Journal of Food Engineering 169 (2016) 61e71

Contents lists available at ScienceDirect

Journal of Food Engineering

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

A vehicle routing problem of both refrigerated- and general-type

vehicles for perishable food products delivery
Byung Duk Song a, Young Dae Ko b, *
a
Department of Industrial and Systems Engineering, Korea Advanced Institute of Science and Technology, Guseong-dong, Yuseong-gu, Daejeon 305-701,
Republic of Korea
b
Data Analytics Group, Deloitte Consulting, Deloitte Touche Tohmatsu Limited, One IFC, 23, Yoido-dong, Youngdeungpo-gu, Seoul 150-945,
Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history: This study examines a vehicle routing problem that encompasses both refrigerated- and general-type of
Received 15 August 2014 vehicles for multi-commodity perishable food products delivery. It is assumed that both the location and
Received in revised form the volume of the ordered food products for each customer are known. Furthermore, the capacity,
30 July 2015
maximum delivery time, and available number of both refrigerated- and general-type of vehicles are
Accepted 24 August 2015
Available online 28 August 2015
predetermined. By reflecting these characteristics, we develop a nonlinear mathematical model and a
heuristic algorithm to generate efficient vehicle routings with the objective of maximizing the total level
of the customer satisfaction which is dependent on the freshness of delivered food products. In addition,
Keywords:
Perishable food product delivery
numerical examples and sensitivity analysis are provided to show the validity of the model. The aim of
Refrigerated-type vehicle this study is to confirm the performance and the availability of refrigerated-type vehicle for perishable
Multi-commodity food products delivery compared with general-type one.
Vehicle routing problem © 2015 Elsevier Ltd. All rights reserved.
Heuristic algorithm

1. Introduction
The development of Information and Communication Technol-
ogies (ICT) has dramatically changed people's daily lives in our
modern society. Among those changes, people's shopping style is
one of the areas most significantly affected by those kinds of trends.
Before the development of ICT, people would visit local stores to
purchase products which they needed in their daily lives. In addi-
tion, they tended to visit large warehouse stores usually located far
from their home to purchase their daily necessities on weekend.
People spent a great deal of time for driving toward shopping
stores, choosing various products and carrying out items to their
home. However, due to the development of ICT, people started to
purchase their daily necessities not only at local shopping stores,
but also on-line shopping stores through internet or phone. As a
result, market size and market share of on-line shopping has
continuously increased. As we can see in Fig. 1, the market volume
of online shopping has increased exponentially in Korea.
In addition, the perishable food products such as vegetables,
fishes and dairy goods can be purchased through on-line shopping
* Corresponding author.
E-mail address: ydko@kaist.ac.kr (Y.D. Ko).
stores. Customers can get their items within several hours during
the working day.
The popularization of on-line shopping has allowed people to
enjoy the convenience of shopping at their home with little effort.
However, at the same time, it causes some troubles for managing
the delivery of food products because of the customer satisfaction
issues coming from the freshness of delivered food products. In
general, food products are characterized as perishable items. Their
freshness is significantly affected by the time duration and tem-
perature environment during the delivery. Thus, when the food
products are delivered to lots of customers, it is hard to keep the
freshness of delivered food products due to extended travel time
and frequent stops to serve customers (Hsu et al., 2007). Therefore,
the efficiency of the vehicle routing scheduling is regarded as an
important issue to maintain the freshness of customer's food
products. To ensure the freshness of the food products, many online
shopping stores that sell perishable food products operate
refrigerated-type vehicles that can control their internal tempera-
ture using cooling equipment. In this case, customers can get more
fresh food products and company can achieve higher level of the
customer satisfaction. However, refrigerated-type vehicles are
more expensive and require more fuel than general-type vehicles;
therefore, it is not easy to operate refrigerated-type vehicles for
every delivery because of the economic issue. As a result, they tend

