Professional Documents
Culture Documents
A R T I C LE I N FO A B S T R A C T
Keywords: The optimal decision of global warehouse/hub deployment is critical to firm profitability. This study resolves the
Warehouse configuration problems of deciding the optimal warehouse location for multiple markets and determining warehouse con-
Inventory control figuration design against stochastic demands. An appropriate inventory policy with owned and rented ware-
Stochastic demands houses for deteriorating items is further designed. The proposed model maximizes the profit under rent ware-
Deteriorating items
house incentives decreasing over time and price-sensitive demands. Furthermore, we propose a solution
algorithm to solve the problem effectively. Sensitivity analysis is conducted to examine the effects of the
parameters for the model of the algorithm.
⁎
Corresponding author at: Department of Industrial Management, National Taiwan University of Science and Technology, No.43, Keelung Rd., Sec.4, Da'an Dist., Taipei City 10607,
Taiwan, ROC.
E-mail addresses: yslin@mail.ntust.edu.tw (Y.-S. Lin), kjwang@mail.ntust.edu.tw (K.-J. Wang).
https://doi.org/10.1016/j.cie.2018.04.008
Received 3 May 2015; Received in revised form 1 January 2018; Accepted 4 April 2018
Available online 06 April 2018
0360-8352/ © 2018 Elsevier Ltd. All rights reserved.
Y.-S. Lin, K.-J. Wang Computers & Industrial Engineering 120 (2018) 83–93
stochastic demand, as well as inventory policy for managing deterior- with finite replenishment rate and unequal length of cycle time. Jaggi,
ating items. Therefore, this study develops a model to select locations Pareek, Khanna, and Sharma (2015) developed an inventory model of
and warehouse configurations for serving multiple markets against two warehouses with price dependent demand with inflation. In which,
stochastic demand and derive an inventory policy for managing dete- they also investigated the effect of FIFO and LIFO policies in the model.
riorating items accordingly. The effect of sale price on demand in In the scenario where FIFO policy is applied, goods in OW will be used
various locations is considered by the parameter of a probability to fulfill customer demand first. In the scenario where LIFO policy is
function. Warehouse rental is decreased to encourage long-term part- applied, goods in RW will be used to fulfill customer demand first. The
nership with RWs. This study aims to maximize total profit of the entire result of their investigation is the application of FIFO or LIFO policies is
system for managing stochastic demands and deteriorating items. only affected by holding cost and the deteriorating rates.
The rest of the paper is organized as follows. A relevant literature Many researchers take into account two-warehouse with deterior-
review is discussed in Section 2. Section 3 details the mathematical ating items. Sarma (1983) developed a two-warehouse model for de-
model under stochastic demand. A hybrid solution algorithm is pro- teriorating items with infinite replenishment rates and shortages.
posed in Section 4. Section 5 illustrates numerical examples and sen- Pakkala and Achary (1991) developed a two-warehouse probabilistic
sitivity analyses. Finally, conclusions are provided in Section 6. order-level inventory model for deteriorating items. Yang (2004) de-
veloped a two-warehouse inventory model with constant deteriorating
2. Literature review items, a constant demand rate, and complete shortages. Singh, Kumar,
and Kumari (2009) developed an inventory policy for a deteriorating
2.1. Location and allocation of facilities item with two separate warehouses, that is, an OW with finite dimen-
sion and an RW with infinite dimension, under inflation and time value
The location and allocation of facilities are important issues because of money. Ghiami, Williams, and Wua (2013) investigated a two-
they significantly affect long-term investments of firms. Researchers echelon supply chain model for deteriorating inventory in which the
have achieved progress on this problem with respect to representations demand rate in the retailer is stock-dependent and in case of any
and solutions. Laporte, Louveaux, and Hamme (1994) considered a shortages, the demand is partially backlogged. Yadav et al. (2015)
class of capacitated facility location problems in which customer de- studied inventory model of two warehouses for deteriorating items with
mands are stochastic. Snyder (2006) reviewed the literature on sto- time-dependent demand and variable holding cost. Yu et al. (2015)
chastic and robust facility location models. He mentioned that the two- examined inventory control for deteriorating items in a two-warehouse
stage nature of facility location problems makes these problems ex- setting with an incentive policy offered by a rented warehouse that
tremely attractive to researchers who are exploring approaches to de- allows the rental fee to decrease over time.
