Flexible Services and Manufacturing Journal

https://doi.org/10.1007/s10696-019-09333-y

Scheduling appointments for container truck arrivals

considering their effects on congestion

Sanghyuk Yi1 · Bernd Scholz‑Reiter1 · Taehoon Kim2 · Kap Hwan Kim3

© Springer Science+Business Media, LLC, part of Springer Nature 2019

Abstract
Trucking companies deliver a large number of containers every day to container ter‑
minals at hub ports. Truck drivers for the delivery operation can experience long
waiting times when they arrive at peak hours. This study proposes a scheduling
method for appointments that considers the cost of trucks staying in the terminal,
demurrage cost, container delivery cost, number of appointments allowed at each
time window and block, and number of trucks available during each time window.
Unlike previous studies, this study considers the effects of the appointments on the
waiting time at the terminal when the appointment schedule is constructed. This
paper introduces a mathematical formulation and a heuristic algorithm based on
the Frank–Wolfe algorithm to solve the problem within a reasonable computational
time. Numerical experiments are conducted to compare the proposed algorithm with
the other heuristic approaches and analyze the effects of the appointments using
empirical data. In addition, the impact of appointments by multiple trucking compa‑
nies is examined.

Keywords Container terminal · Appointment system · Trucking company ·

Scheduling

* Kap Hwan Kim

kapkim@pusan.ac.kr
1
International Graduate School (IGS) for Dynamics in Logistics, Production Engineering,
University of Bremen, Hochschulring 20, 28359 Bremen, Germany
2
Department of Computer Engineering, Pusan National University, Busandaehak‑ro 63
beon‑gil. 2, Busan 46241, Republic of Korea
3
Department of Industrial Engineering, Pusan National University, Busandaehak‑ro 63 beon‑gil
2, Busan 46241, Republic of Korea

13
Vol.:(0123456789)
 S. Yi et al.

1 Introduction

Operations at container terminals have many challenges that can be described as

open systems of material flow with two external interfaces: quayside and landside
(Steenken et al. 2004). In quayside operations, quay cranes load or unload containers
from or to a vessel. In landside operations, the yard cranes are loading or unload‑
ing containers on or off inland transporters, such as trucks. In practice, the prior‑
ity of quayside operations is higher than that of landside operations (Phan and Kim
2015b). In other words, a yard crane prioritizes the handling of internal trucks over
serving the road trucks.
When trucking companies develop schedules of their trucks, they attempt to min‑
imize the various costs considering the delivery requirements of containers, turna‑
round times within a terminal, and availability of trucks at the corresponding time
window. To reduce the unproductive waiting of trucks within a terminal, trucking
companies consider the information about the arrival of trucks at the terminal, par‑
ticularly at some peak times. One of the practical ways for trucking companies to
communicate with terminal operators is the truck appointment system, which has
already been implemented at some container terminals, such as the ports of Los
Angeles and Long Beach (Giuliano and O’Brien 2007), Hong Kong, Jebel Ali, Ant‑
werp Gateway, and Southampton.
In conventional truck appointment systems, truck companies appoint their trucks
to the time windows, while the terminal operator sets in advance the maximum num‑
ber of trucks that can arrive at the gate during each time window (Phan and Kim
2015a). When the maximum value is reached, any additional appointments for truck
arrivals are rejected. On the other hand, from the perspective of trucking companies,
while container terminals can control the congestion easily by setting the maximum
number of truck arrivals, trucking companies have difficulties in satisfying the deliv‑
ery requirements of individual containers, and the various conditions of each truck‑
ing company. Phan and Kim (2015a) developed a new truck appointment system
with the negotiation process considering the demand for both container terminals
and trucking companies, as illustrated in Fig. 1. The new truck appointment system
estimates the expected waiting time of time windows by the information of work‑
load from the terminal operation system. Then, trucking companies submit applica‑
tions for appointments by using the informed expected waiting time.
One of difficulties in preparing the applications for appointments is that the
number of appointments is very large. According to practical data collected from

Fig. 1  A new truck appointment system (Phan and Kim 2015a)

13
Scheduling appointments for container truck arrivals…

a container terminal in Busan (refer to Table 4), the number ranges between 45 and
475 for inbound containers, which are to be sent to 7 blocks in a day. Thus, it is
almost impossible for a human planner to decide the best time window to send a
truck to pick up or deliver each container, which is a motivation of this study. This
study proposes an efficient algorithm for a trucking company to schedule a large
number of appointments considering various practical constraints and cost terms in
a short computational time. The algorithm proposed in this study may be used by
trucking companies for the appointment scheduling. Otherwise, the appointment
schedule, which is embedded in the Internet-based appointment system owned by
container terminals, may be produced upon the requests from trucking companies.
This paper is organized as follows: Sect. 2 provides literature reviews; Sect. 3
describes the appointment system assumed in this study; Sects. 4 and 5 propose a
mathematical model for scheduling appointments and a heuristic algorithm to obtain
the solution; Sect. 6 introduces numerical examples and the results of the numerical
experiments; and Sect. 7 provides the conclusions.

2 Literature review

This paper discusses the truck scheduling problem by appointing trucks within
the new truck appointment system at container terminals. It is attempted to review
researches on the truck scheduling problems for the container delivery, impacts of
truck arrival distribution on terminal operations, and conventional truck appoint‑
ment systems. Then, the research gaps between existing works and the new truck
appointment system are discussed.

2.1 Truck scheduling problem for the container transportation

The problems for scheduling road trucks have been addressed. Nossack and Pesch
(2013) discussed the scheduling of road trucks at a container terminal. The delivery
tasks of trucks were classified into full outbound, empty outbound, full inbound, and
empty inbound containers. The purpose of the scheduling trucks was to minimize
the total operation time, where the strict time constraints requested by the custom‑
ers and terminals should be satisfied. Wang and Yun (2013) addressed the inland
container transportation according to the trucks and trains among the customers,
container terminals, and depots. They formulated a mathematical model through
a graphical representation and developed a hybrid Tabu search algorithm. Zhang
et al. (2009, 2010) dealt with the truck scheduling problem for the container trans‑
portation considering multiple depots and terminals with time windows. The prob‑
lem was formulated as a multi-traveling salesperson problem with time windows
(m-TSPTW). Heilig et al. (2017a) addressed the truck routing problem for the inter-
terminal transportation which accounts for a substantial portion of container trans‑
ports and has gained increased interest in terms of both economic and environmen‑
tal perspectives. They proposed a multi-objective simulated annealing approach in
order to reduce the economic costs and environmental impact. Schulte et al. (2017)

13
 S. Yi et al.

discussed the truck scheduling problem with collaborative appointment for reducing
empty truck emissions. They proposed a collaborative planning model for trucking
companies within a truck appointment system.

2.2 Importance of managing truck arrivals on terminal operations

The management of road truck arrivals strongly influences overall terminal opera‑
tions. Chen et al. (2013) proposed a method for estimating hourly truck arrival rate
from a given time window allowed to containers of each vessel and for estimating the
expected waiting time of trucks at the gate. They showed, by a case study, that truck
arrivals may be distributed over time windows and the gate congestion may be allevi‑
ated considerably by optimizing the time window assignment. Zhao and Goodchild
(2010) explored the effect of the information about truck arrivals on container handling
operations. The results demonstrated that the information can significantly reduce the
unproductive rehandling movements of yard cranes. Lalla-Ruiz et al. (2015) discussed
the management of container flows from container vessels to consignees through a
terminal. They showed that as more road trucks are allowed inside of the terminal, the
company waiting time is significantly reduced by directly delivering containers to con‑
signees without the temporary storage of containers at the terminal yard.

2.3 Truck appointment systems for a container terminal

Many studies have examined the application and optimization of the truck appoint‑
ment system at a container terminal. Murty et al. (2005) addressed the decision-sup‑
port system to provide superior services at the Hong Kong International Terminal,
where the allocation of appointment time slots to road trucks is involved to alleviate
the congestion at the terminal. The appointment is requested only for road trucks
visiting a terminal for inbound containers during the peak time window. Zhang et al.
(2013) attempted to determine the appointment quota for each period by develop‑
ing an optimization model and queuing network. A genetic algorithm and Pointwise
Stationary Fluid Flow Approximation (PSFFA) was used to find the optimal solu‑
tion. The results revealed a decrease in the turnaround time of trucks at the terminal.
Zehendner and Feillet (2014) discussed the appointment system at a container termi‑
nal with straddle carriers (SCs) considering the various inland transport modes, such
as trucks, trains, and barges. A mixed integer linear programming model was pro‑
posed to determine the number of truck appointments and number of SCs allocated
to the different transport modes to reduce the overall delays in a container terminal.