http://dx.doi.org/10.1016/j.jfoodeng.2015.08.027
0260-8774/© 2015 Elsevier Ltd. All rights reserved.
62 B.D. Song, Y.D. Ko / Journal of Food Engineering 169 (2016) 61e71
Fig. 1. Increasing the market size of online shopping in Korea.
to operate both refrigerated- and general-type vehicles to cover
customer orders.
In this research, we consider a vehicle routing problem with
both refrigerated- and general-type vehicles for multi-commodity
perishable food products delivery to maximize the total sum of
the customer satisfaction, which is dependent on the freshness of
the delivered food products. When vehicles start their travel from
the depot, the freshness of each food product type is assumed to be
perfect. During the delivery, the freshness of each food product is
regarded to reduce based on the elapsed traveling time with the
freshness reduction rate of each food product type. However, when
the food products are delivered via refrigerated-type vehicle,
freshness of delivered food is higher due to its freshness-control
function. But when the storage door of both general- and
refrigerated-type vehicle is opened for picking out some food
products during delivery, all food products which remaining at
storage are regarded to lose certain amount of freshness. In addi-
tion, we assumed that the locations, quantity and food product type
of demand (¼customer orders) and required service for each
customer are known and the capacity of each refrigerated- and
general-type vehicle can be designed differently. With these con-
ditions, we try to derive an efficient delivery schedule for all ve-
hicles through mathematical modeling and efficient solving
algorithm. Based on the result of various numerical experiments,
we want to provide a guideline for delivery schedule of multi-
commodity perishable food product via refrigerated- and general-
type vehicle. Because the purchasing and operating cost of both
refrigerated- and general-type vehicles are significantly different
according to the specification of them, we do not investigate cost
related elements in this study.
2. Literature review
In this section, we introduce previous vehicle routing problems
for perishable food products. If someone wants to investigate an
overall vehicle routing problem, please refer to Eksioglu et al.
(2009) for the classification of the vehicle routing problem.
Various researches have been conducted to find an efficient
vehicle routing policy and supply chain design for perishable goods.
Tarantilis and Kiranoudis (2001) solved a heterogeneous fixed-fleet
vehicle routing problem to find a vehicle operation schedule for
fresh milk. A threshold-acceptance-based algorithm was developed
that aimed to satisfy the needs of a company that planned to
generate a schedule repeatedly, many times over a day. Tarantilis
and Kiranoudis (2002) also addressed an open multi-depot
vehicle routing problem for distributing fresh meat from depots
to their customers located in an area of the city of Athens. A sto-
chastic search meta-heuristic algorithm was proposed to solve the
problem. Prindezis et al. (2003) suggested an application service
provider that would offer the services of distribution logistics for
central food markets that sell and distribute fresh food products. A
vehicle routing problem was proposed and solved via appropriate
meta-heuristic techniques. Campbell and Savelsbergh (2005) sug-
gested a decision support tool for consumer direct grocery initia-
tives. In their article, the authors defined routing and scheduling
problems for grocery delivery service and proposed an insertion
heuristic to derive vehicle schedules. In the problem, the company
decides which deliveries to accept or reject as well as vehicle
schedules for the accepted deliveries so as to maximize expected
profits. With an unpredictable demand, the authors define routing
and scheduling problems and solving algorithm. Osvald and Stirn
(2008) presented an algorithm for the distribution of fresh vege-
tables in which the perishability represents a critical factor. They
dealt the problem with time windows and time-dependent travel-
times (VRPTWTD) where the travel-times between two nodes are
related both the distance and on the time of the day. Schmid et al.
(2009) developed hybrid solution approach for ready-mixed con-
crete delivery. Concrete product is a perishable good, in the sense
that it hardens after a certain amount of time. Due to this charac-
teristic, the authors developed an integer mathematical model to
deliver concrete products from plants to construction sites using a
heterogeneous fleet of vehicles. Optimization and heuristic tech-
niques are integrated to derive vehicle schedules. Recently, Hasani
et al. (2012) designed a closed-loop supply chain for perishable
goods. In this paper, multiple periods, multiple products and mul-
tiple supply chain echelons were considered with uncertain de-
mand. Commercial optimization software LINGO version 8 (LINGO
systems Inc.) was applied to derive a solution to the proposed
mathematical model. Amorim et al. (2013) considered the issue of
lot sizing versus batching in the production and distribution
planning of perishable goods. The authors proved the importance
of lot sizing for make-to-order systems when perishability is
explicitly considered. Govindan et al. (2014) suggested a two-
echelon multiple-vehicle location-routing problem for supply
chain network of perishable food. A multi-objective optimization
model for perishable food supply chain network was developed.
The goal was to determine the number and location of facilities and
to optimize the amount of products delivered to lower stages and
routes at each level. In above researches, heterogeneous types of
vehicles are considered to deliver perishable products. However, no
refrigerated type of vehicle was considered to deliver perishable
products for freshness management.
 B.D. Song, Y.D. Ko / Journal of Food Engineering 169 (2016) 61e71 63

In addition, some researchers addressed the distribution of
frozen and chilled food, which is usually called the cold chain.
Zhang et al. (2003) presented a tabu search algorithm that opti-
mizing the structure of cold chains for the distribution of chilled or
frozen food. Physical distribution systems were structured in such a
way that the cost of storage and transportation for the whole dis-
tribution system was minimized while the product quality re-
quirements were fulfilled. Ying and Ying (2012) proposed an
optimization model that considered the food distribution issue
with a time window for refrigerated food under the condition of
utilizing different modes of transport. Furthermore, the establish-
ment of a distribution center was taken into consideration and the
objective function included transportation costs, distribution cen-
ter establishment cost, penalty cost, and damage cost. In above
researches, authors considered cold chain network design to con-
trol freshness issue of delivered products. However, most of them
are focused on the network design and specific vehicle routing is
not considered.
In this study, we are focusing on the real life issue, which means
multi-commodity vehicle routing of daily delivery for perishable
food products. To address the freshness issue, we considered the
use of both refrigerated- and general-type vehicles. To the best of
our knowledge, this is a state of the art approach for a heteroge-
neous fixed-fleet vehicle routing problem that considers the
simultaneous operation of both refrigerated- and general-type
vehicles to maximize the sum of the customer satisfaction that
comes from the freshness of each delivery of food products.
3. Model development
purchase food products by visiting that grocery store, or by using
one of the online channels: Telephone, fax, or internet/mobile
website. Even though certain customers purchase food products
directly in the store, they can ask for their goods to be delivered.
Therefore, this grocery store delivers many perishable food prod-
ucts according to the customer orders.
It is assumed that the grocery store is located in a metropolitan
city. There are several branch stores and each store is assigned at
certain section of the metropolitan city to serve the customer de-
mand. Therefore, the delivery distance and duration time of each
customer demand is relatively short. In general, this kind of grocery
store tends to cover less than 5 km  5 km area and offer delivery
services at every 2 or 3 business hours during its business hour. In
addition, this kind of delivery services tends to be provided as free
of charge for customer conveniences in Korea.
To cope with the delivery service requests, the grocery store has
a certain number of both refrigerated- and general-type vehicles,
and all food products can be delivered by any type of vehicles. The
delivery process of this grocery store is as follows: In general, a
grocery store collects customer orders over a certain time interval,
and then the store's delivery team makes a vehicle routing plan for
each vehicle to deliver the customer orders. However, due to the
capacity limitations, certain customer requests can be denied or
delayed to the next delivery without penalty. Under this situation, a
grocery store wants to generate an optimal vehicle routing plan for
each vehicle to maximize the sum of the total customer satisfaction
which representing the freshness of the delivered food products.
Through an optimal vehicle routing plan of each refrigerated- and
general-type vehicle, we will evaluate and compare both the per-
formance and the availability of them.