cision making with uncertainties. Wang, Makond, and Liu (2011) ad-
dressed facility location and task allocation problems of a two-echelon 2.3. Deteriorating items
supply chain against stochastic demand. Gendron, Khuong, and Semet
(2016) considered the two-level uncapacitated facility location pro- Research on inventory management of deteriorating items is also
blem, arising in industrial applications in freight transportation and important because the deterioration of stock items considerably affects
telecommunications, with single-assignment constraints. industries. Deterioration is generally defined as the decay, damage,
The complexity of this problem could be very high to model and spoilage, dryness, and vaporization of products. The two broad classi-
solve, due to the inclusion of different decisions, various type of deci- fications of inventory subject to decay are (i) items with a fixed lifetime
sion variables, and large-scale scenarios. Survey data from literature and (ii) items with age-independent ongoing deterioration (Cohen,
also indicate that the addressed location-allocation problems are highly 1974; Nahmias, 1978). Items with a fixed lifetime (such as foodstuff,
complex. Laporte et al. (1994) formulated as two-stage stochastic pro- green vegetables, human blood, photographic film, and so on) and a
gramming, with binary variables in the first stage and continuous maximum usable lifetime are known as perishable products. Items with
variables in the second stage. The problems were then solved to achieve age-independent ongoing deterioration, such as volatile liquids (al-
optimality through a branch-and-cut method. Ignacio, Filho, and cohol, gasoline, and radioactive substances), and those without shelf
Galvãoa (2008) used the lower and upper bounds of Lagrangean re- life are known as decaying products. The decrease in volatile liquids
laxation that combine CPLEX tool and Tabu search to determine the can be attributed to evaporation, whereas the decrease in grain in-
numbers, locations, and allocations of concentrators and routers in ventory can be ascribed to pests or pilferage. The latter group of pro-
computer networks. Wang et al. (2011) solved the problem by bi-level ducts generally does not have an expiration date and can be stored
stochastic programming, and presented a genetic algorithm (GA) with indefinitely, although some of these products may undergo natural at-
efficient greedy heuristics. Manatkar et al. (2016) used both Particle tenuation while in inventory. However, their unit value remains basi-
Swarm Optimization (PSO) and GA to solve the problem. Researchers cally the same.
have devoted considerable effort in developing optimization-based Several authors have studied commodities with age-independent
approaches that can help solve these problems. However, such ap- ongoing deterioration. Ghare and Schrader (1963) developed an
proaches are inefficient in a finite-time context, and several applica- equation for an exponentially decaying inventory. They observed that
tions using soft computing remain in their infancy. commodities such as volatile liquids (alcohol, gasoline, and radioactive
substances) shrink with a certain proportion of time, which can be
2.2. Alternative warehouses approximated by a negative exponential function of time. Covert and
Philip (1973) extended the model of Ghare and Schrader (1963) to
In practice, most enterprises store goods in alternative warehouses consider deterioration with a two-parameter Weibull distribution.
to increase flexibility. For example, they purchase more goods during Goyal and Giri (2001) and Bakker, Riezebos, and Teunter (2012) re-
peak seasons, then store them in their OW, whereas excess quantities viewed the advances in the deteriorating inventory literature since
are often stored in an additional RW. These enterprises strategically 1990 and 2001, respectively. Taleizadeh (2014) developed EOQ model
outsource warehousing activities. Researchers have shown interest in for a deteriorating product with and without shortage under con-
this field of study, and numerous companies are facing this critical issue secutive prepayments which allow purchasers to pay all or a fraction of
in practice. Hartely (1976) was the first author to consider the effect of the purchase cost in advance.
a two-warehouse model in inventory research and who developed an Many studies have revealed the significance of further investigation
inventory model with an RW storage policy. Kumar Sett, Sarkar, and and research on facility location and allocation problems in managing
Goswamic (2012) considered an inventory model of two warehouses inventory of decaying products. However, only a few studies
84
Y.-S. Lin, K.-J. Wang Computers & Industrial Engineering 120 (2018) 83–93
Stochastic Two-Stage Location-Allocation Inventory System rate of market i in scenario k. Assume that demand rates
follow a probability distribution function as a parameter of ρi .
Stage I: Location & Warehouse Configuration Design B(Wi, di) The average setup cost function in OW for market i with
Determine the Location of warehouses and warehouse configurations storage capacity, Wi, and demand, di.
