You are on page 1of 21

Simulation Modelling Practice and Theory 106 (2021) 102146

Contents lists available at ScienceDirect

Simulation Modelling Practice and Theory


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

Simulation model and performance evaluation of battery-powered


T
AGV systems in automated container terminals
Ning Maa,c, Chenhao Zhoub,c, , Aloisius Stephenc

a
School of Public Policy and Administration, Xi’an Jiaotong University, Xi’an 710049, PR China
b
School of Management, Northwestern Polytechnical University, Xi’an 710072, China
c
Department of Industrial Systems Engineering and Management, National University of Singapore, 117576, Singapore

ARTICLE INFO ABSTRACT

Keywords: The use of battery power is becoming widespread rapidly among the mega ports worldwide,
Simulation model owing its low emission and high energy efficiency. In this paper, a simulation approach is pre­
System performance sented to configure the charging stations (CSs) and battery-powered automated guided vehicles
Battery-powered AGV (B-AGVs) at automated container terminals. The facility planning of the CSs and the operation
Charging stations
strategies of the B-AGVs are crucial for the system performance of the terminal. A port system is
modeled as a discrete event simulation model, mainly consisting of three parts: vessel generator,
dispatcher, and traffic network. Two types of layout designs and two types of recharging policies
are presented, to deploy the CSs and control the B-AGVs to be recharged, respectively. Extensive
simulation experiments are conducted to analyze the parameter effects on the system perfor­
mance. The results indicate that a decentralized CS layout and a progressive recharging policy
lead to excellent performance. Some management insights are offered to gain a better under­
standing of the key factors when deploying a B-AGV system.

1. Introduction

The continuous increase in fossil fuel consumption and greenhouse gas emissions worldwide are concerning the governments and
public. In port industries, diesel engines are widely utilized to power automated guided vehicles (AGVs) in container terminals. AGVs
are unmanned and controlled by a central system, which contributes to cost-saving and high-efficiency. However, they also produce
many pollutants and greenhouse gas emissions that diffuse across cities easily. Because of the environmental concerns, port operators
look forward to alternative reliable energy sources. Among current clean energies, Zhu et al. [26] pointed electric power has emerged
as one of the most practical and feasible alternative solutions, owing to its advantages of low emission, high energy efficiency, and
stable cost. In the long term, Schmidt et al. [18] noted that battery-powered AGVs (B-AGVs) are cheaper than diesel-powered AGVs
(D-AGVs) because the reduced maintenance and energy cost can offset the higher initial acquisition cost. Utilizing B-AGVs instead of
D-AGVs has attracted some interest from port industries.
With the rapid development and application of electric vehicles, the latest battery technology can support the carrying capability
and durability of B-AGVs in heavily loaded situations. Currently, B-AGVs deployment is spreading rapidly worldwide. For example,
Song and Ravesteijn [20] reported that the Shanghai Yangshan port in China has 130 B-AGVs in operation. The Long Beach Container
Terminal also ordered B-AGVs instead of diesel-powered trucks, which aims to promote clean, green maritime transportation ([6]).
Table 1 compares the differences of B-AGVs and D-AGVs. It can be seen that B-AGVs present numerous advantages in regard to the


Corresponding author.
E-mail address: zhou_chenhao@u.nus.edu (C. Zhou).

https://doi.org/10.1016/j.simpat.2020.102146
Received 25 February 2020; Received in revised form 13 June 2020; Accepted 30 June 2020
Available online 03 July 2020
1569-190X/ © 2020 Elsevier B.V. All rights reserved.
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Table 1
Comparison of a D-AGV and a B-AGV.
Item D-AGV B-AGV Superior