2.4 Collaborative truck appointment systems

For the inland transportation of containers from/to a terminal, it is essential to col‑

laborate with various stakeholders such as truck companies, inland depots, ship‑
pers, shipping lines etc. There have been some researches on the collaborative sys‑
tem for the transport of containers. In particular, Heilig et al. (2017b) proposed a
mobile cloud platform for the real-time communication with various stakeholders

13
Scheduling appointments for container truck arrivals…

of the inter-terminal transportation. The platform embedded the decision support

component for the truck routing with reducing its empty trips. Likewise, this paper
addresses the collaborative system for the truck appointments towards a container
terminal considering not only the restriction of appointment slots, but also the con‑
gestion of terminal operations.
The scheduling problem of this paper was motivated by the collaborative truck
appointment systems proposed by Phan and Kim (2015a) and Phan and Kim (2016).
In the proposed truck appointment system, trucking companies reserve truck
appointments, and the expected waiting times of trucks for each time window and
each block at a terminal are then estimated. The trucking companies modify their
appointments according to the information of the expected waiting time. Trucking
companies pay the appointment charge for each arriving truck, which is proportional
to the time for a truck remaining in the terminal.
Riaventin and Kim (2018) first addressed the problem of scheduling appoint‑
ments. They proposed a two-phase algorithm, in which the transportation problem
is solved first and the resulting solution is modified to satisfy one additional con‑
straint. They proposed two algorithms: one based on a transportation programming
procedure and the other based on a network flow algorithm. On the other hand, they
assumed that the expected system time is fixed and not dependent on the appoint‑
ment schedule. However, this study considers the fact that the expected system time
at each time window again depends on the appointment schedule.
Unlikely from previous studies which have dealt with modelling the waiting
times considering appointments or collaboration among trucking companies by
using appointment systems, this study addresses a scheduling method for appoint‑
ments considering various practical constraints and cost terms from the viewpoint of
a trucking company. Even though there are other previous studies on the scheduling
of appointments (Phan and Kim 2015a, 2016; Riaventin and Kim 2018), this study
has contributions in the following aspects: firstly, this study attempts to provide an
efficient algorithm to schedule a large number of appointments in a short time for
trucking companies; secondly, this study considers various practical constraints and
cost terms such as the available number of trucks, appointment quotas set by the
terminal, detail situation of each yard block at each time window, and the flexibility
of space allocation of outbound containers; thirdly, considering that the number of
appointments is large, the effect of the number of appointments on the waiting time
at the yard is directly considered during the scheduling, which has not been consid‑
ered in the previous study (Riaventin and Kim 2018).

3 A truck appointment scheduling problem and turnaround time

of trucks

This section introduces a method to estimate the expected time for a truck to stay at a
block, which is called the system time, using the information on the number of truck
arrivals. This section proposes a new appointment system to distribute truck arrivals as
evenly as possible over the entire working hours and defines the appointment-schedul‑
ing problem, which is the main issue in this paper.

13
 S. Yi et al.

Fig. 2  Detail explanation of the new truck appointment system

3.1 A new truck appointment system

The truck appointment system is a popular way to provide a communication chan‑

nel between the container terminal and trucking companies. The appointment sys‑
tem in this study is a communication tool to help both parties collaboratively decide
the truck arrival times at the terminal. When a trucking company wants to deliver
outbound containers or pick up inbound containers, it submits an application to the
appointment system. The appointment system can be operated by each terminal or
by an independent agency. The appointment time slots for the containers will then
be recommended by the appointment system considering the estimated system time
of the trucks arriving within each time window and at each yard block, as shown in
Fig. 2. Note that the expected system time (Ŝ ) for each time window t and each block
b was estimated based on the expected number of internal truck arrivals (wbt ) for ship
operations and the estimated number of road truck arrivals ( Zbt ) collected from the
appointment system. Trucking company then makes a new appointment application
for each container again, taking the information on the expected system time pro‑
vided by the appointment system into account. The length of a time window may be
between 0.5 and 2 h depending on the situation of each terminal. A shorter time win‑
dow will be used to control the congestion in more detail. Terminal managers, who
want to reduce the system time of trucks at a yard block during a specific peak hour,
may apply an appointment charge if necessary, which is proportional to the expected
system time of each truck. This process is repeated until all trucking companies con‑
firm their appointments. Using this appointment system (Riaventin and Kim 2018), it
is anticipated that it will be possible to distribute the truck arrival density across the
entire terminal operation time and avoid truck arrivals during peak hours.
The main advantage of the new appointment system is that the terminal has
information on the arrival times of trucks in advance. Thus, trucking companies
can predict more accurately the service completion times of their trucks at the
terminal. As a result, the terminal can reduce the congestion of trucks for hin‑
terland operations, minimizes the number of rehandling steps, and increases the
utilization of yard equipment. Trucking companies can utilize the information on
the respective waiting times of their trucks at the terminal, which in turn enables
more efficient truck operation. Based on the new model of the truck appointment
system, this study proposes a method for scheduling appointments that may be

13
Scheduling appointments for container truck arrivals…

Number of containers to pick-up/deliver

Container terminal
7 Demurrage cost
Depot1
9 B1 Inbound containers are
discharged 4 days ago
8 12 24 Time in a day
Demurrage cost
7 B2
Outbound containers are
Depot2
7 to be loaded after 3 days
Yard blocks Vessels
8 12 24 Time in a day
Average delivery time
Maximum number of appointment slots Number of ship operations
Time window 6 12 18 24 Time window 6 12 18 24
B1 5 5 5 5 B1 10 10 0 0 Inbound task
B2 5 5 5 5 B2 11 20 5 0 Outbound task
6 12 18 24 Time in a day

Fig. 3  An example of the appointment scheduling problem

Table 1  Examples of the Task Type Location Number of Free charge

container inbound (I)/outbound (Block ID) containers period
(O) schedule for a single
trucking company U1 O B1, B2 9 8:00 ~
U2 O B1, B2 7 8:00 ~
U3 I B1 7 ~ 8:00
U4 I B2 7 ~ 8:00

used for the appointment system or by individual trucking companies for their
own scheduling purposes.

3.2 Defining the appointment scheduling problem

Suppose that a trucking company has containers to pick up or deliver on a particular

day, as shown in Fig. 3 and Table 1. A task is defined as a group of pick-up orders
of inbound containers with identical attributes, i.e., containers to pick up from the
same block and with the same deadline for pick-up. For outbound containers, a task
is defined as a group of container deliveries for the same vessel. These tasks should
be appointed at a certain time window minimizing the total operation cost which
consists of the container delivery cost, the demurrage charge, and the cost of a truck
during its stay at the yard. Firstly, the delivery time has tendency to have peaks in a
day, which is shown at the graph in the lower left side of Fig. 3. Thus, the appoint‑
ments at the peak time periods need to be avoided. Secondly, one of the deadlines
for the pick-up/delivery comes from the maximum free-of-charge dwelling time of
containers. If a container stays longer than a pre-specified free-of-charge period,
a demurrage charge is imposed for each time unit exceeding the free-of-charge
period, which is illustrated at the graph in the upper-right side of Fig. 3. Accord‑
ing to the stevedoring tariff document of “S” terminal (Pusan East Container Ter‑
minal Co Ltd 2004), the demurrage fee is charged when an inbound container stays
longer than the free-of-charge period, which is 4 days. Alternatively, for outbound

13
 S. Yi et al.

containers, 3 days are provided to shippers as a free-of-charge period. Hence, truck‑

ing companies attempt to complete their tasks during the free-of charge period. The
third term is the cost of a truck during its stay at the yard which is proportional to
the congestion. The congestion is expressed by the expected system time.
To determine the number of containers to pick up or deliver within a fixed-length
time window for each task, this study considered three constraints for scheduling.
The first constraint was that the maximum number of slots available for appoint‑
ments within each time window of each storage block may be restricted (Maximum
Slot Constraint). The terminal operator may restrict the number of truck arrivals,
which is a popular practice in traditional appointment systems. Therefore, the new
appointment system in this study was designed to satisfy this traditional require‑
ment of terminal operators. The second constraint was that the total allowable opera‑
tion time of trucks of the trucking company is restricted within each time window
because of the limited number of available trucks of the trucking company (Max-
imum Truck Time Constraint). In addition, for a task of outbound containers, the
maximum number of containers that are allowed to be stored into each block is lim‑
ited by the storage space plan (Space Plan Constraint), which is prepared before
the outbound containers begin to arrive. On the other hand, the storage space for
outbound containers for a task may be allocated more than one block. Note that from
the definition of an inbound task, all the inbound containers for a task are stored at
the same block. Table 1 lists the tasks for inbound and outbound containers.
The objective function is the total cost of the trucking companies, which includes
the truck cost within the terminal, the demurrage charge, and the container delivery
cost outside the terminal. The truck cost within the terminal, including the truck opera‑
tion cost and the appointment charge, was assumed to be proportional to the expected
system time of a truck at the corresponding block within the appointment time slot. The
appointment time is the time when the appointment is made, while the appointment
time slot means the time when the truck is supposed to arrive at the terminal specified
by the appointment. Note that the system time depends strongly not only on the number
of ship operations, which may be estimated from the schedule for ship operations, but
also on the hinterland operations, which can be estimated from the appointments, at the
corresponding block within the corresponding time window. In addition, it is assumed
that the travel time outside the terminal is dependent on the appointment time slot for
the truck arrival.

3.3 Estimating the expected system time of trucks

A queuing mode can be used to estimate the expected system time, which consists of
the expected waiting time and the expected service time for a truck by a yard crane.
For example, when the M/M/1 queuing model is used, the expected waiting time can
be expressed as 𝜇(𝜇−𝜆)
𝜆
� , where 𝜇 and 𝜆 represent the expected service rate of the

server and the expected arrival rate of the customers per unit time, respectively.

13
Scheduling appointments for container truck arrivals…

In this study, to derive a formula for estimating the expected system time using
the information on truck arrivals, the cycle times of various types of operations were
collected from P Terminal in Busan, Korea. From the data collected, the following
statistics were estimated: (mean, standard deviation) of discharging, loading, carry-
in and carry-out operations are (2.71, 0.86), (2.42, 0.81), (3.43, 0.87), and (4.72 min,
2.22 min), respectively. Note that ship operations have a higher priority than hinter‑
land operations. A simulation program was developed to estimate the expected sys‑
tem time of each type of operation. Note that the system time includes the operation
time and the waiting time of trucks. Through regression analysis, Eqs. (1)–(2) were
derived with R = 0.883 and coefficients with very small p values (Kim 2017).
( )
Ŝ ibt Zbt = 6.725 + 0.242wbt + 0.459Zbt for i ∈ UI (1)
( )
Sibt Zbt = 6.331 + 0.242wbt + 0.459Zbt for i ∈ U O (2)
where the expected system time is expressed in minutes (min). The expected system
time is estimated by the number of ship operations (wbt ) and the total number of
road truck operations ( Zbt ). The value of wbt is provided by the terminal operation
planning system and Zbt is the total number of appointments for all the tasks at block
b and time window t , which is the decision variable of this problem. The equations
are separated by the types of hinterland operations: (1) for the inbound case and
(2) for the outbound case. This is because the required handling time is different
from each other. More rehandling operations can occur in the inbound case than the
outbound case. In the case of wbt = 1 and Zbt = 3 which means that 1 internal truck
and 3 road trucks are expected to visit block b within time window t , the expected
system time at block b and time window t is 8.344 min (= 6.725 + 0.242 × 1 + 0.4
59 × 3) for inbound road trucks and 7.950 min (= 6.331 + 0.242 × 1 + 0.459 × 3) for
outbound road trucks. Note that the subscript “ t ” in (1) and (2) represents the time
window index when a truck arrives. Even when a truck arrives at the yard at time
window t but departs the yard at time window ( t + 1), the truck is counted to arrive
at time window t . For the deriving (1) and (2), the length of a time window was
assumed to be 1 h.
In the numerical experiments in this study, the expected system of each time
window for a given number of internal and road truck arrivals was estimated using
regression Eqs. (1) and (2).