Configuration decisions Inventory policies R(di) The setup cost function in RW for market i with demand, di.
Stage II: Inventory Policy Design Customer with BG The total budget for overall setup cost of OWs and RWs.
Stochastic Demand ED The mathematical expectation with respect to demand state
. .
q1 d1 D.
OW RW
. .
qi OW RW di
Oi Ordering cost per ordering in market i ($/ordering).
bi The unit administration cost of RW in market i ($/unit/year).
HOi The unit holding cost of OW in market i ($/unit/year).
qn OW RW dn HRi The unit holding cost of RW in market i ($/unit/year), HRi > bi.
Pi The unit deteriorating cost in market i ($/unit).
θOi Deterioration rate of goods of OW in market i.
Fig. 1. Conceptual research framework. θRi Deterioration rate of goods of RW in market i.
U The unit purchasing price.
simultaneously consider the inventory policy and the warehouse loca-
tion and configuration, especially in a two-warehouse case. Several The following notations of decision variables for the first stage are
decision-making issues for location and allocation involved in ware- used:
house configuration about a two-warehouse setting also remain un-
clear, particularly when an incentive policy is incorporated by RWs. xi xi = 1 if an OW is built for market i; otherwise xi = 0.
yi yi = 1 if an RW is rented for market i; otherwise yi = 0.
3. Two-stage model that considers stochastic demand (xi, yi) is a pair notation of the warehouse configuration, as listed in
Table 1.
A two-stage model is developed as a mixed-integer nonlinear pro- Wi Storage capacity of OW for market i.
gramming for location–allocation decision and inventory policy for
deteriorating items. Warehouse configuration in terms of two-ware- The following notations of decision variables for the second stage
house settings for multiple markets is determined in the first stage. The are used:
total budget for the overall setup cost of alternative warehouses is
confined. If an OW is decided to be established, then the storage ca- qi Ordering lot size for market i.
pacity of the OW is assigned further. In the second stage, an inventory tRi Usage duration of RW for market i.
policy is developed for the corresponding warehouse configuration tOi Usage duration of OW for market i.
determined in the first stage. In addition, the use of RWs is considered Ti Replenishment cycle time for market i.
when rental fee decreases with the length of storage, and ordering lot
size is modified by the conditions. A conceptual framework is shown in Since tRi, tOi and Ti are inter-dependent on qi, the inventory policy can
Fig. 1. Our objective is to determine the optimal warehouse config- be defined as qi.
uration and storage capacity of an OW (that is, the location issue) at the
first stage and the optimal inventory policies for managing warehouses
3.1. First stage: determining warehouse location and configuration
and deteriorating items (that is, the allocation issue) on the second
stage.
The location and warehouse configuration problem on whether and
The mathematical model developed in this study is based on the
where to build OW/RW for markets is modeled in the first stage. The
following assumptions. (i) OWs have a limited storage capacity because
objective function (1) maximizes the expected profit of K demand
of budget constraint. By contrast, RWs have no storage capacity limit. A
scenarios by initially adding the setup costs of OW and RW for all
firm can rent alternative warehouses to satisfy its storage requirement.
markets to the inventory cost, and then subtracting the sum from the
(ii) RWs provide an incentive policy in which warehouse rental ex-
gross profit. ED is the mathematical expectation of the total profit with
ponentially decreases over rental (usage) time. Warehouse rental is
respect to K demand scenarios.
usually charged by usage time to encourage firms to rent for a long
duration. An RW owner specifically proposes an incentive policy that I I
⎛ ⎞
rental fee will decrease by ui% per unit of time between timin and timax , Minimize ED ⎜∑ x i B (Wi ,di ) + ∑ yi R (di )−L (x ,y,W ,D) ⎟
where timin is the least amount of storage time desired by the RW owner, ⎝ i=1 i=1 ⎠ (1)
and timax is the amount of time reached by the rental bi (Yu et al., 2015).
(iii) The products have deteriorated property, but the replacement of
deteriorated products is not considered (Yang, 2004). Table 1
Warehouse configuration representation.
The following notations of the parameters for the first stage mod-
eling are used: (xi, yi) Warehouse configuration
85
Y.-S. Lin, K.-J. Wang Computers & Industrial Engineering 120 (2018) 83–93
where L (x ,y,W ,D) is the total inventory cost provided by the second Inventory Level
stage.