3.1. Problem description

3.2. Notations
In this study, we deal with a vehicle routing problem for a local
grocery store that sells perishable food products. The customer can

i, j: Index for individual customer node

p: Index for food product type
0: Index for indicating depot
V: Set of customer nodes
Vþ: Set of customer nodes and depot, V∪{0}
Kr: Set of refrigerated-type vehicles
Kg: Set of general-type vehicles
K: Set of all vehicles, Kr∪Kg
P: Set of food product types.
cski,: Customer satisfaction value when customer node i is served by vehicle k
CSp,min: Minimum customer satisfaction value for food product type p
brp: Reduction rate of the satisfaction function for food product type p via refrigerated-type vehicle delivery[/elapsed time]
bgp: Reduction rate of the satisfaction function for food product type p via general-type vehicle delivery [/elapsed time]
okp: Unit satisfaction reduction by storage door opening for vehicle k, food product type p
noi: Number of storage door opening while customer i served by certain vehicle
fp(t): Customer satisfaction function for food product type p; it can be differently defined according to the food product nature, but decreased function about elapsed time t
from a depot
tij: Traveling time between node i and node j
si: Required service time of customer node i
qi: Volume of food products ordered by customer at node i
M: Large positive number
ckmax: Maximum capacity of vehicle k
tkmax: Maximum delivery time of vehicle k at each operation
tk0: Preparation time of vehicle k in depot
cki : Decision variable on the remaining capacity of vehicle k after serving the customer at node i
tki : Decision variable on the service start time of customer node i by vehicle k
yki : Binary decision variable indicating that customer at node i is served by vehicle k
xij: Binary decision variable indicating whether the customer at node j is visited directly after customer of node i is served
xi0j: Binary decision variable indicating whether the customer at node i, depot, and customer at node j are served sequentially
TDij: Route generation factor; certain vehicle which is located at node i is dispatched at node j which has lowest value of TDij where j2J and TDij ¼ a x Dij þ qj
Dij: Euclidean distance between node i and node j
a: Conversion parameter between distance and volume for route generation factor
64 B.D. Song, Y.D. Ko / Journal of Food Engineering 169 (2016) 61e71

3.3. Mathematical model  

cki  qj xij þ M 1  xij  ckj ; i; j2V; ck (20)
Based on the addressed problem description, the mathematical
formulations are developed as follows.
cski  CSp;min $yki ; i2V; k2K; p2P (21)
X X
Maximize cski yki (1)
k2K i2V x0i ¼ noi ; i2V (22)

Subject to
xij ðnoi þ 1Þ ¼ noj ; i; j2V (23)
X
xij  1; j2V (2)
i2V þ noi is integer; i2V (24)

X
xi0j  1; j2V (3) cki  0; i2Vk ; ck (25)
i2V þ

X tik  0; i2V þ ; ck (26)

xji  1; j2V (4)
i2V þ
yki 2f0; 1g; i2V þ ; ck (27)
X
xj0i  1; j2V (5)
xij 2f0; 1g; i; j2V þ (28)
i2V þ

X X xi0j 2f0; 1g; i; j2V (29)

xij ¼ xji ; j2V (6)
i2V þ i2V þ

X X
x0i ¼ xi0 (7)
i2V i2V 3.3.1. Objective function
The objective function stands for the total sum of the customer
satisfaction. The customer satisfaction value cski is defined as in
xi0 þ xoj  2$xi0j (8)
Equation (30).
   
X
K
cski; ¼ Max fp tik  tjk  t0j x0j $ykj þ si
yki  1; i2V (9) j2V
k¼1
 okp $noi ; i2V; k2K; p2P (30)

X
K Note that customer satisfaction function, fp(t), is decreasing
yki ykj  xij ; i; j2V (10) function according to the elapsed time from a depot to certain
k¼1 customer node when that customer orders food product type p.
The customer satisfaction value is regarded as decreasing func-
X X
x0i yki ¼ xi0 yki ; ck (11) tion according to the elapsed time from a depot. In detail, if both
i2V i2V the customer node i and j are served at same route, the elapsed
time of customer node i can be expressed Min[tki  tkj þ t0j þ si],
h i which is about the minimum value of [inter-arrival time between
Min t0i $x0i $yki þ t0k  tjk ; j2V; ck (12) customer node i and j þ inter-arrival time between a depot and
i2V
customer node j þ required service time of customer node i].
h i Otherwise, if the customer node i is not served with any
Max tik þ si þ ti0 xi0  tmax
k
; j2V; ck (13) customer node j at same route, the elapsed time of customer
i2V
node i becomes [tki þ si], because it is the first route of both
  refrigerated and general-type vehicle. In addition, delivered food
tik þ si þ tij  M 1  xij  tjk ; i; j2V; ck (14) products can be influenced by outside circumstances when the
storage door is opened.
   