4 Formulating the appointment scheduling problem

This section introduces a mathematical model to describe the appointment-sched‑

uling problem. The following notations are used in the formulation:

13
 S. Yi et al.

4.1 Parameters

c Cost per unit time for a truck to stay at the yard, which includes the labor cost for
a driver, fuel cost, and truck overhead cost. When an appointment charge, which
is proportional to the system time, is imposed on trucks to reduce the congestion
in the terminal, this cost term includes the appointment charge of a container per
unit system time, which is set by the terminal operator
dit Demurrage charge of a container for task i when the container is picked up/
delivered within the time window t
b Container delivery cost per unit time of a driving truck, which includes the
overhead cost of a truck, labor cost for the truck driver, and fuel cost for the
truck
sO Average operation (service) time per truck for transferring an outbound con‑
tainer by a yard crane
sI Average operation (service) time per truck for transferring an inbound con‑
tainer by a yard crane
mbt Maximum number of slots available for appointments at block b within time
window t . This parameter will be set by the terminal operator considering the
workload of the corresponding block during the corresponding time window
ait Average time required to deliver a container between a shipper (or consignee)
and a terminal for task i when a truck moves the container within time window
t
ni Number of containers for task i
ut Maximum available number of trucks that may be used to deliver containers
during time window t . The number of trucks available may be limited for a
trucking company during a time window
vibt Number of slots in block b allocated to outbound task i at time window t in the
yard space plan
wbt Number of ship operations planned in block b at time window t . Usually,
the quay crane work schedule for the corresponding vessel includes this
information

Sets

T Set of time windows under consideration (t ∈ T )

B Set of blocks in the terminal (b ∈ B)
U Set of tasks (i ∈ U )
Ub Set of tasks for which containers may be picked up from or delivered to block b
U I Set of inbound tasks
U O Set of outbound tasks
Bi Set of blocks in which space may be allocated to outbound task i or the con‑
tainers for inbound task i are located. For an inbound task, there is only a sin‑
gle element in Bi

13
Scheduling appointments for container truck arrivals…

4.2 Decision variables

Xibt Number of pick-ups (in the case of inbound containers) or deliveries (in
the case of outbound containers) for task i appointed at block b and time
window t
∑
Zbt  Xibt , which is the total number of pick-ups and deliveries for hinterland
i
( ) operations appointed at block b within time window t
Wbt( Zbt) Expected waiting time for a truck arriving at block b in time window t
Sibt Zbt Expected system time when a truck for task i arrives at block b in time
window t . The system time means the time during which a road truck
stays at the block, which includes the(waiting ) time and the container
transfer time by a( yard
) crane, i.e., W bt Z bt + sO for a truck delivering
an outbound; Wbt Zbt + sI for a truck picking up an inbound container.
Note that the container transfer time is different between inbound con‑
tainers and outbound containers because more rehandling operations are
expected for picking up an inbound container than an outbound container

The appointment-scheduling problem can be expressed as follows. The objective

function minimizes the total cost (TC ) which ( is) the following cost terms per unit
time: the truck cost within the terminal (cSibt Zbt Xibt ), demurrage cost (dit Xibt ), and
container delivery cost by a truck (bait Xibt ).
(AS) ∑ ∑ ∑[ ( ) ]
Minimize TC(X) = cSibt Zbt + dit + bait Xibt (3)
i∈U b∈Bi t∈T

subject to
∑∑
Xibt ≤ ut for all t∈T
(4)
i∈U b∈Bi

∑
Xibt ≤ mbt for all t ∈ T, b ∈ B
(5)
i∈Ub

∑∑
Xibt = ni for all i ∈ U
(6)
b∈Bi t∈T

Xibt ≤ vibt for all i ∈ U O , b ∈ Bi , t∈T (7)

Xibt ∶ a non - negative integer (8)

Constraint (4) indicates that within each time window, there is a limitation in the
total operation time of trucks, which is the number of trucks available multiplied by
the length of a time window (Maximum Truck Time Constraint). A trucking com‑
pany can be assumed to have a limited number of trucks available at each time
window. Constraint (5) prevents the number of appointments from exceeding the

13
 S. Yi et al.

Table 2  Example of the input variables

Parameters ut mbt wbt ait (h) dit ($) vibt c ($/min) b ($/h)

Task – – – – – U1 U2 U1 U2 U1 U2 2 10
Block – B1 B2 B1 B2 – – – – B1 B2 B1 B2
Time T1 10 8 8 2 1 0.5 0.5 5 5 4 5 3 4
window T2 10 8 8 3 3 1.0 1.0 0 0 4 5 3 4

Tt , Bb, and Ui represents time window t , block b, and task i , respectively

number of slots available within each time window at each block (Maximum Slot
Constraint). This constraint follows the practice of the traditional appointment sys‑
tem, in which the terminal operator restricts the maximum number of truck arrivals
considering the ship operation schedule. Constraint (6) represents the number of
pick-ups or deliveries necessary to satisfy the requirements defined for each task.
Constraint (7) restricts the maximum number of slots allocated to an outbound task
at each block in each time window by the space plan (Space Plan Constraint). Note
that most of the terminals prepare a storage space plan for outbound containers so
that their final layout becomes as efficient as possible for the loading operation.
The total cost function can be rewritten as
∑∑ ∑ [ ( ) ]
TC(X) = cSibt Zbt + dit + bait Xibt
b∈B t∈T i∈Ub
∑∑ ∑ [ ( ) ]
= cWbt Zbt + csi + dit + bait Xibt
b∈B t∈T i∈Ub

where si = sI when task i is for inbound containers, while it is sO when task i is for
outbound containers. The following property is then useful for designing the solu‑
tion algorithm.
( )
Property 1 If W bt Zbt is a non-decreasing and convex function of Zbt, then TC(X)
is a convex function of X.

Proof Refer to Appendix 1.

Note that the expression for the expected waiting time in the M/M/1 queue (Refer
to Appendix 2) and regression Eqs. (1) and (2) satisfy the condition for Property 1.
The example of outbound tasks in Table 1 can be formulated as follows using the
input parameters in Table 2:
[ ( ) ] [ ( ) ]
TC(X) = 2S111 Z11 + d11 + 10a11 X111 + 2S112 Z12 + d12 + 10a12 X112
[ ( ) ] [ ( ) ]
+ 2S121 Z21 + d11 + 10a11 X121 + 2S122 Z22 + d12 + 10a12 X122
[ ( ) ] [ ( ) ]
+ 2S211 Z11 + d21 + 10a21 X211 + 2S212 Z12 + d22 + 10a22 X212
[ ( ) ] [ ( ) ]
+ 2S221 Z21 + d21 + 10a21 X221 + 2S222 Z22 + d22 + 10a22 X222

subject to

13
Scheduling appointments for container truck arrivals…

X111 + X121 + X211 + X221 ≤ u1 ,

X112 + X122 + X212 + X222 ≤ u2 ,
X111 + X211 ≤ m11 , X112 + X212 ≤ m12 ,
X121 + X221 ≤ m21 , X122 + X222 ≤ m22 ,
X111 + X112 + X121 + X122 = n1 ,
X211 + X212 + X221 + X222 = n2 ,
X111 ≤ v111 , X112 ≤ v112 , X121 ≤ v121 , X122 ≤ v122 ,
X211 ≤ v211 , X212 ≤ v212 , X221 ≤ v221 , X222 ≤ v222

5 Algorithms for solving the appointment scheduling problem

This section proposes an appointment-scheduling algorithm (ASA) for solving (AS),

which is based on the Frank–Wolfe algorithm. The Frank–Wolfe algorithm is a pop‑
ular approach( for)the optimizing the problem with non-linear objective terms.
Let X = Xibt . The Problem (3)–(8) is linearized. Starting with an initial point,
X0, at the k th step, TC(X) is replaced with( )its first (order
)T (Taylor )series expansion
about the solution, Xk , i.e., Fk (X) = TC Xk + ∇TC Xk X − Xk . The following
problem is solved iteratively until a sufficiently converged solution is obtained.
(LPk : Linearized problem at iteration k)
( )T
Minimize ∇TC Xk X

subject to constraints (4)–(7) and the constraint that Xibt is a non-negative real
number. ( ) ( ) ( )
Let ∇TC Xk be Qibt Zbt . When Qibt Zbt is fixed, the problem ( LPk ) can be for‑
mulated as a minimum st maximum flow network (MCMFN) model, as illustrated
in Fig. 4 for the case with outbound tasks. Using the network illustrated in Fig. 4,
the minimum cost maximum flows can be found using the shortest augmenting path
(SAP) algorithm.
The weighted directed graph of the transportation network (G(V, E)), as shown in
Fig. 4, can be constructed as follows. Three types of nodes are provided: time nodes,
of which node “Tt ” represents the time window “ t ”; block-time nodes, of which “Bbt ”
represents block “b” at time window “ t ”; and task nodes, of which “Ui” represents
task “i ”. Two nodes, S and E (source and sink, respectively), are added to the network.
Note that for an outbound task, the arcs are directed from the preceding block-time
nodes for more than one block to a task node, while, for an inbound task, the arcs are
directed from the preceding block-time nodes for only a single block to a task node.
This is because the containers for an outbound task may have a space allocation at
more than one block, while an inbound task is related to the containers stored at only
a single block. Figure 5 presents the network for the problem in Tables 1 and 2. The
capacity ([ut ]) of the arc from “S ” to “Tt ” represents the maximum number of trucks
available at time window t . The capacity ([mbt ]) of the arc from “Tt ” to “Bbt ” is the