Subjectto x i ∈ {0,1}, yi ∈ {0,1}, 0 ⩽ Wi ⩽ x i di qi
(2) IRi(t )
I I
IOi(t )
∑ x i B (Wi ,di ) + ∑ yi R (di ) ⩽ BG Wi
i=1 i=1 (3)
Li
Constraint (3) is a limitation of the total budget for the whole setup cost
in OWs and RWs.
, 0 ⩽ t ⩽ timin qi
⎧ HRi
⎪ min )
hi (t ) = HRi ·e −λi (t − ti , timin ⩽ t ⩽ timax
⎨ max min
⎪ bi = HRi ·e−λi (ti − ti ) , timax < t (4)
⎩
I Oi (t)
where timin is the least amount of storage time desired by the RW owner,
timax is the longest amount of storage time desired by the RW owner and
λi is given by λi = −ln(1−ui /100) . The relation of hi(t) to various times is
depicted in Fig. 2.
Unit holding
cost t Ri
Ti Time
th
Fig. 5. Inventory level as the i market uses RW only (i.e., xi = 0 and yi = 1).
di θRi (tRi− t )
IRi (t ) = (e −1), 0 ⩽ t ⩽ tRi, if yi = 1
min
θRi (6)
i (tRi ti )
H Ri e
The corresponding time is
1 ⎡ (qi−x i Wi ) θRi
tRi = ln + 1⎤, if yi = 1
θRi ⎢ ⎣ di ⎥
⎦ (7)
The Eq. (6) can be combined as
yi ⎡ (qi−x i Wi ) θRi
t imin timax Time tRi = ln + 1⎤
θRi ⎢ ⎣ di ⎥
⎦ (8)
Fig. 2. Unit holding cost decreases with time. th
IOi(t) is the inventory level in OW of the i market at time t. The change
86
Y.-S. Lin, K.-J. Wang Computers & Industrial Engineering 120 (2018) 83–93
tRi
in inventory level of OW during an infinitesimal time can be formulated
as
(
DCRi = Pi IRi (0)− ∫0 )
di dt = Pi (qi−x i Wi −di tRi )
(17)
( )
dIOi (t ) tOi
dt
= −θOi IOi (t ), 0 ⩽ t ⩽ tRi, if yi = 1,x i = 1
(9)
DCOi = x i Pi IOi (0)−IOi (tRi ) + IOi (tRi )− ∫t Ri
di dt
x i ⎡ Li θOi The problem that includes integer and continuous decision variables
tOi = x i tRi + ln + 1⎤
θOi ⎢ ⎣ di ⎥
⎦ (14) is highly complex; thus, we develop a hybrid solution algorithm that
uses a GA for the first stage and a gradient method for the second stage,
The RW holding cost in the ith market during the period Ti can be
as shown in Fig. 6. The GA is developed with a repair mechanism. The
derived as
proposed GA consists of making following two decisions. (i) What kind
t
Ri min of warehouse configuration should be established, and (ii) what the
⎧ HRi ∫0 IRi (t ) dt , 0 ⩽ tRi ⩽ ti
⎪ storage capacity of the OW should be designed. This study uses a ge-
min
⎪ H ⎛∫ti I (t ) dt + ∫tRi I (t )·e−λi (t − timin) dt ⎞, t min < t ⩽ t max netic search strategy to identify the most appropriate values for the
⎪ Ri ⎝ 0 Ri timin Ri i Ri i
⎪ ⎠ decision variables in the first stage and a gradient method to calculate
HCRi = min max
continuous variables in the second stage. The general procedures for the
⎨ HRi ⎛∫ti IRi (t ) dt + ∫minti min
IRi (t )·e−λi (t − ti ) dt
⎪ ⎝ 0 ti proposed GA and gradient method are presented in Figs. 7 and 8, re-
⎪ spectively.