tik þ si þ ti0 þ t0k þ t0j  M 1  xi0j  tjk ; i; j2V; ck (15)
3.3.2. Constraints
Constraints (2) to (8) are used to represent the routing flow.
tik  M$yki ; i2Vk ; ck (16) Constraints (2) to (5) indicate that there is at most one preceding
job and one subsequent job for each customer node. Constraint (6)
is used to balance the input flow and output flow at each customer
cki  M$yki ; i2V; ck (17)
node. Constraint (7) indicates that each route starts and ends at a
depot. Constraint (8) describes the relationship between notation
xi0j  xi0 $x0j ; i; j2V; ck (18) xij and xi0j. Constraint (9) ensures that each customer is served at
most once. Constraints (10) and (11) are used to represent the
relationship between xij and yki . Constraints (12) to (18) are
ckmax  qi $yki  cki ; i2V; ck (19)
developed to determine both the routing sequence and the service
start time. Constraints (12) and (13) provide the minimum and
 B.D. Song, Y.D. Ko / Journal of Food Engineering 169 (2016) 61e71
maximum boundaries for the service start time variable. Con-
straints (14) and (15) are related with the precedence relationship
between the customer service start times. In particular, constraint
(15) represents those relationships when the vehicle visits
customer node i, the depot, then customer j sequentially. Con-
straints (16), (17), and (18) are applied to link the decision vari-
ables. Constraints (19) and (20) are used to ensure that the
capacity of each vehicle is not exceeded. Constraint (21) is intro-
duced as the minimum allowable value of each customer's satis-
faction. Finally, the remaining constraints explain the decision
variables that are suggested in this problem. Constraints (22) and
(23) are about the calculation for number of storage door opening
while customer i served by certain vehicle. When vehicles drop by
for loading food products at depot, the number of storage door
open is initialized as zero. Constraint (24) represents the feature of
storage door open notation as integer variable. Constraints (25)
and (26) are about the non-negative constraints for the remaining
capacity and the service start time at customer node i, respec-
tively, and constraints (27), (28), and (29) are made to describe the
binary decision variables.
4. Solution approaches
To overcome the fundamental complexity of vehicle routing
problems, numerous heuristic algorithms such as exact algorithm,
saving algorithm, and generalized assignment heuristic have been
developed by Agarwal et al. (1989), Altinkemer and Gavish (1991),
and Fisher and Jaikumar (1981) over the last several decades. In
addition, some authors have developed meta-heuristics to derive
vehicle schedules efficiently. Gendreau et al. (1994) developed tabu
searchebased heuristics and demonstrated the performance of
their proposed heuristic. Yousefikhoshbakht et al. (2014) suggested
a combined meta-heuristic called the bone route algorithm that
uses a tabu search as an improved procedure for a heterogeneous
fixed-fleet vehicle routing problem.
In this paper, to solve the nonlinear mathematical formulation,
we suggest an efficient heuristic algorithm called a priority-based
heuristic (PBH), where priority is calculated based on both the
distance and density of the set of nodes.
4.1. Priority-based heuristic
A PBH is developed to address both the complexity and
nonlinearity of the proposed mathematical formulation. The goal of
the PBH is the maximization of the total sum of the customer
Table 1
Pseudo-code of PBH.
Priority based heuristic
STEP 1: Let clk be the current location of vehicle k and ctk be the current time of vehicle k. Initially, clk ¼ 0 and, ctk ¼ 0 for ck2{1,2,…,k}. Select vehicle dispatching rule and
arrange vehicles according to the rule. Set k ¼ 1.
STEP 2: Calculate pi for every i2{V ∩ yki ¼ 0 for every k}. IF k ¼ jKj, k)1.
STEP 3: Selected seed i ¼ max{pi } and define corresponding pir.
STEP 4: Let TDij be the route generation function. Calculate route generation factor TDij ¼ aDij þ qj where 0  a  1
for ci, j2{pir ∪0}
STEP 5: Let t be the index of customer node in pir.
STEP 6: t ¼ min{jjTDclk,j}.
STEP 7: clk)t, ctk)tclk,t þ st and ykt ¼ 1. TDit ¼ INT_MAX for every i2V.
STEP 8: IF any constraint (13), (19), (20) and (21) is not violated, GO TO STEP 6
ELSE IF one of (19), (20) and (21) violated, k)kþ1. GO TO STEP 2.
ELSE IF (a) vehicle(s) violate constraint (13), delete vehicle index k in K and GO TO STEP 2.
P
ELSE IF yik ¼ jvj or ctk > tkmax for ck2{1,2,…,k}, GO TO STEP 9.
i2V;k2K
P P
STEP 9: Calculate the total customer satisfaction, cski yki
STEP Finish PBH k2K i2V
10:
65
satisfaction, which can be achieved by serving as many customers
as possible, while maintaining the freshness of each type of food
product. As such, in the proposed heuristic algorithm, priority is
calculated for every uncovered customer node. Let pi be the priority
value of the ith customer node and R be the Euclidean distance. Let
pir be the set of customer nodes located within R from each node i.
Then the elements of pir are selected via Equation (31).
n o
pir ¼ jjDij  R and ykj ¼ 0 for ck (31)
For every unassigned customer node i, a priority value pi is
calculated with Equation (32). It is assumed that the priority value
of each node i is described as “many customers with a lesser ca-
pacity requirement,” which can be calculated from the number of
nodes in pir, divided by the average customer demand in pir.
8 ,0 19 ,
< X . = X
pi ¼ jpir j @ qj jpir j A ¼ jpir j 2
qj (32)
: ;
j2pir j2pir
For each iteration during PBH, the best priority node is selected
as a seed. And the vehicle schedule is generated via route genera-
tion factor to serve the nodes within radius R from the selected best
priority node. This allows vehicles to visit as many customers as
possible in a single journey while satisfying the capacity limit and
the minimum customer satisfaction level of each type of food
product. In addition, in this problem, both refrigerated- and
general-type vehicles are considered. As a vehicle dispatching
methodology, we introduce three vehicle dispatching rules. 1)
refrigerated-type vehicle first, 2) General-type vehicle first and 3)
Random vehicle dispatching. The overall procedure of PBH is
described as shown in Table 1.
5. Numerical examples
To validate the proposed nonlinear mathematical formulation
and heuristic algorithm, computational experiments are designed
with various operating situations. Small-size demonstrations are
provided to extrapolate important insights from the derived results
while medium- and large-size implementations are presented to
test the performance of the proposed heuristic algorithm.
Furthermore, sensitivity analysis of the customer satisfaction
function is performed to know better the nature of the problem
according to the customer's behavior. For the numerical examples,
we selected vehicle dispatching rule 1, which means refrigerated-
66 B.D. Song, Y.D. Ko / Journal of Food Engineering 169 (2016) 61e71

Table 2
Systemic parameters of an example problem.

Capacity of refrigerated-type vehicle, crmax 12 m3(2 m  3 m  2 m)

Capacity of general-type vehicle, cgmax 12 m3(2 m  3 m  2 m)
Traveling speed of refrigerated-type vehicle 500 m/1 min
Traveling speed of general-type vehicle 500 m/1 min
Maximum value (intercept) of the customer satisfaction function, a 100
Reduction rate of the satisfaction function of product type 1 by refrigerated-type vehicle delivery, br1 0
Reduction rate of the satisfaction function of product type 2 by refrigerated-type vehicle delivery, br2 1
Reduction rate of the satisfaction function of product type 3 by refrigerated-type vehicle delivery 3, br3 1.5
Reduction rate of the satisfaction function of product type 1 by general-type vehicle delivery, bg1 1
Reduction rate of the satisfaction function of product type 2 by general -type vehicle delivery, bg2 2
Reduction rate of the satisfaction function of product type 3 by general -type vehicle delivery 3, bg3 3
Minimum customer satisfaction level for product type 1, CS1,min 70
Minimum customer satisfaction level for product type 2, CS2,min 65
Minimum customer satisfaction level for product type 3, CS3,min 60
R of priority based heuristic 1000 m
Maximum delivery time of the refrigerated-type vehicle, trmax 120 min
Maximum delivery time of the general-type vehicle, tsmax 120 min
Satisfaction reduction in by storage door opening of refrigerated-type vehicle (k ¼ 1) for product type 1, o11 0.5
Satisfaction reduction in satisfaction by storage door opening of refrigerated-type vehicle (k ¼ 1) for product type 2, o12 1
Satisfaction reduction in satisfaction by storage door opening of refrigerated-type vehicle (k ¼ 1) for product type 3, o13 2
Satisfaction reduction in satisfaction by storage door opening of general-type vehicle (k ¼ 2) for product type 1, o21 1
Satisfaction reduction in satisfaction by storage door opening of general-type vehicle (k ¼ 2) for product type 2, o22 1.5
Satisfaction reduction in satisfaction by storage door opening of general-type vehicle (k ¼ 2) for product type 3, o23 3
Conversion parameter at route generate function, a 0.1