13
 S. Yi et al.

[m 11 ],0 [v111 ],Q111

T1 B11 U1

[u 1 ],0
[m 21 ],0
[n 1 ],0
B21

S E

B12 [n 2 ],0
[m 12 ],0
[u 2 ],0

T2 [m 22 ],0
B22 [v222 ],Q222
U2

Fig. 4  Network representation of MCMFN model for the outbound tasks

[8],0 [4],30.97
X111=2
T1 B11 U1

[10],0
[8],0 X122=3
[9],0
B21

S E

B12 [7],0
[8],0
[10],0 X211=2

TC=438.77
T2 [8],0 B22 [4], 31.46 U2
X112=1

Fig. 5  Network representation for MCMFN for the outbound tasks with a solution

maximum number of slots available for appointments at block b within time window t .
The capacity of the arc ([vibt ]) from “Bbt ” to “Ui” means the number of slots in block
b allocated to outbound task i at time window t in (the yard
) space plan. The cost (Qibt )
of the arc from node “Bbt ” to “Ui” represents Qibt Zbt . Figure 5 gives an example of
the MCMFN model for the example of outbound tasks in Table 1.
The main solution procedure for the appointment-scheduling algorithm (ASA)
can be described as follows:
( )
Step 0 Set iteration = 0. Let Xibt =(0 for( all))
i , b, and t . Calculate Qibt Zbt . Solve
MCMFN model ( for) a given Qibt Zbt . Let the solution be X
Step 1 Calculate Qibt Zbt using the solution X . Solve the MCMFN model. Let the
solution be X ′. Let D = X � − X . Minimize TC(X + 𝜏D) by the golden sec‑
tion search for the optimal value of 𝜏, 𝜏 ∗ (0 ≤ 𝜏 ∗ ≤ 1). Let X = X + 𝜏 ∗ D
Step 2 If the change of TC(X) is smaller than 𝜀(0 < 𝜀 ≪ 1), then go to Step 3, oth‑
erwise, set iteration = iteration + 1. and go to Step 1
Step 3 Convert the solution X to an integer-valued solution and stop

13
Scheduling appointments for container truck arrivals…

The algorithm used to solve the MCMFN model is the shortest augmenting path
algorithm (Ahuja et al. 1993). Duan (1994) modified this algorithm and proposed
a more efficient shortest path algorithm (i.e., the shortest path = faster algorithm:
SPFA). This algorithm extends the Bellman–Ford algorithm (Bellman 1958) and
Dijkstra’s algorithm (Dijkstra 1959) to find the shortest path for graphical models
that even have negative edge weights.
Step 3 of the main solution procedure (ASA) converts the real-valued solution
into an integer-valued one. For this conversion, a heuristic rule is proposed. Note
that the procedure should not make the solution violate constraints (4)–(7). Values
of the solution are repaired so that the modified solution satisfies the constraints in
the order of (6), (4), (5), and (7) one by one group. In the process, it is attempted
that the preceding groups of constraints, for which the modifications of the solution
have already been completed for the feasibility, are not violated again by a succeed‑
( ) the value of an Xibt is to be decreased, the value
ing step. When ( of) Xibt with the
largest Qibt Zbt is decreased, because Xibt with the largest Qibt Zbt is expected to
reduce TC in a highest rate. (When) the value of an Xibt is to be increased, the value
of Xibt with the smallest Qibt Zbt is increased, because it is expected to increase TC
in a lowest rate. When constraint (6) is violated, after one Xibt is decreased, another
Xibt must be increased. Whenever
( ) there is a change in the value of an Xibt , the value
of the corresponding Qibt Zbt is updated. The heuristic procedure for converting the
solution to integer-valued one (HPCI) can be described in detail as follows:

Step 1: Round off

Step 2: (Check whether constraints (6) are satisfied)
while ( ) do

Select

Select

end while
Step 3: (Check whether constraints (4) are satisfied)
while ) do
Select

Select incurring no
additional violation of constraints (4).
(for maintaining the feasibility to constraints (6))
end while
Step 4: (Check whether constraints (5) are satisfied)
Repeat the procedure similar to Step 3 until all the constraints (4), (5), and (6) are sat-
isfied.
Step 5: (Check whether constraints (7) are satisfied)
Repeat the procedure similar to Step 3 until all the constraints (4), (5), (6), and (7) are
satisfied.

13
 S. Yi et al.

5.1 Numerical example

Using the example of outbound tasks in Tables 1 and 2 and the regression equations
for the
( expected
) system time (1) and (2),
( the
) solution procedure was applied. Let
cSibt Zbt + dit + bait be denoted by Eibt Zbt .
∑ ∑ ∑[ ( ) ]
TC(X) = cSibt Zbt + dit + bait Xibt
i∈U b∈Bi t∈T
∑ ∑ ∑[ ( ) ]
= cSibt Zbt + dit + bait Xibt
i∈U I b∈Bi t∈T
∑ ∑ ∑[ ( ) ]
+ cSibt Zbt + dit + bait Xibt
i∈U O b∈Bi t∈T

( ) ( )
For i ∈ U I , Eibt Zbt = cSibt Zbt + dit + bait
[ ]
∑
≅ c 6.725 + 0.242wbt + 0.459 Xibt + dit + bait
i∈Ub

( ) ( )
For i ∈ U O , Eibt Zbt = cSibt Zbt + dit + bait
[ ]
∑ .
≅ c 6.331 + 0.242wbt + 0.459 Xibt + dit + bait
i∈Ub

I
Let r = 0.459c and ribt = 6.725c + 0.242cwbt + dit + bait ,
O
ribt = 6.331c + 0.242cwbt + dit + bait .

( ) ∑ ( ) ∑
Then, for i ∈ U I , Eibt Zbt = r I
Xibt + ribt , for i ∈ U O , Eibt Zbt = r O
Xibt + ribt .
i∈Ub i∈Ub

𝜕TC(X) ( ) ∑
I I
= Qibt Zbt = 2r Xibt + 2rXibt + ribt = 2rZbt + ribt for i ∈ U I ,
𝜕Xibt i∈U −{i} b
O
= 2rZbt + ribt for i ∈ U O .
(9)
Figure 6 shows the change of the objective value as the iteration progresses for
the example of outbound tasks in Table 1. 𝜀 was set to 10−3. The expected total cost
(TC ) converged to $438.60 during 11th iterations. This expected total cost in Fig. 6
was evaluated using the real numbered solution. The final real numbered solution
should be converted to the integer numbered one using the procedure of the HPCI.
Table 3 lists the final integer numbered solution ( X ) whose total cost is $438.77.
Table 3 lists the results of the solution procedure with various cost terms.

13
Scheduling appointments for container truck arrivals…

Total cost ($)

439.40
439.280
439.20
439.00
438.754
438.80 438.857
438.635 438.617 438.601 438.595
438.60 438.686 438.625 438.609 438.597
438.40
438.20
1 2 3 4 5 6 7 8 9 10 11 Iteration

Fig. 6  Change in the objective value by appointment scheduling algorithm (ASA) as the iteration pro‑
gresses

6 Numerical experiments

This section consists of 3 sub-sections. Firstly, an experiment was carried out to

compare the various algorithms with that proposed in this study for randomly gener‑
ated problems. Secondly, some experiments were performed to examine the effects
of the appointment system and explore useful characteristics of the optimal solu‑
tion by a sensitivity analysis. The practical data was used, which includes the infor‑
mation on the arrival time window of each truck and the assigned storage location
of each container. Thirdly, the effect of appointments by multiple trucking compa‑
nies was examined. For the experiment, one of randomly generated problems was
selected and modified to the problem with multiple trucking companies.
In Sect. 6.1, Table 5 compares various approaches using 28 randomly gener‑
ated problems, where c = 2 ($/min), b = 20($/h), mbt = 10, and ut = 30 for all b and t .
The demurrage charge (dit) is $ 15 and applied on randomly chosen 20% of tasks dur‑
ing one-third time windows for outbound ones or during two-third time windows for
inbound ones. The truck delivery time (ait) has a value between 0.5 and 1.1 h with the
peaks twice a day. The arrivals of the internal trucks for ship operations (wbt) were gen‑
erated randomly.
In Sect. 6.2, the data from “D” container terminal in Busan, Korea, were collected
over a 30-day period, for the analysis based on the empirical data. The number of
blocks was 7. The length of the scheduling time window in a day was 24 h. The sched‑
uling was assumed to have been done once a day. The other input parameters were set
to be the same as those in the experiment of Sect. 6.1. All the containers to be delivered
were assumed to be inbound ones. Table 4 lists the number of trucks and the workload
level of each scheduling problem, which is the number of trucks divided by the maxi‑
mum number of available trucks. The original number of truck arrivals on each day
was used to illustrate the performance of the proposed algorithm and the original
arrival times of trucks were used for a comparison with the appointments from the pro‑
posed algorithm, which was shown in Table 7. The results of the following experiments
were based on the average value of 30 problems in Table 4. In addition, it shows the
truck cost within the terminal, the demurrage cost, the container delivery cost, the
∑ ∑ ∑ � � ��
expected system time per truck ( Sibt Zbt Xibt.) and container delivery
∑ ∑ ∑ � � i∈U b∈Bi t∈T
time per truck ( ait Xibt).
i∈U b∈Bi t∈T