⎪ max min tRi
+ e−λi (ti − ti ) ∫t max IRi (t ) dt ⎞, timax < tRi
⎪ i ⎠
⎩
4.1. GA implementation
(15)
The OW holding cost in the ith market during the planning horizon 4.1.1. Chromosome design and fitness value
Ti is Each chromosome is cascaded by three sets of genes, namely, the
binary codes of xi (build an OW) and yi (rent an RW), and the real
HCOi = HOi (∫ 0
tRi
IOi (t ) dt + x i ∫t
Ri
tOi
IOi (t ) dt ) (16)
number of Wi (storage capacity of the OW). A two-dimensional chro-
mosome matrix, A, is constructed to conform to its domain applications,
th
The deteriorating cost of the i market during the planning horizon as shown in Fig. 9. A population has P chromosomes, and each chro-
Ti can be expressed as mosome has K demand scenarios (Fig. 10).
Step 3: Evolution
Step 2: Evaluation
No Reproduction
Step 1 Constrained
Initialization Gradient
Initialize method for Yes Crossover
population Stage II
(for Stage I End
variables) Mutation
Evaluate the
whole system Next generation
Repair
87
Y.-S. Lin, K.-J. Wang Computers & Industrial Engineering 120 (2018) 83–93
The fitness function of the proposed algorithm is assigned as Eq. proposed GA. Two random numbers between 1 and K are selected, and
(21). The difference between Eqs. (21) and (1) lies in the evaluation of each corresponding substring between the selected points is swapped to
the expected value of Eq. (1). The expected value in this study adopts render two child chromosomes, as illustrated in Fig. 11. The mutation
the sample average approximation (SAA) scheme that assumes K de- chromosomes are selected, according to the mutation rate. A gene of the
mand scenarios. Then, Eq. (1) is approximated by the SAA function of selected mutation chromosome is a uniform mutation (also refer to Gen
Eq. (21). & Cheng, 2000; Haupt & Haupt, 1998; and Wang, Wang, & Chen, 2008).
n n
When a chromosome is operated by mutation operator, it becomes
1⎛ ⎞ infeasible in Wi against Constraints (2) or (3). Thus, a repairing op-
Maximize ∑ xi B (Wi ,di) +
K ⎜ i=1
∑ yi R (di )−L (x ,y,W ,D) ⎟
⎝ i=1 ⎠ (22) eration is required. If a chromosome completes the process but does not
satisfy Constraint (2), then the values of Wi are set as the nearest value
K demand scenarios are randomly selected from a lognormal distribu- fitted in the constraint interval. If Constraint (3) is not satisfied, a lo-
tion as customer demand in each generation. The sampling procedure is cation is randomly chosen to reduce the storage capacity or abandon
defined as D ∼ LogN(μ, σ2), where μ and σ2 are the mean and variance RW until Constraint (3) is fitted.
of customer demands, respectively. Lognormal distributions are well
suited for modeling stochastic economic variables, such as demands
(Kamath & Pakkala, 2002). 4.2. Gradient method for the second stage
4.1.2. Crossover, mutation, and repair mechanism HCRi values differ with respect to tRi intervals, thus the second stage
Fitness-proportionate selection (Beasley, Bull, & Martin, 1933; is a piecewise continuous optimization problem. Thus, the problem
Jadaan, Rajamani, & Rao, 2005) is implemented to select the chromo- must be solved by dividing TCi in terms of tRi, which is referred to as
somes (the structure is shown in Fig. 10) for the operation of crossover. TPi. A simple constrained optimization procedure for solving TPi(qi) is
A two-point crossover operation (Chambers, 1995) is applied to the provided in Fig. 8.
88
Y.-S. Lin, K.-J. Wang Computers & Industrial Engineering 120 (2018) 83–93
For j=a to c
* dTCi j (qij ) j *
(i) Find the optimal q i j by 0 for each TPi . If q i j conforms
dqij
j
to the constraint of each TPi (i.e., Formula B2, B4, and B6), then go to
*
(iii) Substitute q i j into Equation (19).
End for j
*
The global optimal is the maximum among TC i j ( q i j ) .
89
Y.-S. Lin, K.-J. Wang Computers & Industrial Engineering 120 (2018) 83–93
Table 2 Table 3
Parameters setting. The best solution.