type vehicle first. Please note that in PBH heuristic, customer dis-
tribution congestion (density) is an important factor to derive
vehicle schedules and it decreases as the algorithm runs. Therefore
we assigned refrigerated-type vehicle preferentially to increase the
total customer satisfaction. These experiments use a personal
computer with Intel(R) Core(TM) i5-4460 CPU, 3.20 GHz and 8.00
GB RAM.
5.1. Small size demonstration
The proposed nonlinear mathematical formulation and heuris-
tic algorithm were tested with a randomly generated example
problem. Fifty customer nodes are randomly generated on the two-
dimensional plane with a range between (0 m, 0 m) and (5,000 m,
5,000 m). The depot is located at the center of the plane (2500 m,
2500 m) and the preparation time at depot was set to 5 min. Two
vehicles: One refrigerated-type (k ¼ 1) and one general-type (k ¼ 2)
are considered. There are assumed three types of food products and
the linear trend reduction was considered for the customer satis-
faction function of each food product. That means that the
customer satisfaction function can be described as follows: In
Equation (33), a indicates the maximum value (intercept) of the
customer satisfaction function while brp and bgp stand for the
satisfaction reduction rate of food product type p according to the
elapsed time t from a depot by refrigerated- and general-type
vehicle delivery, respectively. In usual, brp  bgp for p2P due to

Table 3
Details of customer node data with 50 customers.

Customer index x y Demand (m3) Product type Service time (min) Customer index x y Demand (m3) Product type Service time (min)

1 1109 1490 0.4 2 3 26 3403 1368 1.5 1 1

2 2765 2179 0.5 2 2 27 2042 699 1.8 3 2
3 975 2998 0.9 3 3 28 1598 2151 0.7 3 2
4 90 842 1.6 3 1 29 933 58 0.3 2 2
5 938 1208 1.7 2 2 30 2792 811 0.7 1 1
6 3908 1005 0.9 1 3 31 3378 1073 0.3 3 3
7 1223 1590 0.4 3 1 32 4980 3935 0.6 1 1
8 4654 3092 1.4 2 1 33 161 1906 0.5 2 1
9 2930 208 1.8 1 3 34 3293 2871 0.4 1 1
10 1675 2458 1.2 3 2 35 2763 3169 1.7 1 2
11 425 2213 0.9 2 2 36 3366 1493 0.8 3 1
12 1947 3108 0.4 1 2 37 2839 4964 0.4 3 3
13 4307 1275 1.3 2 3 38 2870 4650 0.6 1 1
14 3627 4873 1.3 1 2 39 4583 2600 0.9 1 2
15 1666 4325 0.9 2 2 40 1436 4002 1.6 2 2
16 2021 1984 1.1 2 2 41 4782 1486 1.8 3 1
17 1235 466 1.5 1 3 42 23 3866 1.2 1 1
18 3437 2020 0.8 3 1 43 3030 1489 1.7 3 1
19 2480 2877 0.7 2 2 44 4092 4156 0.3 2 1
20 1898 3563 0.8 1 3 45 4020 2598 1.1 3 1
21 1126 199 1.4 2 2 46 942 691 0.5 2 3
22 112 4397 0.5 3 3 47 1647 4798 0.7 1 3
23 362 1860 0.6 1 1 48 4812 1674 1.4 2 3
24 1137 2712 1.0 1 2 49 4332 3428 1.8 3 2
25 1203 1789 1.6 2 3 50 4952 2609 1.6 3 1
 B.D. Song, Y.D. Ko / Journal of Food Engineering 169 (2016) 61e71 67

Table 4
Detail results of both the refrigerated-type and the general-type vehicles.

Vehicle schedule Number of Average satisfaction of customer

served
customer

The refrigerated-type vehicle

Route 1 0 / 16/28 / 10/24 / 11/23 / 1/7 / 25/5 / 0 91.0234
10
Route 2 0 / 19/35 / 34/12 / 20/0
5
Route 3 0 / 45/39 / 8/49 / 32/0
5
Route 4 0 / 40/15 / 47/0
3
The general-type vehicle
Route 1 0 / 2/18 / 36/26 / 31/6 / 13/30 / 0 83.3156
8
Route 2 0 / 38/37 / 14/44 / 0
4
Route 3 0 / 46/17 / 21/29 / 4/0
5
Route 4 0 / 42/22 / 0
2
Average customer satisfaction of refrigerated-type vehicle per Average customer satisfaction of general-type vehicle per product type
product type
Product type 1 Product type 2 Product type 3 Product type 1 Product Product type 3
type 2
98.1111 87.3356 84.9036 90.6003 80.5687 77.5636

the freshness-control function (some of values in brp can be zero customer node. The demand of each customer node was a
based on the product characteristic). randomly generated value between 0.3 m3 and 1.8 m3 and service
time was assigned as 1, 2 or 3 min.
8 With the above hypothetical data set, the proposed heuristic
< 0; t<0
a  brp t; t  0; p2P; k2Kr algorithm was implemented. In this example, we assumed that
fp ðtÞ ¼ (33)
: g freshness reduction of product type 2 and 3 happens during the
a  bp t; t  0; p2P; k2Kg
delivery via refrigerated-type vehicle by setting (br1, br2, br3) ¼ (0, 1,
More detailed information about the system parameters is 1.5). Also freshness of food products was reduced by the storage
shown in Table 2 and Table 3 shows detailed information about door opening for both refrigerated- and general-type vehicle.
customer data, such as their x and y coordinates, and both the Table 4 illustrates detail delivery results of both refrigerated-
demand (qi), food product type (Pi), and the service time (si) of each and general-type vehicles including specific routes and customer

Fig. 2. Geographical descriptions for operation routes of refrigerated-type vehicle.