Powertrain Complex system Simple system B-AGV


Fueling cost Higher Lower B-AGV
Maintenance cost Higher Lower B-AGV
Noise Heavy Light B-AGV
Emissions NOx, Solid particulate Zero emission B-AGV
Purchase cost Lower Higher D-AGV
Operation range Longer Shorter D-AGV
Refueling time Shorter Longer D-AGV

powertrain, operational cost, and environmental issues. A D-AGV has a complex powertrain consisting of a starter generator, an
internal combustion engine, an AC/DC converter, a DC/AC converter, and an electric motor, whereas a B-AGV has a relatively
simpler powertrain with a battery, a DC/AC converter, and an electric motor. However, compared to D-AGVs, the current drawbacks
of B-AGVs are their limited range of operation and longer recharge time. Hence, appropriate facility planning and effective opera­
tional strategies for a B-AGV system are extremely importance when substituting D-AGVs at container terminals.
Unlike conventional transportation systems, the B-AGV system needs to deploy charging stations (CSs) to recharge the batteries.
While diesel vehicles only need a few minutes to refuel, the B-AGV may need 1 h to charge a flat battery to full ([11]). These regular
breaks in the work cycle entail increasing the size of vehicles to offset the recharging liabilities. Poor deployment of battery infra­
structure and operational strategies of charging the B-AGVs can cause major operational delays. Hence, proper facility planning and
effective operational strategies of the B-AGV system are of significant importance when substituting the diesel-powered AGVs in
container terminals. To maintain acceptable performance of the B-AGV system, the following key questions should be answered.

1. What is the minimum number of B-AGVs required to achieve an acceptable waiting time when visiting a CS?
2. What is the best configuration of CS when deploying them at the terminal?
3. What is the best recharging policy for charging B-AGVs, and how is it different from other policies?
4. What is the minimum ratio of B-AGVs to D-AGVs that achieves similar performance?

The deployment of a B-AGV system needs to address long-term decisions on the facility planning of CSs and short-term decisions
on recharging operation for B-AGVs. The facility design is a strategic decision, which comprises location and capacity planning of
infrastructure ([12]).Taner et al. [22] concluded that facility layouts directly affect the terminal performance through various si­
mulation experiments. Recently, the CS location problem has attracted increasing attention in many fields ([13]). Chung and Kwon
[4] investigated the multi-period facility planning of CSs deployed in expressways. Zhu et al. [27] presented a novel model for CS
location problem of plug-in electric vehicles. Liu and Song [15] proposed a robust optimization model for the wireless charging
infrastructure location planning problem for electric buses with uncertainties in energy consumption and travel time. Zhang et al.
[23] studied the facility planning of fast-charging stations, simultaneously considering the transportation and electrical power
networks. Numerical experiments were conducted to illustrate the two proposed planning methods. Bai et al. [2] considered the
design problem of the CS network in a city with low penetration rate of electric vehicles. They presented a cell-based model to decide
locations, capacity options, and service types for CSs.
Considering the recharging operation for B-AGVs, Hausler et al. [8] tackled the charging scheduling problem of electrical vehicles
by minimizing the waiting time of customers.Bayram et al. [3] studied a charging network that could accommodate customer classes
with different charging preferences. Sweda et al. [21] investigated the optimal recharging policy for an electric vehicle, minimizing
the total cost including all stopping, charging, and overcharging costs. Kabir and Suzuki [9] proposed a flexible recharging approach
to increase the total productive hours. Simulation results showed that the performance of the manufacturing system increased
significantly by adopting the proposed approach. Kabir and Suzuki [10] evaluated the influences of four routing heuristics for battery
AGVs on the system performance. Simulation results showed that the best productivity was achieved when the travel distance and
waiting time were minimized simultaneously at the CS.Seitaridis et al. [19] investigated the scheduling problem of charging electric
vehicles with multiple CSs. They proposed an agent-based simulation scheme for the CS to answer the requests of vehicles using the
integer linear programming approach. Fernandez et al. [7] presented a real-time forecasting application to provide recommendations
for electric vehicles reaching a charging station (CS), aiming to avoid the waiting time and reduce the charging time. The application
utilizes the Internet of Things (IoT) technology and can run on a low-cost test system.Ryck et al. [17] proposed a decentralized
method of charging AGVs in industrial systems, where an AGV can independently choose a CS and determine the charging time. A
general constrained optimization algorithm was presented and compared with the current industrial method.
Although many studies have focused on the development of facility planning of CSs and operation of battery vehicles, no efficient
work has evaluated the performance of the B-AGV system in the container terminals. In this study, we concentrate on the deployment
of a B-AGV system and evaluate system performance. Considering that the B-AGV system is one of the sub-systems in the container
terminal, the performance of the B-AGV system should be evaluated through integration into the whole system. A container terminal
is a large-scale complex system; hence, multiple components interact with each other. Moreover, port operators prefer to know the
long-term performance of the B-AGV system. Because it is difficult or impractical to mathematically formulate an answer to the above