Parameter Description Value sampled from The first stage The second stage
Table 4
5.2. Solution Factorial design of experiments for the parameters of GA.
Factor Level
Fig. 12 indicates the time consumed by our hybrid algorithm. The
proposed algorithm achieves the best solution −864,008.76 at the 33rd −1 (Low) +1 (High)
generation and consumes 7795 s. The best solution for each decision
Population size 30 100
variable is shown in Table 3. Crossover rate 0.6 0.85
Mutation rate 0.05 0.1
Termination condition 50 100
5.3. Sensitivity analysis on algorithmic parameters
This section examines the effects of the parameters on the proposed −868,763.10 is population size = 100, crossover rate = 0.6, and mu-
algorithm through sensitivity analysis. A two-level factorial design, tation rate = 0.05. The GA is terminated when the fitness no longer
including 24 = 16 combinations, is constructed. Several vital factors produces better results after 100 iterations.
that influence the fitness value and time consumed are identified. The Through ANOVA, and using the experimental results in Table 5 as
size of the problem being investigated is five demand scenarios and ten basis, population size and crossover rate are determined to be sig-
candidate markets. Table 4 depicts a 4-factor and 2-level factorial de- nificant factors that affect the fitness value [Table 6(a)], whereas po-
sign. Other parameters are the same as those in Section 5. The results, pulation size and terminal condition are factors that significantly affect
including the fitness value, time consumed, and GA generation for each time consumed [Table 6(b)]. Although a higher population size sig-
experiment of the factorial design, are shown in Table 5. The best nificantly contributes to the fitness value, it is also time consuming [the
combination of the parameters with the highest fitness value of
Generation
5
x 10 22 44 66 88 110 132
-6.8
-7
-7.2
-7.4
-7.6
Fitness
-7.8
-8
-8.2
-8.4
-8.6
-8.8
0 0.5 1 1.5 2 2.5 3
4
CPU time (Seconds) x 10
Fig. 12. Fitness evolution of the proposed hybrid algorithm.
90
Y.-S. Lin, K.-J. Wang Computers & Industrial Engineering 120 (2018) 83–93
Table 5
The results for each experiment of factorial design.
Run Population size Crossover rate Mutation rate Terminal condition Fitness value (best solution) Time consumeda GA generation achieved
a
In seconds.
Table 6 Table 7
Analysis of Variance (ANOVA). The results from different standard deviation.
Source DF SS MS F p-value σ Fitness valuea, F1 Fitness valueb, F2 TDPc PTDPd
91
Y.-S. Lin, K.-J. Wang Computers & Industrial Engineering 120 (2018) 83–93
Fig. 13. The fitness histogram by different standard deviations of the demand.
policy. By adopting the proposed model and solution approach, firms than 0.6 because a higher crossover rate can provide higher fitness
have a potential to optimize the location for their warehouses/hubs value without increasing the amount of time consumed. Crossover rate
strategically, lower operational cost, and achieve high agility. Further, has few effect on the time consumed, but the fitness value significantly
the inherent complexity of the addressed problem is high. Our proposed increases as crossover rate increased. Therefore, we suggest that
two-stage programming enables efficient and effective selection of the crossover rate should be tuned to obtain efficient fitness and consuming
warehouse location and configuration to serve multiple markets against time.
stochastic demands as well as optimize inventory policy to manage Several interesting extensions of this research can be conducted in
items with deterioration, which commonly exist, yet unsolved in in- the future. First, the optimal parameter setting of the proposed GA is
dustry. Industries that need dual warehouses, such as global logistics, worthy to be fully examined. Second, other significant characteristics
international trading, and large-scale manufacturing firms, will benefit commonly occurred in the industry, such as decreasing price and per-
from our proposed method if the model is appropriately adopted. iodical demand, can be further considered in the model. Third, a multi-
The experiments reveal the several parameters are significant. stage programming for a multi-echelon supply chain system is worthy
Appropriate selection on the parameter will facilitate the usage of the of being further studied by extending the proposed model.
proposed model and solution method. The best solution, −864,008.76,
is found, in which our experiment suggests optimal settings of popu-
lation size with 30, crossover rate with 0.85, and mutation rate with Acknowledgement
0.05. These parameters are optimized by using design of experiment.
The population size of P = 30 is preferred than P = 100 because higher The authors gratefully acknowledge the comments and suggestions
population size provides higher fitness value, but it is time-consuming. of the editor and the anonymous referees. This work is partially sup-
The mutation rate neither affects the fitness value nor time consumed; ported by the Ministry of Science and Technology of the Republic of
thus, mutation rate = 0.05, which provides the highest fitness value in China (Taiwan) under the grant number MOST 106-2917-I-564-016
the experimental design, is chosen. Crossover rate = 0.85 is preferred and 105-2221-E-011-106-MY2.