68 B.D. Song, Y.D. Ko / Journal of Food Engineering 169 (2016) 61e71

Fig. 3. Geographical descriptions for operation routes of general-type vehicle.

Table 5
Computational results of medium-size problems.

jVj jKrj jKgj (tr,max tgmax) Number of served customer Average customer satisfaction CPU time (sec)

Refrigerated type General type Refrigerated type General type

100 1 1 (60,60) 19 14 89.1846 80.3847 0.130

(120,120) 32 30 89.9613 79.4283 0.143
150 2 2 (60,60) 35 29 88.6014 81.1147 0.345
(120,120) 61 57 86.9772 80.3402 0.402
200 3 3 (60,60) 54 47 89.7217 80.8751 0.648
(120,120) 96 82 87.8296 80.3292 0.741

satisfactions. The proposed heuristic algorithm generated four
routes for both refrigerated-type vehicle and general-type vehicle.
The four routes of refrigerated-type vehicle covered 23 customer
demands. Meanwhile, the four routes for general-type vehicle only
covered 19 customer demands. The number of customers served by
refrigerated-type vehicle is greater than that of general-type
vehicle because refrigerated-type vehicle has smaller reduction
rates of customer satisfaction for during the delivery and door
opening. Refrigerated-type vehicle guarantees the higher customer
satisfaction for every type of product. In addition, as the route index
increased, fewer customers were served. This is obvious
commonsense because both the number of uncovered customers
and the customers in range R decreased. The average customer
satisfaction from refrigerated-type and general-type vehicles was
91.0234 and 83.3156 respectively, and the computational time was
0.151 s.

Table 6
Systemic parameters of large-size problem.

Minimum customer satisfaction level for product type 1, CS1,min 75.0

Minimum customer satisfaction level for product type 2, CS2,min 72.5
Minimum customer satisfaction level for product type 3, CS3,min 70.0
Minimum customer satisfaction level for product type 3, CS4,min 67.5
Minimum customer satisfaction level for product type 3, CS5,min 65.0
Reduction in satisfaction by door opening of refrigerated-type vehicle for product type 1, Ok1 k2Kr 0.5
Reduction in satisfaction by door opening of refrigerated-type vehicle for product type 2, Ok2 k2Kr 0.75
Reduction in satisfaction by door opening of refrigerated-type vehicle for product type 3, Ok3 k2Kr 1.0
Reduction in satisfaction by door opening of refrigerated-type vehicle for product type 3, Ok4 k2Kr 1.25
Reduction in satisfaction by door opening of refrigerated-type vehicle for product type 3, Ok5 k2Kr 1.5
Reduction in satisfaction by door opening of general-type vehicle for product type 1, Ok1 k2Kg 1.0
Reduction in satisfaction by door opening of general-type vehicle for product type 2, Ok1 k2Kg 1.5
Reduction in satisfaction by door opening of general-type vehicle for product type 3, Ok1 k2Kg 2.0
Reduction in satisfaction by door opening of general-type vehicle for product type 3, Ok1 k2Kg 2.5
Reduction in satisfaction by door opening of general-type vehicle for product type 3, Ok1 k2Kg 3.0
 B.D. Song, Y.D. Ko / Journal of Food Engineering 169 (2016) 61e71 69

Figs. 2 and 3 describe geographical description for operation

route of each refrigerated- and general-type vehicle respectively. As

Service coverage CPU time

we can see in the figures, vehicles visit customer nodes sequentially

1.619
1.772
2.404
2.454
3.043
3.118
3.914
3.969
(sec.)
in terms of geographical location due to the characteristic inherent
in the PHB heuristic. For example in route 1 of Fig. 2, vehicle visit
customer 7 instead of 25 after serving customer 1. This result comes
from the route generation factor, TDij. In the route generation factor,
i indicate the current location of the vehicle, customer 1 in this case.
TD1,7 and TD1,25 are calculated as follow: TD1,7 ¼ a  D1,7 þ

28.20
43.80
43.40
63.60
55.80
84.60
69.40
98.00
q7 ¼ 0.1  151.64 þ 0.4 ¼ 15.564 and
(%)

76.05(12)
TD1,25 ¼ a  D1,25 þ q25 ¼ 0.1  313.43 þ 1.6 ¼ 32.943. Therefore,
75.98(16)
76.20(16)
76.90(20)
78.49(15)
77.94(25)
75.82(17)
76.93(27)
customer 7 was served instead of 25 after serving customer 1 in this
example. Conversion parameter a was assumed as 0.1 to consider
P5

the relative magnitude of both distance and demand volume.

78.18(12)
78.31(18)
77.95(16)
78.15(21)
79.71(23)
80.41(33)
79.10(28)
78.58(45)
Average customer satisfaction per product type (Number of served customer per product)

5.2. Medium- and large-size implementation

P4

82.54(12)
80.96(23)
81.41(19)
80.79(30)
81.10(21)
80.96(38)
80.40(35)
80.56(45)

To demonstrate the applicability of the proposed mathematical

formulation and heuristic algorithm, tests with medium- and large-
P3

size datasets were also conducted. While the system parameters

83.15(15)
82.53(26)
83.02(30)
82.70(40)
81.78(36)
82.51(51)
82.49(39)
83.39(53)

were set to the same as in Table 2, a variety of quantities of

customer nodes, such as 100, 150 and 200 customers were
P2
General type

considered in a range between (0 m, 0 m) and (5,000 m, 5,000 m)

85.31(17)
85.97(25)
85.19(25)
85.60(42)
86.23(42)
86.88(58)
86.38(50)
86.53(60)

with three types of food products. The number of both refrigerated-

and general-type vehicles were assumed to be 2, 4, and 6, while the
P1

maximum delivery time of both vehicles (tr,max tgmax) was set to

83.96(11)
83.48(20)
83.97(21)
82.88(28)
83.32(30)
82.13(48)
82.90(40)
83.22(56)

(60 min, 60 min) and (120 min, 120 min).