13
13
Table 3  Costs for the final solution with outbound tasks
( )
i b t Xibt Zbt Expected system Qibt Zbt Truck cost within Dem. Cost ($) Container delivery Total cost ($)
time (min) terminal ($) cost ($)

1 1 1 2 4 17.30 30.97 34.60 10.00 10.00 54.60

1 2 1 2 4 17.79 31.46 35.57 10.00 10.00 55.57
1 1 2 2 4 16.82 30.49 33.64 0.00 20.00 53.64
1 2 2 3 4 26.68 31.46 53.36 0.00 30.00 83.36
2 1 1 2 4 17.30 30.97 34.60 10.00 10.00 54.60
2 2 1 2 4 17.79 31.46 35.57 10.00 10.00 55.57
2 1 2 2 4 16.82 30.49 33.64 0.00 20.00 53.64
2 2 2 1 4 8.89 31.46 17.79 0.00 10.00 27.79
Sum 139.38 – 278.77 40.00 120.00 438.77
S. Yi et al.
Scheduling appointments for container truck arrivals…

Table 4  Input data of each day Day No. of Workload ­levela Day No. of Workload level
used in the experiment trucks trucks
(A) (A)

1 346 0.481 16 381 0.529

2 237 0.329 17 341 0.474
3 230 0.319 18 199 0.276
4 276 0.383 19 237 0.329
5 232 0.322 20 82 0.114
6 137 0.190 21 150 0.208
7 125 0.174 22 403 0.560
8 475 0.660 23 399 0.554
9 383 0.532 24 206 0.286
10 174 0.242 25 225 0.313
11 144 0.200 26 161 0.224
12 188 0.261 27 102 0.142
13 45 0.063 28 142 0.197
14 97 0.135 29 394 0.547
15 352 0.489 30 213 0.296
a
Workload level = (A)/(No. of available trucks per hour × 24)

For the experiments in Sect. 6.3, Problem ID 24 in Table 5 was used where it is
assumed that 20 tasks were randomly assigned to either of 2 trucking companies.

6.1 Comparison among various solution procedures

When regression Eqs. (1) and (2) are used, the mathematical model in this sec‑
tion becomes a quadratic programming (QP) model, which was solved using the
CPLEX.1 In addition, the performance of the proposed appointment scheduling
algorithm (ASA) was compared with other algorithms introduced below.

6.1.1 Myopic heuristic algorithm (Myopic)

A simple heuristic algorithm is proposed where appointments for each task are
assigned to the least cost time window task by task. They are assigned to time win‑
dows in a way of minimizing the cost of the task considering the expected system
times updated by appointment assignments of tasks scheduled so far.

Step 0 Let Xibt = 0 for all i , b, and t ( )

Step 1 Update the expected system time (Sibt Zbt ) using the solution X.

1
https​://www.ibm.com/analy​tics/cplex​-optim​izer.

13
 Table 5  Comparison of the performance of various approaches
No. U B T No. of trucks Type of tasks QP Myopic TP-FST ASA
a
TC ($) (A) Gap (%) TC ($) (B) (B/A)-1 (%) TC($) (C) (C/A)-1 (%) TC($) (D) (D/A)-1 (%)

13
1 2 2 2 14 O 397 0.00 421 6.29 397 0.00 397 0.00
2 3 3 6 25 I/O 728 0.00 790 8.43 807 10.84 728 0.00
3 4 3 6 30 I/O 892 0.00 987 10.70 993 11.29 893 0.09
4 5 4 6 35 I/O 1037 0.00 1155 11.38 1164 12.22 1038 0.05
5 6 4 6 40 I/O 1188 0.00 1283 8.04 1321 11.21 1189 0.07
6 6 4 6 80 I/O 2553 0.00 2803 9.77 2777 8.76 2555 0.06
7 6 4 12 80 I/O 2358 0.03 2646 12.23 2632 11.65 2358 0.03
8 6 4 15 80 I/O 2245 0.08 2572 14.53 2534 12.83 2247 0.06
9 6 4 18 80 I/O 2239 0.10 2525 12.78 2548 13.83 2239 0.02
10 6 4 24 80 I/O 2212 0.03 2567 16.04 2532 14.48 2214 0.10
11 7 7 24 125 I/O 3490 0.13 4185 19.92 4009 14.88 3490 0.00
12 7 7 24 232 I/O 6872 0.09 8105 17.95 7808 13.62 6878 0.09
13 7 7 24 280 I/O 8479 0.06 10,126 19.42 9800 15.58 8495 0.19
14 7 7 24 341 I/O 10,569 0.09 12,307 16.44 11,940 12.97 10,609 0.37
15 7 7 24 450 I/O 14,430 0.07 16,725 15.90 16,356 13.35 14,469 0.27
16 8 7 24 450 I/O 14,138 0.06 16,554 17.08 16,199 14.57 14,162 0.17
17 8 8 24 450 I/O 14,125 0.08 16,388 16.02 16,153 14.36 14,132 0.05
18 8 9 24 450 I/O 14,075 0.09 16,386 16.42 16,061 14.11 14,089 0.10
19 8 10 24 450 I/O 13,959 0.09 16,389 17.41 16,199 16.05 13,974 0.11
20 8 11 24 450 I/O 13,877 0.11 16,386 18.08 16,199 16.73 13,881 0.02
21 10 12 24 500 I/O 15,194 0.10 18,325 20.61 18,188 19.70 15,222 0.19
22 12 12 24 550 I/O 17,160 0.09 20,246 17.98 20,135 17.33 17,183 0.13
23 15 12 24 600 I/O 18,513 0.14 22,034 19.02 22,119 19.48 18,527 0.08
24 20 12 24 650 I/O 20,286 0.12 24,135 18.98 24,119 18.90 20,345 0.30
S. Yi et al.
 Table 5  (continued)
No. U B T No. of trucks Type of tasks QP Myopic TP-FST ASA
a
TC ($) (A) Gap (%) TC ($) (B) (B/A)-1 (%) TC($) (C) (C/A)-1 (%) TC($) (D) (D/A)-1 (%)

25 25 12 24 700 I/O 21,985 0.16 26,324 19.73 26,528 20.66 22,017 0.14
26 50 12 24 1030 I/O 33,092 0.11 37,134 12.21 37,900 14.53 33,156 0.18
27 75 12 24 1545 I/O 52,333 0.06 56,122 7.24 56,958 8.84 52,388 0.11
28 100 12 24 2060 I/O 73,305 0.07 76,230 3.99 76,933 4.95 73,365 0.08
Average – 0.07 – 14.45 – 13.49 – 0.11
a
Gap between the lower bound and the best feasible solution obtained within the earlier one between two pre-specified deterministic tuning times (400,000 ticks or 3600 s)
in ­CPLEX®
Scheduling appointments for container truck arrivals…

13
 S. Yi et al.

Step 2 If all tasks are assigned, then stop. Otherwise, select a task ( i ) among unas‑
signed tasks randomly. Assign the task ( i ) to time window t (and block b
for the outbound
( ) tasks) with the least cost which is calculated by using the
updated Sibt Zbt , while the solution should satisfy constraints (4)–(7). If
there are multiple least cost solutions, then choose b and t randomly among
them for the assignment. Update the solution Xibt . Go to Step 1.

6.1.2 Two phase algorithm with fixed system time (TP‑FST) (Riaventin

and Kim 2018)

Riaventin and Kim (2018) proposed a two phase heuristic algorithm for schedul‑
ing multiple appointments considering
{ ( )a known expected
} system time at each block
and time window. The term, cSibt Zbt + dit + bait , in the objective function of (3)
is dependent on the solution ( Xibt ). On the other hand, Riaventin and Kim (2018)
assumed that this term is given under the assumption that that Xibt = 0 for all i , b,
and t . That is, the fixed expected system time was assumed to have been evaluated
based on the estimated number of arrivals of internal trucks for ship operations and
of road trucks for the appointments confirmed previously. From this assumption, the
objective function of (3) and constraints (5) and (6) may be considered to be the
transportation problem. The two-phase algorithm consists of the first phase for solv‑
ing the transportation problem and the second phase for modifying the solution to
satisfy some additional constraints.

6.1.3 Iterative two phase algorithm with dynamic system time (TP‑DST)

The TP-DST decomposes the set of tasks into multiple groups and the groups of
tasks are then scheduled one by one. The expected system time is updated between
the scheduling for each group. This process is repeated until the expected system
times converge. This algorithm was designed to complement the weakness of TP-
FST, which is based on the fixed expected system time. According to the experi‑
ment, the solution oscillates between two solutions without any convergence in the
objective value or solutions.
In this experiment, the CPLEX was run to solve the QP formulation within a pre-
specified deterministic tuning time due to time limitations. The gap in the QP solu‑
tion indicates the difference between the best feasible solution and the lower bound
obtained within the pre-specified time limit.
As illustrated in Tables 5 and 6, a comparison of the ASA with the QP showed
that the ASA in this study obtained solutions close to those by QP within 0.11% in
TC (total cost) within reasonable computation times. The myopic heuristic algorithm
and the previous work (TP-FST) showed the TC larger than that by QP by 14.45 and
13.49% on average. Table 6 shows that problems 1-6 may be solved optimally by QP
within a reasonable time. The largest size of problems 1-6 is 4 tasks, 4 blocks, and
80 trucks, which is far below of the practical size.

