on the delivery time of the item in both EOQ and EPQ ABt
not.
(economic production quantity) models.
This paper presents a new, single-item, multi-supplier lot- AEt 1 if there is inventory at the end of period t, 0 if not.
sizing problem with quantity discounts and shortages. Next, 1 if there is a shortage at the beginning of period t, 0 if
BBt
the MIP model is developed in section 2, the heuristic solution not.
method is presented in section 3, and numerical examples are BEt 1 if there is a shortage at the end of period t, 0 if not.
solved in section 4. Finally, conclusions are given in section 1 if a purchase from a supplier i is made in period t, 0 if
5. Fit
not.
1 if a purchase from a supplier i with price break k is
2. Model Formulation Uitk
made in period t, 0 if not.
M A large number.
The objective of the lot-sizing problem is to find the
optimal order quantities when the demand varies over time. In Model formulation and related costs:
this paper, demand is assumed to be known, and no more than
The lot-sizing problem in this paper is formulated using
one order can be made per period. Several suppliers are mixed-integer programming (MIP). Eq. (1) is used for
available, with different purchase prices and quantity discount calculating the total ordering cost (CO) from all suppliers
policies: both all-units or incremental. Full orders are received during the planning horizon. Eq. (2) and (3) are used for
at the start of each period. The transportation cost is calculating the purchase cost from supplier i in period t under
considered and calculated per vehicle. The planning horizon is all-units and incremental discounts, respectively.
known and finite, and all demands must be satisfied by the
end of the horizon. T I
Parameters: CO o F
t 1 i 1
i it (1)
dt Demand in period t. K
h Unit holding cost per period. CPit pitkU itk Qit , i = 1, , i, and t = 1, , T (2)
l Unit shortage cost per period. k 1
K
oi Ordering cost from supplier i. CPit p itkU itk QPit , i = i +1, , I, and t = 1, , T (3)
ci Containers volume per vehicle of supplier i. k 1

si Transportation cost per vehicle from supplier i.

The total holding cost for each period (CI) is obtained by
Unit purchase cost from supplier i with price break k multiplying the unit holding cost with the average of the
pitk
under all-units quantity discounts. beginning and ending inventory levels. There are three
Subtracted unit purchase cost from supplier i with price possible cases, as shown in Figure 1. For case (a), BEt = 0,
psik
break k under all-units quantity discounts. and Eq. (4) is active. For case (b), BEt = 1, BBt = 0, and Eq.
(5) is active. For case (c), BEt = 1, BBt = 1, and both Eq. (4)
p itk Average unit purchase cost from supplier i in period t and (5) are inactive because there is no inventory.
with incremental quantity under price break k.
The upper bound order quantity for item j of supplier i
qik
with price break k.
ci Capacity (volume) per vehicle for supplier i in period t.
v Volume per unit of the item.

Variables:

TC Total cost for all planning periods.
QPit Purchase quantity from supplier i in period t.

CO Total ordering cost. (a) (b)

Purchase cost from supplier i with order quantity QPit in
CPit
period t.
CIt Total inventory cost in period t.

CSt Total shortage cost in period t.

Number of transportation vehicles from supplier i in
Nit
period t. (c)
IBt Beginning inventory level in period t.
IEt Ending inventory level in period t. Figure 1. Three Different Cases of Inventory Level
 Hesham K. Alfares and Rio Turnadi / Procedia CIRP 56 (2016) 199 202 201

h Constraints on the number of orders are given by:

CIt t ( IBt  IEt )  M u BEt (4)
2
QPit d MuFit, i = 1, , I, and t = 1, , T (12)
h( IBt2 )
CIt t  M (1  BEt  BBt ) (5)
2d t Since quantity discounts are offered, the model must ensure
that a unique a unique unit purchase cost is used, and that it is
The total shortage cost for each period (CS) is obtained by consistent with the order quantity (i.e. unit cost = pitk if qi,k1 <
multiplying the unit shortage cost with the average of the QPit d qik). Assuming either all-units or incremental quantity
beginning and ending inventory levels. There are three discounts, the following constraints are used:
possible cases, as shown in Figure 2. For case (a), ABt = 0,
and Eq. (6) is active. For case (b), ABt = 1, AEt = 0, and Eq. qi(k 1) + M(Uitk 1) < QPit d qik + M(1 Uitk),
(7) is active. For case (c), ABt = 1, AEt = 0, and both Eq. (6) i = 1, , I, k = 1, , K, and t = 1, , T (13)
and (7) are inactive because there is no shortage.
QPit d MuUitk, i = 1, , I, and t = 1, , T (14)
K

k 1
U itk Fit , i = 1, , I, and t = 1, , T (15)

The following constraints are needed to calculate the average
unit purchase cost under incremental quantity discounts.