The computational results from the medium-size problems
P5

were summarized at Table 5. For every case, the proposed heuristic

86.98(18)
86.69(26)
85.97(28)
85.67(45)
86.82(31)
86.49(46)
85.45(40)
85.50(62)

algorithm generated efficient solutions within a very short time,

less than 1 s for all cases. As we mentioned above, we can confirm
P4

that refrigerated-type vehicles were used to cover more customer

87.74(15)
88.62(21)
88.41(24)
88.34(42)
89.68(31)
87.89(58)
88.19(40)
88.27(59)

nodes than general-type vehicles. Furthermore, when (trmax,

tgmax) ¼ (120 min, 120 min), more customers could be served
P3

because of the relatively long delivery time.

89.98 (14)
90.86(20)
91.36(16)
92.01(24)
92.21(25)
91.97(36)
91.30(32)
91.44(45)

In the large size implementation, we considered 500 customer

Refrigerated type

nodes in a range between (0 m, 0 m) and (5,000 m, 5,000 m) with

P2

five types of food products. We performed computational experi-

96.33 (15)
96.56(24)
96.55(22)
96.54(26)
95.98(25)
95.97(30)
96.17(26)
96.20(38)

ments with various operating vehicle numbers and delivery times.

Five types of food products are delivered to the customer. System
parameters for deterioration rates of five types of food products and
P1

minimum levels are provided in Table 6. Other parameters such as

a, R, crmax, crmax, a and vehicle velocity are same with those in Table 2.
81.4508
81.3190
81.4491
81.7393
82.3318
82.5625
81.9744
81.9555
General

Among the set of vehicles, first partition (k ¼ 1,…, jKj/2) are

type
Average customer

refrigerated-type vehicles and second partition (element k¼ (jKj/

2 þ 1),…, jKj) are general-type vehicles because dispatching rules 1
Refrigerated
satisfaction

(refrigerated-type vehicle first) was applied. (br1, br2, br3, br4, br5)
89.1770
89.3626
88.9929
88.5098
89.2661
88.1113
88.1093
88.2303

and (bg1, bg2, bg3, bg4, bg5) were set to (0, 0.75, 1, 1.25, 1.5) and (1.0, 1.5,
type

2.0, 2.5, 3.0).

Table 7 shows the number of served customers, average satis-
factions, service coverage and computational time. As we can see in
General

Table 7, 500 customer nodes were considered, served with four, six,
type
Computational results of large-size problems.

68
108
106
153
137
205
169
230
jKrj jKgj (tr,max tgmax) Number of served

eight and ten of both vehicles. Because the proposed mathematical

formulation and heuristic algorithm were not designed to serve all
Refrigerated

customers, we introduce the concept of service coverage, which is

customer

the ratio of served customer nodes per all customer nodes. Obvi-
type

ously, when the number of operating vehicles and the maximum

73
111
111
165
142
218
178
260

delivery time were set to large, then more customers could be

(120,120)
(180,180)
(120,120)
(180,180)
(120,120)
(180,180)
(120,120)
(180,180)

served and the service coverage was higher. Although both types of
vehicles are functionally same in term of travelling speed, capacity
and maximum delivery time, refrigerated-type vehicles serve more
customers due to the freshness control function with higher level of
2