13
Scheduling appointments for container truck arrivals…

Table 6  Comparison of the computation time of various approaches

No. U B T No. of Type of QP Myopic rule TP-FST ASA
trucks tasks Comp. time Comp. time Comp. time Comp. time
(s) (s) (s) (s)

1 2 2 2 14 O 0.6 3.4 0.0 0.0

2 3 3 6 25 I/O 0.7 3.7 0.0 0.2
3 4 3 6 30 I/O 0.7 3.9 0.0 0.1
4 5 4 6 35 I/O 0.9 4.6 0.0 0.5
5 6 4 6 40 I/O 0.9 4.7 0.0 0.4
6 6 4 6 80 I/O 68.6 5.4 0.0 0.2
7 6 4 12 80 I/O 2508.7 7.0 0.0 0.6
8 6 4 15 80 I/O 3517.2 7.1 0.0 0.5
9 6 4 18 80 I/O 1710.8 7.4 0.0 1.3
10 6 4 24 80 I/O 1844.4 7.9 0.0 0.4
11 7 7 24 125 I/O 1381.8 9.5 0.1 2.2
12 7 7 24 232 I/O 3001.2 9.9 0.1 2.9
13 7 7 24 280 I/O 1153.7 9.4 0.1 2.3
14 7 7 24 341 I/O 925.8 8.9 0.0 15.7
15 7 7 24 450 I/O 685.4 9.7 0.1 21.5
16 8 7 24 450 I/O 2497.1 11.2 0.1 34.1
17 8 8 24 450 I/O 769.2 10.4 0.1 40.5
18 8 9 24 450 I/O 824.2 11.7 0.1 15.3
19 8 10 24 450 I/O 588.0 10.8 0.1 49.1
20 8 11 24 450 I/O 562.8 12.4 0.1 54.9
21 10 12 24 500 I/O 1166.0 13.5 0.1 73.0
22 12 12 24 550 I/O 1372.0 15.5 0.2 151.4
23 15 12 24 600 I/O 1073.6 18.6 0.2 161.1
24 20 12 24 650 I/O 847.9 23.3 0.2 257.5
25 25 12 24 700 I/O 2949.8 34.1 0.2 355.3
26 50 12 24 1030 I/O 3600+ 16.8 0.3 1076.8
27 75 12 24 1545 I/O 3600+ 22.3 0.5 2380.9
28 100 12 24 2060 I/O 3600+ 26.8 1.0 3022.1

6.2 Analysis on the appointment scheduling by using empirical data

6.2.1 Effect of scheduling truck appointments

To evaluate the effect of the truck appointment on the various cost terms, the cases
without the appointment were compared with those with it. For this experiment, the
problems in Table 4 were used so that all of the tasks were for inbound containers. For
the cases without appointments, the original arrival times of the trucks collected from
“D” terminal were used. Table 7 shows the result of the comparison, which estimates

13
 S. Yi et al.

Table 7  Comparison of the various cost terms between cases with and without the appointment system
for the problems with inbound containers
Truck cost within Dem. cost ($/day) Container Total cost ($/day)
the terminal ($/ delivery cost
day) ($/day)

Case without appointment 4295.85 479.50 3851.67 8627.01

scheduling (A)
Case with appointment 4089.35 70.50 3174.00 7333.85
scheduling (B)
Relative value (B/A) (%) 95.2 14.7 82.4 85.0

that the daily total cost can be reduced by 15.0% on average by making appointments.
In particular, the container delivery cost was reduced by 17.6% on average.

6.2.2 Impact of the workload level of a terminal on the appointment scheduling

Figure 7 examines the influence of workload level of a terminal on the cost terms.
The problems in Table 4 were used for this experiment. Figure 7 shows that all the
cost terms increase with increasing workload level. In particular, when the expected
system time increased, not only the truck cost within the terminal increased but also
the demurrage and the delivery costs increased. The increasing rate of the latter two
cost terms was higher than the former one.

6.2.3 Impact of the type of containers on various cost terms

This experiment compares the cost terms of three cases: case with only inbound con‑
tainers; case with only outbound containers; case with both inbound and outbound
containers. For the case with both types of containers, the problems in Table 4 were
modified so that a half of the tasks are for outbound containers and the remaining are
for inbound containers. For the case with only outbound containers, the problems in
Table 4 were modified so that two candidate blocks are assigned to each task consid‑
ering the flexibility of locating outbound containers. The case with both inbound and
outbound containers refers to the result in Table 7. Table 8 summarizes the results of
the comparison. The problems with only outbound tasks showed 7.1% lower average
total cost per day than the ones with only inbound tasks, especially 8.1% in the cost
within the terminal and 100.0% in demurrage charge. Besides, it also showed a lit‑
tle lower cost than the case with both types of containers. In the case with outbound
containers, truck appointments could be allocated to any block between two blocks,
which gave appointments a higher chance to reduce the cost terms.

13
Scheduling appointments for container truck arrivals…

Truck cost within terminal (per truck, $/day) Dem. Cost + Delivery cost (per truck, $/day)
18.0 18.0
17.0 17.0
16.0 16.0
15.0 15.0
14.0 14.0
13.0 13.0
12.0 12.0
11.0 11.0
10.0 10.0
9.0 9.0
0.0 0.2 0.4 0.6 0.8 0.0 0.2 0.4 0.6 0.8
Workload level Workload level

Fig. 7  Cost per truck for various workload levels

Table 8  Comparing various cost terms among the cases with various types of containers
Truck cost within Dem. cost ($/day) Container Total cost ($/day)
the terminal ($/ delivery cost
day) ($/day)

Case with only inbound 4089.35 70.50 3174.00 7333.85

containers (A)
Case with both types of 3970.27 81.50 3167.33 7219.10
containers (B)
Case with only outbound 3759.75 0.00 3054.67 6814.42
containers (C)
Relative value (B/A) (%) 97.1 115.6 99.8 98.4
Relative value (C/A) (%) 91.9 0.0 96.2 92.9

6.2.4 Sensitivity analyses of various cost parameters

Sensitivity analyses for the parameters were carried out using the problems with
the half inbound and half outbound tasks, as illustrated in the problems in Table 8.
Tables 9, 10, 11, and 12 present the results of the sensitivity analyses. The results in
Tables 9, 10, 11, and 12 summarize the average values of problems 1–5. The values
of the parameters are the same as those in Table 4.
Table 9 lists the results of the sensitivity analysis for the various levels of truck
cost per unit time within the terminal (c). As the truck cost per unit time within the
terminal increases, the truck cost within the terminal increases rapidly compared to the
other cost terms. On the other hand, appointments move to time windows with shorter
expected system times at a cost of a slight increase in the container delivery time.
Thus, a container terminal with high congestion can induce trucking companies to
move road truck arrivals to less congested time periods by increasing the appointment
charge which may be included into the truck cost per unit time within the terminal.
Table 10 lists the impact of the container delivery cost per hour which can be
influenced by the variation in the overhead cost of a truck, labor cost for the truck
driver, and fuel cost for the truck. Hence, while the container delivery cost per
hour decreases, the arrivals of trucks are distributed uniformly to various time win‑
dows to reduce their appointment and demurrage charges. On the other hand, as

13
 S. Yi et al.

Table 9  Sensitivity analysis for various values of truck cost per unit time within the terminal (c)
c Truck cost Dem. Container Total cost ($/ Expected sys‑ Container
within the ter‑ cost ($/ delivery cost truck) tem time (min/ delivery time (h/
minal ($/day) day) ($/day) truck) truck)

0.5 4.70 0.00 12.42 17.12 9.41 0.62

1 8.91 0.00 12.82 21.73 8.91 0.64
1.5 12.86 0.02 13.21 26.09 8.58 0.66
2 16.78 0.18 13.42 30.38 8.39 0.67
5 38.77 1.20 14.34 54.31 7.75 0.72
10 75.01 2.12 15.08 92.21 7.50 0.75
20 149.08 2.50 15.33 166.92 7.45 0.77

Table 10  Changes in the cost terms for various values of container delivery cost per unit time (b)
b Truck cost Dem. Container Total cost ($/ Expected sys‑ Container
within the ter‑ cost ($/ delivery cost truck) tem time (min/ delivery time (h/
minal ($/day) day) ($/day) truck) truck)

5 15.91 0.00 3.79 19.69 7.95 0.76

10 16.29 0.00 7.11 23.41 8.15 0.71
15 16.60 0.02 10.32 26.94 8.30 0.69
20 16.78 0.18 13.42 30.38 8.39 0.67
50 17.45 1.01 31.05 49.52 8.73 0.62
100 17.76 1.35 61.12 80.23 8.88 0.61
200 17.78 1.58 121.81 141.17 8.89 0.61

Table 11  Sensitivity analysis for various values of the maximum availability of trucks (ut)
ut Truck cost within Dem. Container Total cost Expected system Container
the terminal ($/ Cost ($/ delivery cost ($/truck) time (min/truck) delivery time (h/
day) day) ($/day) truck)

15 16.28 0.13 14.77 31.18 8.14 0.74

20 16.54 0.20 14.00 30.73 8.27 0.70
25 16.70 0.21 13.60 30.51 8.35 0.68
30 16.78 0.18 13.42 30.38 8.39 0.67
40 17.01 0.13 13.13 30.27 8.50 0.66
50 17.04 0.11 13.10 30.25 8.52 0.66

the container delivery cost increases, the trucking company attempts to reduce the
container delivery time by bearing the cost for the appointment and demurrage of
containers.
This study examined how much the maximum available number of trucks per
time window (ut ) affects the various cost terms of the truck appointment system.
Table 11 shows the results. If a trucking company has a small number of available

13
Scheduling appointments for container truck arrivals…

Table 12  Sensitivity analysis for various values of the maximum availability of appointment slots (mbt)
mbt Truck cost within Dem. Container Total cost Expected system Container
the terminal ($/ cost ($/ delivery cost ($/truck) time (min/truck) delivery time (h/
day) day) ($/day) truck)

3 15.81 1.63 13.88 31.32 7.91 0.69

4 16.33 0.82 13.57 30.72 8.17 0.68
5 16.61 0.41 13.50 30.52 8.30 0.68
6 16.69 0.32 13.41 30.42 8.34 0.67

trucks per time window, then the total cost increases because of the lower flexibility
in assigning trucks to the available time windows, which results in a loss of opportu‑
nity to reduce the total cost. Note that the increase in the maximum available num‑
ber of trucks contributes to the reduction in the demurrage and container delivery
cost (or time), while it adversely affects the truck cost within the terminal or the
system time of trucks.
In a container terminal, internal trucks may arrive at a block for ship operations,
which increases the system time (congestion) of road trucks at the block. Because
the terminal operator has information on the ship operation schedule, the terminal
operator may restrict the maximum number of road trucks or appointments. This
experiment tests the effect of the maximum number of appointment slots on the var‑
ious performance measures or cost terms, whose results are listed in Table 12. As
the maximum number of available slots decreases, the total cost increases but the
expected system time decreases; hence, the appointment cost decreases. On the other
hand, the demurrage cost of containers and the container delivery cost increase.