According to this RW incentive policy, the following lemma regarding timax is proved.
Lemma 1.
1 ⎛ HRi ⎞
timax = ln ⎜ + timin ⎟
λi ⎝ bi ⎠
Proof.. An incentive policy of RW is stopped by timax . Hence, the unit holding cost of RW, hi(t), at time timax is
max − t min )
HRi ·e−λi (ti i = bi
One has
1 ⎛ HRi ⎞
timax = ln ⎜ + timin ⎟ □
λi ⎝ bi ⎠
Appendix B. Computing the bounds of qi by converting the bounds of tRi for each interval
92
Y.-S. Lin, K.-J. Wang Computers & Industrial Engineering 120 (2018) 83–93
1 ⎡ (qi−Wi ) θRi
tRi = ln + 1⎤ ⩽ timin, if yi = 1
θRi ⎢ ⎣ di ⎥
⎦ (A1)
Thus,
min θ
di (eti Ri−1)
qi ⩽ + Wi , if yi = 1
θRi (A2)
Interval 2 (timin < tRi ⩽ timax ): Substitute Eq. (4) into timin < tRi ⩽ timax as
1 ⎡ (qi−Wi ) θRi
timin < tRi = ln + 1⎤ ⩽ timax, if yi = 1
θRi ⎢ ⎣ di ⎥
⎦ (A3)
Thus,
min θ max θ
di (eti Ri−1) di (eti Ri−1)
+ Wi < qi ⩽ + Wi , if yi = 1
θRi θRi (A4)
Interval 3 (timax < tRi ): Substitute Eq. (4) into timax < tRi as
1 ⎡ (qi−Wi ) θRi
tRi = ln + 1⎤ > timax, if yi = 1
θRi ⎢ ⎣ di ⎥
⎦ (A5)
Thus,
max θ
di (eti Ri−1)
qi > + Wi , if yi = 1
θRi (A6)
References Manatkar, R. P., Karthik, K., Kumar, S. K., & Tiwari, M. K. (2016). An integrated inventory
optimization model for facility location-allocation problem. International Journal of
Production Research, 54(12), 3640–3658.
Angalakudati, M., Balwani, S., Calzada, J., Chatterjee, B., Perakis, G., Raad, N., & Melo, M. T., Nickel, S., & Saldanha-da-Gama, F. (2009). Facility location and supply chain
Uichanco, J. (2014). Business analytics for flexible resource allocation under random management – A review. European Journal of Operational Research, 196, 401–412.
emergencies. Management Science, 60(6), 1552–1573. Nahmias, S. (1978). Perishable inventory theory: A review. Operation Research, 30(4),
Bakker, M., Riezebos, J., & Teunter, R. H. (2012). Review of inventory systems with 680–708.
deterioration since 2001. European Journal of Operational Research, 221, 275–284. Pakkala, T. P. M., & Achary, K. K. (1991). A two warehouse probabilistic order level
Beasley, D., Bull, D., & Martin, R. (1933). An overview of genetic algorithms: Part 1. inventory model for deteriorating items. Journal of the Operational Research Society,
Fundamentals, University Computing, 2, 58–69. 42, 1117–1122.
Chambers, L. (1995). Practical handbook of genetic algorithms: Applications (1st ed.). Boca Porter, M. E. (2000). Location, competition, and economic development: Local clusters in
Raton, FL: Chapman & Hall/CRC Press106–113. a global economy. Economic Development Quarterly, 14(1), 15–34.
Cohen, M. A. (1974). Inventory control for a perishable product optimal critical number or- Sadegheih, A. (2006). Scheduling problem using genetic algorithm, simulated annealing
dering and applications to blood inventory management. Unpublished Ph.D. dissertation. and the effects of parameter values on GA performance. Applied Mathematical
Northwestern University. Modelling, 30, 147–154.
Covert, R. P., & Philip, G. C. (1973). An EOQ model for items with Weibull distribution Sarma, K. V. S. (1983). A deterministic inventory model with two levels of storage and an
deterioration. AIIE Transaction, 5, 323–326. optimum release rule. Opsearch, 20, 175–180.
Gen, M., & Cheng, R. (2000). Genetic algorithms and engineering optimization. John Wiley & Singh, S. R., Kumar, N., & Kumari, R. (2009). Two-warehouse inventory model for de-
Sons Inc. teriorating items with shortages under inflation and time-value of money.