D
Pitk pik  ik
,
QPit
(a) (b) i = i + 1, , I, t = 1, , T, and k = 1,, K (16)

K
D ik qik ( pi,k 1  pik ), i = i + 1, , I (17)
k' 1

Inventory balance constraints are listed below:
(c)
ABt + BBt = 1, t = 1, , T (18)
Figure 2. Three Different Cases of Shortages Level AEt + BEt = 1, t = 1, , T (19)

l IBt d MuABt, t = 1, , T (20)

CSt t ( SBt  SEt )  M u ABt (6) SBt d MuBBt, t = 1, , T (21)
2
IEt d MuAEt, t = 1, , T (22)
l ( SBt2 ) SEt d MuBEt, t = 1, , T (23)
CSt t  M (1  ABt  AEt ) (7)
2d t
Constraint (24) ensures demand satisfaction, while (25)
Since we have a deterministic objective function along with a specifies binary variables and non-negativity restrictions:
set of defined constraints, and because some variables should
I T T
be integer and some should be binary, then MIP is the best
choice for formulating the model. The objective function is to
i 1 t 1
QPit d t
t 1
(24)
minimize the total cost, which is the sum of cost components:
ordering + purchasing + transportation + shortage + holding. Fit, Uitk, ABt, BBt, AEt, BEt {0, 1}, and all variables t 0 (25)
T I
Min TC o F
t 1 i 1
i it  CPit  sit Nit  CIt  CSt (8) 3. Modified Silver-Meal Heuristic

To obtain solutions of realistic lot-sizing problems, a

Ending inventory constraints and beginning inventory modified Silver-Meal (MSM) heuristic was developed. The
constraints are given by (9) and (10), respectively: well-known SM heuristic finds the number of subsequent
periods to include in the given order to minimize the per-
IEt SEt = IBt SBt dt (9) period average cost of ordering and holding. The MSM rule,
I described below, allows an order for a given period to be
IBt SBt = IE(t 1) SE(t 1) + QPit (10) placed in later periods, and it minimizes average purchasing,
i 1 ordering, transportation, holding, and backordering costs.