2
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Fig. 1. Topological map of a perpendicular layout terminal.

questions, we present a simulation approach to evaluate of the performance of the B-AGVs system. This study makes several con­
tributions to the academic and industrial communities. First, it presents a flexible discrete event simulation model to capture the main
activities in container terminals including vessel arrivals; container loading/unloading; dispatching of quay cranes (QCs), yard cranes
(YCs), and B-AGVs; and consuming/recharging of B-AGVs. Second, it provides two location layouts of CSs and practical recharging
policies for B-AGVs, which can be easily implemented in the central control system of the terminals. Third, extensive simulation
experiments are conducted to analyze the performance differences of the proposed layout designs and recharging policies, which can
help port operators select the best choice for deploying the B-AGV system. Finally, it offers some management insights for the better
understanding of the key factors affecting the required capacity of CSs and for achieving the best performance of the entire terminal.
The remainder of this paper is organized as follows. Section 2 describes the simulation model for the container terminal with B-
AGVs. Section 3 presents the candidate deployment designs for CS and B-AGVs. Section 4 shows the analysis results to answer the
concerns of port operators. Section 5 concludes the study, offering some management insights for applying B-AGVs.

2. Simulation model

This section introduces a discrete event simulation model capable of capturing the main operations in the container terminal. The
proposed model combines the accurate modeling of the various equipment in the terminal the integration of operational strategies.
We first briefly present the operation process in the container terminal, and then we describe the detailed modules in the simulation
model.

2.1. Terminal operation process

The container terminal is a complex system comprising multiple types of equipment such as QCs, YCs, and AGVs. Fig. 1 shows the
topological map of one container terminal with a perpendicular layout. According to the functional partitioning, the terminal can be
divided into three main areas: wharf area, yard block, and land side. The vessels arrive at the terminal and are assigned to the berths
according to their sizes and residual wharf capacity. The information of loading/unloading containers of the vessels is sent to the
terminal, and the central system determines the schedules of processing the containers. The schedules are sent to the equipment
control system, which handles the QCs, YCs, and AGVs to perform the schedules. Considering that the internal and external traffics
are fully isolated in the automated container terminals, the operations of external trucks are ignored in this study.
Fig. 2 shows the operation processes in the container terminal, where CSs are added. The solid lines indicate the operation process
of container unloading from the vessel and restoring in the yard area. The direction of container loading to the vessel is the opposite
of container unloading, which is expressed by the dotted lines. The loading containers are first retrieved from yard blocks by YCs, and
then conveyed to berth areas by B-AGVs. Finally, they are lifted to the vessels by QCs. The unloading containers go through this
process in reverse. In addition, the B-AGVs are recharged according to the recharging rules.
In the terminal operation system, the key control rules include vessel assignment, container processing, B-AGV routing, and B-
AGV recharging. We use the first-in-first-out (FIFO) rule to assign berths to vessels. If no berth is available, the vessels wait outside
until the wharf has sufficient space. When processing containers, the system assigns the nearest idle B-AGV to the QCs and YCs. If
there is no idle B-AGV, the containers are postponed until an available B-AGV is found. We use the built-in methods to arrange the
routing of each B-AGV, which records passing time and status of each B-AGV. The statuses are used to calculate the battery state of
charge (SOC) of B-AGVs. The recharging approach will be presented in the following section.

3
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Fig. 2. Schematic of terminal operation process.