Gendron, B., Khuong, P. V., & Semet, F. (2016). A Lagrangian-based branch-and-bound International Journal of Computational and Applied Mathematics, 4(1), 83–94.
algorithm for the two-level uncapacitated facility location problem with single-as- Snyder, L. V. (2006). Facility location under uncertainty: A review. IIE Transactions, 38,
signment constraints. Transportation Science, 50(4), 1286–1299. 537–554.
Ghare, P. M., & Schrader, S. F. (1963). A model for an exponentially decaying inventory. Taleizadeh, A. A. (2014). An economic order quantity model for deteriorating item in a
Journal of Industrial Engineering, 14, 238–243. purchasing system with multiple prepayments. Applied Mathematical Modelling,
Ghiami, Y., Williams, T., & Wua, Y. (2013). A two-echelon inventory model for a dete- 38(23), 5357–5366.
riorating item with stock-dependent demand, partial backlogging and capacity con- Tsao, Y. C., Mangotra, D., Lu, J. C., & Dong, M. (2012). A continuous approximation
straints. European Journal of Operational Research, 231, 587–597. approach for the integrated facility-inventory allocation problem. European Journal of
Goyal, S. K., & Giri, B. C. (2001). Recent trends in modeling of deteriorating inventory. Operational Research, 222, 216–228.
European Journal of Operational Research, 134, 1–16. Wang, K. J., Makond, B., & Liu, S. Y. (2011). Location and allocation decisions in a two-
Hartely, V. R. (1976). Operations research—A managerial emphasis. California: Good Year. echelon supply chain with stochastic demand – A genetic-algorithm based solution.
Haupt, R. L., & Haupt, S. E. (1998). Practical genetic algorithms. Canada: John Wiley & Expert Systems with Applications, 38(5), 6125–6131.
Sons. Wang, K. J., & Nguyen, P. H. (2017). Capacity planning with technology replacement by
Ignacio, A. A. V., Filho, V. J. M. F., & Galvãoa, R. D. (2008). Lower and upper bounds for a stochastic dynamic programming. European Journal of Operational Research, 260,
two-level hierarchical location problem in computer networks. Computers and 739–750.
Operations Research, 35(6), 1982–1998. Wang, K. J., Wang, S. M., & Chen, J. C. (2008). A resource portfolio planning model using
Jadaan, O. A., Rajamani, L., & Rao, C. R. (2005). Improved selection operator for GA. sampling-based stochastic programming and genetic algorithm. European Journal of
Journal of Theoretical and Applied Information Technology, 4(4), 269–277. Operational Research, 184, 327–340.
Jaggi, C. K., Pareek, S., Khanna, A., & Sharma, R. (2015). Two-warehouse inventory Yadav, A. S., Gupta, K., Garg, A., & Swami, A. (2015). A two warehouse inventory model
model for deteriorating items with price-sensitive demand and partially backlogged for deteriorating items with shortages under genetic algorithm and PSO. International
shortages under inflationary conditions. International Journal of Industrial Engineering Journal of Emerging Trends & Technology in Computer Science, 4(5(2)), 40–48.
Computations, 6, 59–80. Yang, H. L. (2004). Two-warehouse inventory models for deteriorating items with
Kamath, K. R., & Pakkala, T. P. M. (2002). A Bayesian approach to a dynamic inventory shortage under inflation. European Journal of Operational Research, 157, 344–356.
model under an unknown demand distribution. Computers and Operations Research, Yang, C., & Cheng, S. (2003). Myths of OBM and OEM – A case of ACER. Chung Hua
29, 403–422. Journal of Management, 4(1), 89–100 (In Chinese).
Kumar Sett, B., Sarkar, B., & Goswamic, A. (2012). A two-warehouse inventory model Yu, J. C. P., Wang, K. J., & Lin, Y. S. (2015). Managing dual warehouses with an incentive
with increasing demand and time varying deterioration. Scientia Iranica, Transactions policy for deteriorating items. International Journal of Systems Science, 47(3),
E: Industrial Engineering, 19, 1969–1977. 586–602.
Laporte, G., Louveaux, F. V., & Hamme, L. V. (1994). Exact solution to a location problem
with stochastic demands. Transportation Science, 28(2), 95–103.
93