customer satisfactions. In here, we can confirm that the maximum

500 2

5
Table 7

service coverage was 98.00% for the case of five refrigerated-type

jVj

and five general-type vehicles with a maximum delivery time of

70 B.D. Song, Y.D. Ko / Journal of Food Engineering 169 (2016) 61e71
Table 8
Sensitivity analysis of the reduction rate of the customer satisfaction function.
a (bg1,bg2, bg3) Number of served customer
Refrigerated type General type
100 (0.0, 0.0, 0.0) 67 61
(0.25, 0.5, 0.75) 67 61
(0.5, 1.0, 1.5) 66 60
(0.75,1.5, 2.25) 68 57
(1.0, 2.0, 3.0) 67 56
(1.25, 2.5, 3.75) 67 52
(1.5, 3.0, 4.5) 64 50
180 min. Therefore, decision makers who want to adopt these
research results at their businesses can deploy a suitable number of
vehicles to cope with their target service coverage and to create
potential revenue in a real business situation.
In this study, we more focused on efficiency of refrigerated-
type vehicle when it is assumed to apply at food products de-
livery. Therefore, there are several assumptions to simplify the
problem situation. To apply this research results in real situation,
more considerations about how to determine the value of system
parameters are needed. For instance, if there is sufficient data
about deterioration status in terms of types of deterioration and
contributing factors, we can estimate deterioration values such as
brp, bgp and okp by statistical methodologies such as multivariate
regression and so on. It is up to the quality and quantity of related
data at real situation. If there is no data, then actual tests
(including customer surveys) will be very helpful to estimate such
values.
5.3. Sensitivity analysis of the customer satisfaction function
The deterioration rates of perishable food products depend on
numerous factors such as the category of the food product, tem-
perature, humidity, and so on. In this section, we will conduct
sensitivity analysis on the reduction rate of the customer satisfac-
tion function, which is related to the deterioration rate of the
perishable food products. For the test, (jVj, jKrj, jKgj, jPj, trmax, tgmax)
were set to (150, 2, 2, 3, 120, 120), respectively. Also systemic pa-
rameters are set to those of Table 2 except br1, br2, br3, bg1, bg2, and
bg3. To investigate the effect reduction rate of customer satisfaction
by general-type vehicle delivery, we set (br1, br2, br3) ¼ (0, 0, 0) and
changes the values in bg1, bg2, and bg3.
Table 8 shows the results with a variety of values on the
reduction rates per food products of the customer satisfaction
function via general-type vehicle delivery. In common with our
expectation, the service coverage and the average value of
customer satisfaction increase as the value of the reduction rates
of the customer satisfaction function decreases. In every case, the
average customer satisfactions of refrigerated-type vehicle are
similar by setting (br1, br2, br3) ¼ (0, 0, 0). However due to the
freshness reduction by storage door opening, customer satisfac-
tion cannot perfectly be achieved. In the case of general-type
vehicle, the number of served customer and the average
customer satisfaction were hugely affected by the changes of (bg1,
bg2, bg3). When (bg1, bg2, bg3) ¼ (0, 0, 0) general-type vehicle
serves the maximum number of customers. However, both the
number of customer served by general-type vehicle and average
customer satisfaction decrease as the values of (bg1, bg2, bg3) in-
crease. Also, there is a difference in the average customer satis-
faction which comes from the reduction value by storage door
opening. The satisfaction reductions by storage door opening
were set to (0.5, 1, 2) and (1, 1.5, 3) for refrigerated-type and
general-type vehicles.
Average customer satisfaction Service coverage (%) CPU time (sec)
Refrigerated type General type
93.5597 89.2131 85.33 0.359
93.5597 85.6241 85.33 0.391
93.4394 82.9062 84.00 0.356
93.1250 81.6992 83.33 0.358
93.4179 80.5743 82.00 0.361
93.4030 79.3652 79.33 0.349
93.1875 78.4110 76.00 0.356
6. Conclusions
This study aims to confirm the performance and the availability
of refrigerated-type vehicle for multi-commodity perishable food
products delivery comparing general-type one. To achieve this, a
nonlinear mathematical model and an efficient heuristic algorithm
were developed to derive the vehicle routings for both refriger-
ated- and general-type vehicles. The numerical examples were
conducted with a hypothetical data set to evaluate proposed
nonlinear mathematical model and heuristic algorithm. The
average number of served customers per each route by
refrigerated-type vehicle tended to be greater than that of general-
type vehicle due to the constraint of the minimum customer
satisfaction. In addition, the hypothetical data with various sizes of
orders placed by customer nodes were tested to evaluate the
performance of proposed heuristic algorithm. Even in case of 500
customer nodes, five food product types and ten vehicles, the
proposed heuristic algorithm derived an efficient solution within a
few seconds. Moreover, the effectiveness of both refrigerated- and
general-type vehicle were tested under the various deteriorating
situation.
Based on the research results of this study, we expect that a
practitioner can get insights when they design and operate a
perishable food product delivery service. They can decide on a
suitable number of both refrigerated- and general-type vehicles
with the factor of service coverage for various situations based on
the performance and the availability of them. In addition, the
computational results which about the effects of the capacity of
each vehicle and the maximum delivery time on the customer
satisfaction will be helpful for developing the delivery system
efficiently.
As a further study, we will consider the effects of various kinds
of deterioration factors at perishable food products delivery
because the kinetics of freshness losses is depending on several
factors such as temperature and moisture and so on. Then, the
deterioration function will be regarded more generalized numerical
expression to reflect various deterioration factors. In addition, we
will address the various kinds of refrigerated-type vehicle in the
aspect of both capacity and capability.
References
Agarwal, Y., Mathur, K., Salkin, H.M., 1989. A set-partitioning-based exact algorithm
for the vehicle routing problem. Networks 19 (7), 731e749.
Altinkemer, K., Gavish, B., 1991. Parallel savings based heuristic for the delivery
problem. Operations Res. 39 (3), 456e469.
Amorim, P., Belo-Filho, M.A.F., Toledo, F.M.B., Almeder, C., Almada-Lobo, B., 2013. Lot
sizing versus batching in the production and distribution planning of perishable
goods. Int. J. Prod. Econ. 146 (1), 208e218.
Campbell, A.M., Savelsbergh, M., 2005. Decision support for consumer direct gro-
cery initiatives. Transp. Sci. 39 (3), 313e327.
Eksioglu, B., Vural, A.V., Reisman, A., 2009. The vehicle routing proglem: a taxo-
nomic review. Comput. Ind. Eng. 57 (4), 1472e1483.
Fisher, M.L., Jaikumar, R., 1981. A generalized assignment heuristic for vehicle
routing. Networks 11 (2), 109e124.
 B.D. Song, Y.D. Ko / Journal of Food Engineering 169 (2016) 61e71
Gendreau, M., Hertz, A., Laporte, G., 1994. A tabu search heuristic for the vehicle
routing problem. Manag. Sci. 40 (10), 1276e1290.
Govindan, K., Jafarian, A., Khodaverdi, R., Devika, K., 2014. Two-echelon multiple-
vehicle location-routing problem with time windows for optimization of
sustainable supply chain network of perishable food. Int. J. Prod. Econ. 152,
9e28.
Hasani, A., Zegordi, S.H., Nikbakhsh, E., 2012. Robust closed-loop supply chain
network design for perishable goods in agile manufacturing under uncertainty.
Int. J. Prod. Res. 50 (16), 4649e4669.
Hsu, C.I., Hung, S.F., Li, H.C., 2007. Vehicle routing problem with time-windows for
perishable food delivery. J. Food Eng. 80 (2), 465e475.
Osvald, A., Stirn, L.Z., 2008. A vehicle routing algorithm for the distribution of fresh
vegetables and similar perishable food. J. Food Eng. 85 (2), 285e295.
Prindezis, N., Kiranoudis, C.T., Marinos-Kouris, D., 2003. A business-to-business fleet
management service provider for central food market enterprises. J. Food Eng.
60 (2), 203e210.
71
Schmid, V., Doerner, K.F., Hartl, R.F., Savelsbergh, M.W.P., Stoecher, W., 2009.
A hybrid solution approach for ready-mixed concrete delivery. Transp. Sci. 43
(1), 70e85.
Tarantilis, C.D., Kiranoudis, C.T., 2001. A meta-heuristic algorithm for the efficient
distribution of perishable foods. J. Food Eng. 50 (1), 1e9.
Tarantilis, C.D., Kiranoudis, C.T., 2002. Distribution of fresh meat. J. Food Eng. 51 (1),
85e91.
Ying, W., Ying, Y.L., 2012. Optimization model of refrigerated food transportation.
In: Proceedings of International Conference of Service Systems and Service
Management, Beijing, pp. 220e224.
Yousefikhoshbakht, M., Didehvar, F., Rahmati, F., 2014. Solving the heterogeneous
fixed fleet open vehicle routing problem by a combined metaheuristic algo-
rithm. Int. J. Prod. Res. 52 (9), 2565e2575.
Zhang, G., Habenicht, W., Ludwig Spieb, W.E., 2003. Improving the structure of deep
frozen and chilled food chain with tabu search procedure. J. Food Eng. 60 (1),
67e79.