Constraints relating the lot-size to transportation capacity and Step 1. Set t = 1. Choose the highest purchase cost among all
the number of vehicles are given by: the unit costs from all suppliers and all price breaks. Highest
cost = pmax = max(pik).
vuQPit d Nit ci, i = 1, , I, j = 1, , J, and t = 1, , T (11)
202 Hesham K. Alfares and Rio Turnadi / Procedia CIRP 56 (2016) 199 202
Step 2. From all unit costs provided by all suppliers in all
price breaks, subtract the highest unit cost found in step 1.
Subtracted cost = psik = pik pmax. This subtracted cost will be
either zero or a negative value, indicating the savings in unit
purchase cost if we order from a given supplier.
Step 3. Set n = 0. Assume the current time period is t, and the
order Qit covers periods t, , t + n. Total cost for each
supplier (TCit) is the sum of the subtracted purchasing cost,
ordering cost, and transportation cost, if an purchase is made
from the given supplier at time t.
Step 4. Based on the results in step 3, choose the supplier with
the least cost (TCit).
Step 5. Add the holding cost for periods t, , t + n to the
minimum (TCit) of step 4 and divide it by the number of
periods (n + 1) to determine the average cost per period.
Step 6. Repeat steps 3 to 5, after including the demand for the
next period (i.e. let n = n + 1) in the current order. Keep
adding the next periods, one by one, as long as the minimum
average cost (of step 5) is decreasing. Stop as soon as the
average cost per period begins to increase.
Step 7. Confirm the stopping decision by comparing two
options: (i) stopping immediately and proceeding to the next
order, or (ii) adding one more period (i.e. let n = n + 1) to
cover the current order. Add this period if the holding cost
increase is less than the savings; otherwise, stop the iteration.
Step 8. The algorithm terminates the current iteration (order)
according to two stopping criteria, whichever comes first: (i)
the average cost per period is increasing, or (ii) capacity
constraints (11) are exceeded.
Step 9. Instead of placing the current order in period t to cover
demands for periods (t, , t + n), place it in later periods (t +
1, , t + n). Compare the added backordering cost with the
savings in inventory cost. Compare all possible ordering times
and choose the time that gives the minimum total cost.
Step 10. Restart the algorithm from the next time period (let t
= t + n + 1), repeating steps 3 to 9 until all demand periods
are covered.
4. Numerical Examples
Data for the numerical examples was adapted from Lee et
al. [4]. To incorporate the additional aspects of our model, we
added the following data values, which are the same for all
suppliers: shortage cost per period (l) = 0.12, volume of each
item (v) = 3, and vehicle volume capacity (ci) = 75. The
example considers both all-units and incremental quantity
discounts, and two demand scenarios (scenario 1 for 10
periods, and scenario 2 for 20 periods). Four suppliers (A, B,
C, D) are assumed, with different purchase costs and discount
schemes.
Using LINGO 9.0., the MIP model produced a scenario
1 solution whose total cost is $38,985.39. Using MATLAB,
the MSM heuristic produced a similar scenario 1 solution,
with a total cost of $39,004.40. However, while the MIP
model needed 13 hours of computation, the MSM took only
two seconds. Next, we attempted to solve the 20-period
scenario 2, but LINGO was not able to solve the MIP
problem. In contrast, the MSM heuristic produced a solution
whose total cost is $60,442.3 in a computation time of only
three seconds.
5. Conclusions
This paper presented an optimization model and a heuristic
solution algorithm for single-item lot-sizing considering
multiple suppliers, quantity discounts, and backordering of
shortages. The model considers the costs of ordering,
inventory holding, transportation, and shortages. The problem
was formulated as a mixed-integer programming (MIP)
model. LINGO was used to obtain the optimal solution for
small examples, but the MIP model becomes too difficult to
solve for practical large-size problems. Therefore, a heuristic
method, based on the Silver-Meal algorithm, was proposed to
produce good solutions in reasonable computational times.
The heuristic method provides near-optimal solutions, and it
is much faster than the MIP method.
References
[1] I. Okhrin and K. Richter, The linear dynamic lot size problem with
minimum order quantity, Int. J. Prod. Econ., vol. 133, no. 2, pp. 688
693, Oct. 2011.
[2] M. R. Wagner, Online lot-sizing problems with ordering, holding and
shortage costs, Oper. Res. Lett., vol. 39, no. 2, pp. 144149, Mar. 2011.
[3] C. Chu, F. Chu, J. Zhong, and S. Yang, A polynomial algorithm for a
lot-sizing problem with backlogging, outsourcing and limited inventory,
Comput. Ind. Eng., vol. 64, no. 1, pp. 200210, Jan. 2013.
[4] A. H. I. Lee, H.-Y. Kang, C.-M. Lai, and W.-Y. Hong, An integrated
model for lot sizing with supplier selection and quantity discounts, Appl.
Math. Model., vol. 37, no. 7, pp. 47334746, Apr. 2013.
[5] H. Jung and C. M. Klein, Optimal inventory policies for profit
maximizing EOQ models under various cost functions, Eur. J. Oper.
Res., vol. 174, no. 2, pp. 689705, Oct. 2006.
[6] R. V. Ramasesh, Lot-sizing decisions under limited-time price
incentives: A review, Omega, vol. 38, no. 34, pp. 118135, Jun. 2010.
[7] A. K. Maiti, A. K. Bhunia, and M. Maiti, An application of real-coded
genetic algorithm (RCGA) for mixed integer non-linear programming in
two-storage multi-item inventory model with discount policy, Appl.
Math. Comput., vol. 183, no. 2, pp. 903915, Dec. 2006.
[8] M. J. Drake and D. W. Pentico, Price discounts for increased
profitability under partial backordering, Int. Trans. Oper. Res., vol. 18,
no. 1, pp. 87101, Jan. 2011.
[9] L. A. San-Jos and J. Garca-Laguna, Optimal policy for an inventory
system with backlogging and all-units discounts: Application to the
composite lot size model, Eur. J. Oper. Res., vol. 192, no. 3, pp. 808
823, Feb. 2009.
[10] L. A. San-Jos, J. Sicilia, and J. Garca-Laguna, An economic lot-size
model with partial backlogging hinging on waiting time and shortage
period, Appl. Math. Model., vol. 31, no. 10, pp. 21492159, Oct. 2007.
[11] L. A. San Jos, J. Sicilia, and J. Garca-Laguna, Analysis of an
inventory system with exponential partial backordering, Int. J. Prod.
Econ., vol. 100, no. 1, pp. 7686, Mar. 2006.
[12] C. Toews, D. W. Pentico, and M. J. Drake, The deterministic EOQ and
EPQ with partial backordering at a rate that is linearly dependent on the
time to delivery, Int. J. Prod. Econ., vol. 131, no. 2, pp. 643649, Jun.
2011.