6.3 Appointments by multiple trucking companies

As in previous sections, it was assumed that a single trucking company performed

the appointment scheduling and all the tasks were known from the beginning. On
the other hand, when multiple trucking companies deliver containers for a container
terminal, more than one trucking company may do the appointment scheduling at dif‑
ferent times. The appointments of one company can affect the result of the appoint‑
ment scheduling of the next trucking company. This section discusses the impacts of
a situation with multiple trucking companies on the performance of the algorithm.
An experiment was conducted to examine the effect of appointments by multiple
trucking companies. The tasks for each day in Table 5 were assigned to one between
two trucking companies. All tasks assigned to one company (let the company be
company A) were scheduled, and then using the appointments of all the tasks of
company A, the other company (company B) schedules the appointments for all the
assigned tasks. Company A then schedules the appointments again based on the
appointment information by company B. This process is repeated until the maxi‑
mum difference between the values of the new Ŝ ibt (X) and the previous one for all

13
 S. Yi et al.

Table 13  Convergence of the total cost from AITER to the optimal total cost
Iteration number Company A Company B Both companies

TC ($) Ratio of TC TC ($) Ratio of TC TC ($) Ratio of TC(%)

(%) (%)
(I) (I/A) (I/B) (II) (II/A) (II/B) (III) (III/A) (III/B)

0 11,084 109.5 115.9 12,916 127.0 118.7 24,000 118.3 117.4

1 9665 95.5 101.1 11,235 110.5 103.2 20,901 103.0 102.2
2 9566 94.5 100.0 11,094 109.1 101.9 20,660 101.9 101.0
3 9561 94.5 100.0 11,039 108.6 101.4 20,601 101.6 100.7
4 9558 94.5 99.9 11,013 108.3 101.2 20,572 101.4 100.6
5 9557 94.5 99.9 10,987 108.1 100.9 20,544 101.3 100.5
6 9552 94.4 99.9 10,967 107.9 100.7 20,519 101.2 100.3
7 9551 94.4 99.9 10,960 107.8 100.7 20,511 101.1 100.3
8 9550 94.4 99.9 10,954 107.7 100.6 20,504 101.1 100.3
9 9550 94.4 99.9 10,943 107.6 100.5 20,493 101.0 100.2
10 9550 94.4 99.9 10,937 107.6 100.5 20,487 101.0 100.2
Optimal solution by QP (A) 10,118 10,166 20,285
Final solution by AITER (B) 9563 10,885 20,449

i , b, and t become less than 𝜀(0 < 𝜀 ≪ 1). This procedure is called the “alternating
iterative procedure (AITER)”.
Table 13 lists the solution from the alternating iterative solution by two trucking
companies converging as the iteration progresses, during which they compromise
their appointments by sharing the information on the appointments between each
other. In addition, the total cost obtained from the alternating iterative procedure
by the two companies (AITER) approaches. The result showed a gap level of only
0.80% (III/B) within the ­10th iteration in comparison with the final objective value
of this approach with = 10−4.
At the beginning, both trucking companies did not have information about
appointments from the other, which led to an unstable solution and total cost. Nev‑
ertheless, the total cost converged gradually by sharing information on the appoint‑
ments of the other company. The total costs of trucking companies A and B were
reduced by 13.8 and 15.3%, respectively, after 10th iterations. Note that the reduction
was already reduced by 13.7 and 14.5%, respectively, after 3rd iterations. That is, the
largest portion of the reduction occurs at the first few iterations. Table 13 shows that
the gap of the proposed procedure (AITER) from the optimal solution as the central‑
ized scheduling of the QP becomes less than 2% (III/A) at the 3rd iteration.
This experiment suggests that the proposed algorithm can be used for multiple
trucking companies, in which case multiple companies can modify their appointments
at different times using the information on the appointment from other companies. The
appointments from this alternating iterative procedure (AITER) rapidly converge to a
satisfactory solution close to the final solution by scheduling all the tasks for the ter‑
minal by a centralized decision maker. This indicates that even when multiple trucking

13
Scheduling appointments for container truck arrivals…

companies make appointments asynchronously at different times, they can reach

appointments very close to the optimal solution from the entire system’s point of view
within a small number of communications with the appointment system.

7 Conclusions

As more mega-vessels call at hub ports, the congestion at container terminals in hub
ports becomes increasingly severe over time, particularly at peak hours. To motivate
trucking companies to alleviate the congestion, a scheduling algorithm for the truck
appointment is proposed, which considers different expected waiting times at differ‑
ent times in a day. A non-linear mathematical model was formulated. The objective
function was convex only if the expression of the expected waiting time was non-
decreasing and convex, which is satisfied by most waiting time estimators. Based
on the convexity of the objective function, a heuristic scheduling algorithm based
on the Frank–Wolfe algorithm was developed to find the near optimal appointments
within a reasonable computation time.
To evaluate the proposed solution algorithm, numerical experiments were per‑
formed using the linear regression equations to estimate the expected waiting time,
which results in a quadratic programming (QP) model for the appointment-sched‑
uling problem. The QP model could be solved by the CPLEX for comparing the
solutions with those by the heuristic algorithm. The proposed heuristic algorithm
obtained solutions close to the optimal ones with an average gap of 0.11% in the total
cost and outperformed the previous work by approximately 13.49% of the total cost.
Sensitivity analyses of the empirical data were conducted for some factors that
can influence the performance of the appointment system. The total cost tended to
decrease when the number of available trucks in each time window increased. The
problems with only outbound tasks showed a 7.1% lower total cost than the ones
with only inbound tasks. In addition, the total cost could be reduced by 15.0% by
applying the appointment system. The impacts of various input parameters on the
system (waiting) time, container delivery time, and various cost terms were ana‑
lyzed. The changes in the performance of the appointment schedule when multiple
trucking companies construct appointment schedules were also discussed.
A numerical experiment showed that the proposed algorithm can be used for
multiple trucking companies, in which case multiple companies can modify their
appointments at different times using the information on appointments from other
companies. Asynchronous appointments by multiple trucking companies may con‑
verge rapidly to a satisfactory solution close to the final solution by scheduling all
the appointments for the terminal through a centralized decision maker.
The solution algorithm in this study can solve the problems with other types of
expressions for estimating the expected waiting times only if they satisfy the non-
decreasing and convexity properties. More numerical experiments can be done to
examine the effects of different types of waiting time expressions on the performance of
the algorithm and individual cost terms of the objective function. This study addressed
the appointment-scheduling problem for one container terminal. On the other hand, for
the trans-shipment containers, the appointments of two terminals are required, which

13
 S. Yi et al.

was not discussed in this study. In addition, how to motivate road trucks to arrive at the
terminal as close as possible to the appointed times and make appointments as early as
possible was not discussed. These will be promising issues for future studies.

Funding Funding was provided by National Research Foundation of Korea

(2016R1D1A3B03934161).

Appendix 1: Proof of the convexity of TC(X) (Property 1)

∑ ∑ ∑[ ( ) ]
TC(X) = cSibt Zbt + dit + bait Xibt
i∈U b∈Bi t∈T
∑ ∑∑[ ( ) ]
= cSibt Zbt + dit + bait Xibt
i∈Ub b∈B t∈T
∑ ∑∑[ ( ) ]
= cWbt Zbt + csi + dit + bait Xibt ,
i∈Ub b∈B t∈T

where si = sI when task i is for inbound containers, while it is sO when task is for
outbound containers.
∑∑ ∑ [ ( ) ( ) ]
TC(X) = cWbt Zbt Xibt + csi + dit + bait Xibt
(10)
b∈B t∈T i∈Ub

An attempt is made to show that the Hessian matrix (hij) is positive definite, where
𝜕 2 TC(X)
hij = 𝜕X . Note that the second term in (10) vanishes in the second order deriva‑
ibt 𝜕Xjbt � �
∑
tives of (hij). Therefore, this study will only show that Wbt Zbt Xibt is convex with
i∈Ub
respect to (Xibt |i ∈ Ub ) for a given b and t , which leads to the convexity of TC(X) from
the theory that the sum of the convex functions is again a convex function. Note that
∑
Zbt = Xibt. The subscripts b and t were omitted in this proof for the simplicity.
i∈Ub ∑
Then, Xi = Xibt, Z = Zbt, U = Ub, and W = Wbt . Let Y = Xi, F(Y) = Xk W(Z).
k∈U
𝜕F(Y) 𝜕W(Z) 𝜕W(Z) ∑ 𝜕W(Z)
= W(Z) + Xi + Xj + Xk
𝜕Xi 𝜕Xi 𝜕Xi k∈U−{i,j}
𝜕Xi
𝜕W(Z) 𝜕Z 𝜕W(Z) 𝜕Z ∑ 𝜕W(Z) 𝜕Z
= W(Z) + Xi + Xj + X
𝜕Z 𝜕Xi 𝜕Z 𝜕Xi k∈U−{i,j} k 𝜕Z 𝜕Xi
𝜕W(Z) 𝜕W(Z) ∑ 𝜕W(Z)
= W(Z) + Xi + Xj + Xk
𝜕Z 𝜕Z k∈U−{i,j}
𝜕Z
( )
𝜕Z ∑
note that = 1 because Z = Xi .
𝜕Xi i∈U