2.2. Modules of the model

We model the port system as a discrete event simulation model by referring to an open-source discrete event simulation fra­
mework-O2DES. Several publications have shown its advantages in reliability and flexibility, compared with typical simulation
software. Zhou et al. [24] investigated the traffic efficiency in an container terminal by modeling the traffic system as a network of
servers. Li et al. [14] presented a decision-making process that optimized the capacity planning of large-scale container terminals.
Zhou et al. [25] proposed a coordination strategy to reduce the congestion of vessels and vehicles, with a bi-objective simulation
optimization approach.
Fig. 3 shows the main structures of the simulation model consisting of three parts: vessel generator, dispatcher, and traffic
network. The vessel generator samples the arrivals of vessels, including arrival time, vessel type, and number of containers. The
dispatcher assigns QCs, berth area, and yard area to the arrived vessels; registers jobs for containers in the vessels; and dispatches B-
AGVs to process the container jobs. The traffic network comprises moving paths and topological working points. B-AGVs travel on
this network to finish the tasks. The working points denote the junctions where containers are exchanged among QCs, YCs, and B-
AGVs. The detailed functions of these modules will be described in the following sections.

2.2.1. Vessel generator


In this study, we consider multiple types of vessels, where the main parameters are listed in Table 2. The vessels are classified into
three types (k = 1, 2, 3) according to their physical sizes. The number of Twenty-feet Equivalent Units (TEUs) in vessels is randomly
generated from the range [Ml, Mu], where the mean TEUs of vessel of type k can be denoted as MkTEU . The percentage pk expresses the
percentage of generated vessels of type k among all vessels. For simplicity, we assume the ratio of twenty-feet and forty-feet con­
tainers is 1, the ratio of loading and discharging containers is 1, and the percentage of lifting two twenty-feet containers simulta­
neously is 50%.

Fig. 3. Simulation modeling.

4
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Table 2
Parameters of vessel types.
Type (k) Length [Ll, Lu](m) TEUs [Ml, Mu] Percentage p(%)

Small, 1 50–150 500–1000 50


Medium, 2 150–300 1000–4000 45
Large, 3 300–400 4000–9000 5

The annual arrival rates of these vessels are determined with respect to the vessel percentage and the given expected annual
throughput Eannual
TEU
. Let μ be the global annual arrival rate of vessels, and μk be the corresponding annual arrival rate of vessels of type
k, where µk = pk × µ . Hence, μ can be calculated as follows:
TEU
Eannual
µ= 3
k=1
pk × MkTEU (1)

In the simulation model, we assume that the events of vessel arrivals are independent, and the inter-arrival time of vessels follows
a gamma distribution. The average inter-arrival time of vessels of type k can be denoted as 365/(pk × μ). However, it is difficult to
obtain the parameters of shape α and rate β of the gamma distribution in practice. If the coefficient of variation of the inter-arrival
time of vessels is fixed as cv, the αk and βk of vessels of type k can be calculated as follows:
1 pk × µ
= , =
k
cv 2 k
365cv 2 (2)

Without loss of generality, when cv is equal to 0, the generator will return the mean value.