13
Scheduling appointments for container truck arrivals…

𝜕 2 F(Y) 𝜕W(Z) 𝜕Z 𝜕 2 W(Z) 𝜕Z 𝜕W(Z) 𝜕 2 W(Z) 𝜕Z ∑ 𝜕 2 W(Z) 𝜕Z

= + Xi 2
+ + Xj 2
+ Xk
𝜕Xi 𝜕Xj 𝜕Z 𝜕Xj 𝜕Z 𝜕Xj 𝜕Z 𝜕Z 𝜕Xj k∈U−{i,j} 𝜕Z 2 𝜕Xj
𝜕W(Z) ∑ 𝜕 2 W(Z)
=2 + Xk .
𝜕Z k∈U
𝜕Z 2

𝜕 2 F(Y) 𝜕W(Z) 𝜕Z 𝜕W(Z) 𝜕 2 W(Z) 𝜕Z 𝜕 2 W(Z) 𝜕Z ∑ 𝜕 2 W(Z) 𝜕Z

2
= + + Xi 2
+ Xj 2
+ Xk
𝜕Xi 𝜕Z 𝜕Xi 𝜕Z 𝜕Z 𝜕Xi 𝜕Z 𝜕Xi k∈U−{i,j} 𝜕Z 2 𝜕Xi
𝜕W(Z) 𝜕W(Z) 𝜕 2 W(Z) 𝜕 2 W(Z) ∑ 𝜕 2 W(Z)
= + + Xi 2
+ Xj 2
+ Xk
𝜕Z 𝜕Z 𝜕Z 𝜕Z k∈U−{i,j}
𝜕Z 2
𝜕W(Z) ∑ 𝜕 2 W(Z)
=2 + Xk .
𝜕Z k∈U
𝜕Z 2

� �
� � ∑ 2 W(Z)
The Hessian matrix H = hij = 2 𝜕W(Z)
𝜕Z
+ Xk 𝜕 𝜕Z 2
J�U�, where J|U| is the
k∈U
|U| × |U| matrix of ones. It is known that J|U| is a positive semi definite matrix
(Stanley 2013, Horn and Johnson 2013). From the assumption that W(Z) is a non-
2 W(Z)
decreasing and convex function of Z , 𝜕W(Z)
𝜕Z
≥ 0 and 𝜕 𝜕Z 2
> 0, H is a positive semi
definite matrix. Hence, TC(X) is convex. Q.E.D.

Appendix 2: Proof that the expression for the expected waiting time
of M/M/1 queue satisfies the condition for Property 1
Z
W(Z) = 𝜇(𝜇−Z) in the M/M/1 queuing model, where 𝜇 represents the expected service
rate of the server. Note that 𝜇 > Z must hold. dW(Z)
dZ
1
= (𝜇−Z)2 > 0.
d2 W(Z)
dZ 2
= 2
(𝜇−Z)3
> 0. Hence, the conclusion holds. Q.E.D.

References
Ahuja RK, Magnanti TL, Orlin JB (1993) Network flows. Prentice Hall, New York
Bellman R (1958) On a routing problem. Q Appl Math 16:87–90
Chen G, Govindan K, Yang Z (2013) Managing truck arrivals with time windows to alleviate gate con‑
gestion at container terminal. Int J Prod Econ 141(1):179–188
Dijkstra E (1959) A note on two problems in connexion with graphs. Numer Math 1:269–271
Duan F (1994) A faster algorithm for shortest path-SPFA. J Southwest Jiao Tong 2:207–212
Giuliano G, O’Brien T (2007) Reducing port-related truck emissions: the terminal gate appointment sys‑
tem at the ports of Los Angeles and Long Beach. Transp Res Part D 12(7):460–473
Heilig L, Lalla-Ruiz E, Voß S (2017a) Multi-objective inter-terminal truck routing. Transp Res Part E
106:178–202
Heilig L, Lalla-Ruiz E, Voß S (2017b) Port-IO: an integrative mobile cloud platform for real-time inter-
terminal truck routing optimization. Flex Serv Manuf J 29(3–4):504–534
Horn RA, Johnson CR (2013) Matrix analysis. Cambridge University Press, New York

13
 S. Yi et al.

Kim HJ (2017) An operation method of truck appointment system utilizing estimated waiting time in
container terminals. Pusan National University, Thesis Presentation
Lalla-Ruiz E, Armas J, Expósito-Izquierdo C, Melián-Batista B, Moreno-Vega JM (2015) A multi-stage
approach aimed at optimizing the transshipment of containers in a maritime container terminal. In:
15th international conference on computer aided systems theory, LNCS, vol 9520. Springer Interna‑
tional Publishing, pp 255–262
Murty KG, Wan YW, Liu J, Tseng MM, Leung E, Lai KK, Chiu HWC (2005) Hongkong international
terminals gains elastic capacity using a data- intensive decision-support system. Interfaces 35:61–75
Nossack J, Pesch E (2013) A truck scheduling problem arising in intermodal container transportation.
Eur J Oper Res 230:666–680
Phan TMH, Kim KH (2015a) Negotiating truck arrival times among trucking companies. Transp Res Part
E 75:132–144
Phan TMH, Kim KH (2015b) Truck appointment system for transshipment containers in terminals. In:
Proceedings of the Asia Pacific industrial engineering and management systems conference 2015,
pp 1560–1567
Phan TMH, Kim KH (2016) Collaborative truck scheduling and appointments for trucking companies
and container terminals. Transp Res Part B 86:37–50
Pusan East Container Terminal Co Ltd (2004) Shinsundae terminal stevedoring tariff. Pusan East Con‑
tainer Terminal Co Ltd, Busan
Riaventin VN, Kim KH (2018) Scheduling appointments of truck arrivals at container terminals. Int J Ind
Eng 25(5):590–603
Schulte F, Lalla-Ruiz E, González-Ramírez RG, Voß S (2017) Reducing port-related empty truck emis‑
sions: a mathematical approach for truck appointments with collaboration. Transp Res Part E
105:195–212
Stanley R (2013) Algebraic combinatorics: walks, tree, tableaux, and more. Springer, New York
Steenken D, Voss S, Stahlbock R (2004) Container terminal operation and operations research—a clas‑
sification and literature review. OR Spectr 26:3–49
Wang WF, Yun WY (2013) Scheduling for inland container truck and train transportation. Int J Prod
Econ 143(2):349–356
Zehendner E, Feillet D (2014) Benefits of a truck appointment system on the service quality of inland
transport modes at a multimodal container terminal. Eur J Oper Res 235:461–469
Zhang R, Yun WY, Moon IK (2009) A reactive tabu search algorithm for the multi-depot container truck
transportation problem. Transp Res Part E 45:904–914
Zhang R, Yun WY, Kopfer H (2010) Heuristic-based truck scheduling for inland container transportation.
OR Spectr 32(3):787–808
Zhang X, Zeng Q, Chen W (2013) Optimization model for truck appointment in container terminals. Pro‑
cedia Soc Behav Sci 96:1938–1947
Zhao W, Goodchild AV (2010) The impact of truck arrival information on container terminal rehandling.
Transp Res Part E 46(3):327–343

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published
maps and institutional affiliations.

Sanghyuk Yi studied at the Department of Industrial Engineering, Pusan National University (Bachelor,
Master). He is a Ph.D. candidate at the International Graduate School (IGS) for Dynamics in Logistics,
Production Engineering, University of Bremen, Germany. His research is focused on the operational
problem of logistics, particularly container terminal.

Bernd Scholz‑Reiter studied industrial engineering and management at the Technical University of
Berlin. After completing his doctorate in computer science, he was a postdoctoral Fellow at the IBM
Thomas J. Watson Research Center in the USA and a research associate at the TU Berlin. In 1994 he was
appointed to the Chair of Industrial Information Technology at the Technical University of Cottbus. From
1998 to 2000 he was also head of the Fraunhofer Application Center for Logistics System Planning and
Information Systems, which he founded. In the year 2000 he was appointed to the Chair for the Planning
and Control of Production Systems in the Faculty of Production Engineering at the University of Bremen.
From 2002 to 2012, he also headed the Bremen Institute for Production and Logistics (BIBA). He is

13
Scheduling appointments for container truck arrivals…

a Fellow of the International Academy for Production Engineering (CIRP), member of the Scientific
Society for Production Engineering (WGP), the Scientific Society for Industrial Organization (HAB), the
German Academy of Engineering Sciences (acatech), and the Berlin-Brandenburg Academy of Sciences.
From 2007 to 2011 he was Vice President of the German Research Foundation. Since 2012 he has held
the office of President of the University of Bremen. He also serves as Vice President for International
Affairs of the German Rectors’ Conference since 2018.

Taehoon Kim studied at the Department of Computer Engineering, Pusan National University (Bache‑
lor). He is working at the Samsung SDS Co Ltd which is an Information Technology service provider.
His area of expertise is to develop graphical models and heuristic algorithms for optimization problems.

Kap Hwan Kim working at the Department of Industrial Engineering of Pusan National University. He
studied at the Seoul National University (Bachelor) and the Korea Advanced Institute of Science and
Technology (Master, Ph.D.). He was a visiting scholar at Purdue University, Montreal University, and
Shanghai Maritime University. He was the director of the Institute of Logistics Innovation and Network‑
ing at Pusan National University and the president of the Korean Institute of Industrial Engineers. His
research is focused on the design and operational problems of container terminals. He published many
papers at international journals such as Flexible Services and Manufacturing Journal, Transportation Sci‑
ence, Transportation Research B, Transportation Research E, European journal of Operational Research,
and so on.

13