2.2.2. Dispatcher
The dispatcher is a central system assigning resources to the arrival vessels, such as berths, QCs, yard blocks, and B-AGVs. When
berthing vessels, the length of the vessel and safety distance from other vessels should be considered. The safety distances of vessels
are specified by their lengths. For example, if the vessel length is less than 100 m, the safety distance is set as 15 m. If the vessel length
is greater than 300 m, the safety distance is set as 50 m. Arrival vessels will be allocated to the available berths with the constraints of
safety distances. If sufficient berth space is not available, the vessels will wait outside. When there are many available spaces for the
vessels, they are assigned according to the meet-in-middle principle proposed by Côté and Iori [5]. This principle is proposed to solve
the cutting and packing problems, which also gives management insights on berth allocation problems.
The QCs are allocated to the vessel according to vessel length and the number of containers. For example, if the vessel length is less than
200 m, the dispatcher will assign 2 QCs when the number of containers is between 1000 and 3000; the dispatcher will assign 3 QCs when the
number is greater than 3000. If the vessel length is greater than 300 m, 5 QCs will be assigned when the number is between 3000 and 5000;
and 6 QCs will be assigned when the number is greater than 5000. The assigned QCs are also constrained by the available cranes in the
specific berth. The assigned QCs will evenly process the containers in the vessels. We assume discharging (to yard) and loading (to vessel)
activities are performed separately. The QCs will first finish all discharging containers and then start processing the loading containers.
We used a revised eighty–twenty principle to assign yard blocks to store or retrieve the corresponding containers, in which eighty
percent of containers are stored or retrieved from the nearest twenty percent of yard areas, while the other twenty percent of
containers are from the other eighty percent of yard areas. In practice, the containers loaded to the vessels are stored more dis­
persedly than those discharged from the vessels. Thus, when processing discharging containers, we choose the nearest twenty percent
yard blocks among all yard areas and then utilize the eighty–twenty principle to generate the destination points for the containers.
However, for loading containers, the whole yard area contributes to generating the destination points.
Each QC and YC has four working points. This simulation model neglects the exact positions of containers, while the container is
characterized by the starting point and destination. For discharging containers, the starting points are evenly generated from the
working points of assigned QCs, while the destinations are evenly generated from the working points of QCs in the assigned yard
blocks. The choices are exactly the opposite of those made for loading containers. All containers with the same working point of QCs
are added into a job list. The job lists in the same QCs will be processed individually.
The cranes are modeled as servers, where the processing time follows a negative exponential distribution. The expected servicing
time of a single container by QC and YC is 2.5 min and 3 min, respectively. In practice, the gantry of the QC only moves after finishing
all containers in one bay. Thus, the gantry moving time of the QC can be ignored. However, the gantry moving time of the YC should
be calculated, and the gantry traveling speed of the YC is set to 2 m/s.
When processing the container job, the dispatcher will check the full B-AGV list under the idle status, then assign the nearest B-
AGV to handle the corresponding container. A B-AGV is referred to as idle if it is neither handling containers nor being recharged. If a
B-AGV is assigned to this container, the B-AGV and QC/YC will start working simultaneously. When the B-AGV reaches a QC/YC, a
handshake activity is conducted to hand over the container. The handshake time is denoted by either the waiting time of the B-AGV
for the QC/YC or the waiting time of QC/YC for the vehicle. If the current container is a twenty-feet container, the dispatcher will
select another twenty-feet container in the same job list and assign it to the B-AGV. If there is no twenty-feet container, the B-AGV will
handle the previous single twenty-feet container. When the B-AGVs complete their operation and are not dispatched more jobs, they
will move to the nearest buffer areas among the quay and yard sides.

5
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

2.2.3. Traffic network


The traffic network in the container terminal is affected by many design factors, where the travel efficiency is also dynamically
affected by the nature of job sequences in the specific container terminal and number of deployed vehicles. In [24], we modeled the
traffic system as a network of servers that represents both paths and junctions, for which the service rates are dynamically adjusted
according to the vehicle densities on them. The traveling time of vehicles through the paths and junctions dynamically depends on
congestion. Mohammadi and Shirazi [16] utilized a simulation metamodel to achieve a high-degree flexible routing of AGVs in
flexible manufacturing systems. The results of the metamodel demonstrated that the velocity and acceleration/deceleration have
reasonable effects on the throughput. Therefore, in this study, we apply the previous work to model the traffic network for the
container terminal.
For each path, its configuration consists of several static parameters, such as the length from the entrance to the exit, vehicle
capacity, starting point, and ending point. The starting point denotes which path it connects from, and the ending point expresses
which path it connects to. For simplicity, the starting and ending points of paths are set as the working points of cranes processing
containers from or to the B-AGVs. The positions of CSs are set as the ending points of the paths nearby. The dynamic properties of any
path involve all indexes of vehicles currently traveling on the path, which consist of the traveling distances of vehicles on the path,
expected time of vehicles to complete traveling on the path, and vacancy of the path. The instantaneous traveling speed on the path is
dynamically determined by the vehicle density ρ, which is the number of vehicles per unit distance. The traveling speeds of vehicles
are constrained by the safety distance between two consecutive vehicles, which should be sufficient for vehicles decelerating to stop
completely. Let a be the deceleration of the vehicle, the relationship between traveling speed v and safety distance d is v = 2ad .
Moreover, given the vehicle length L and number of lanes N in the path, the average safety distance can be written as d = N / L.
Let vmax be the maximum traveling speed without limitation, and vi(ρ) be the instantaneous traveling speed on the path i. vi(ρ) can be
calculated as follows:

Ni
vi ( ) = min vmax , 2a ( L )+
(3)

where the symbol + denotes if the safety distance is negative, and it is set as 1 m for calculation (see [24]). The other parameters in
function (3) are set as follows: a is 1.55 m/s2, vmax is 4.5 m/s, and L is 14 m. The speed function is adopted only for multi-lane paths,
whereas for junctions, it is assumed that vehicles travel at half the maximum speed, throughout the longest Manhattan-distance in the
junction (width + length).
The static properties of the traffic system involve the full set of all paths, as well as the full set of all vehicles. In this study, the
routes of vehicles were generated by Dijkstra’s algorithm. The dynamic properties of the traffic system comprise of the states of all
paths and the vehicles that are departing to travel through the network. When the vehicle reaching the end point of the current path,
the dispatcher will check if the consecutive path has vacancy to allow any incoming vehicles to continue traveling; in the absence of
space for the incoming vehicle, the vehicle will wait until the state of the consecutive path changes. The travel to the next path will
commence when the path becomes vacant. Once a vehicle exits from a path, it either moves on to the next path or arrives at its
destination, and the path becomes vacant to accommodate a waiting incoming vehicle. This ensures that all vehicles move forward to
their respective destinations. When the end point of the consecutive path is the destination of the vehicle and is available, the vehicle
exits the traffic system.

3. Designs of the B-AGV system

3.1. Facility planning of CSs

Facility planning enables the determination of the number and location of the CSs deployed in terminals that can satisfy the
charging demands of the B-AGVs. The number of CSs is a significant factor affecting the system performance. It is important to build
an adequate number of CSs, to achieve acceptable waiting time when recharging the B-AGVs. However, the construction cost of
building CSs is considerable. Hence, the terminals should build a suitable number of CSs to fulfill the charging task of B-AGVs.
The location choices of CSs take topological structure and operational processes of terminals into consideration. Based on the real-
life conditions, CSs can be deployed around the yard blocks. The location design of CSs can either be centralized or decentralized, as
illustrated in Fig. 4. In the centralized layout, the CSs are present together in the top area of the yard blocks, whereas in the
decentralized layout, the CSs are uniformly distributed at the end points of the yard blocks.

3.2. Recharging policy of B-AGVs

The recharging policy determines as to when the vehicles should be recharged; this also significantly affects the system perfor­
mance. It is impractical to recharge the B-AGV when it is in operation; it can be recharged only happen after it finishes its task. Based
on the operational practices in the port, we propose two recharging policies in this study: conservative and progressive policies. Both
the policies can be easily implemented in the terminal operation system. The details of these policies are presented as follows.

3.2.1. Conservative policy


The conservative policy denotes that the B-AGVs are less frequently recharged to maintain more idle vehicles. The B-AGV will

6
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Fig. 4. Two types of CS layouts..

continuously work until its battery SOC reaches the safety threshold. Herein, we set the threshold at 15% as per the suggestion of the
B-AGV manufacturing company. When the B-AGV reaches the SOC threshold, it will be assigned the quickest CS for being fully
recharged. Fig. 5 illustrates the framework of the conservative recharging policy.
The quickest station is defined as the one with least expected time to finish recharging all B-AGVs in or waiting at the station. The
proposed approach has been verified based on its industry-level application and is also proven to be efficient by the work of [10]. The
proposed dispatching rule is a static approach, which does not take system dynamic information into account. Fig. 6 illustrates an
example about this. The B-AGV spends 10 min moving to Station1, and 15 min to Station2. However, Station1 needs 20 min to finish
current recharging task, and Station2 needs 5 min. Although the B-AGV is close to Station1, it is better that the B-AGV is assigned to
Station2 for recharging.

3.2.2. Progressive policy


Compared with the conservative policy, the progressive policy allows the B-AGV to be recharged more frequently when there is an
idle CS. We utilized a multiple controlling approach to determine when the B-AGV should be recharged. Before allocating a new job
to any idle B-AGV, the dispatcher checks the SOC of the B-AGV. If the remaining SOC is less than 50%, the dispatcher further checks if
the nearest CS from the B-AGV is idle. If yes, the dispatcher assigns the B-AGV to be recharged, or the B-AGV continues to fetch new
job. However, if the remaining SOC is less than 30%, the dispatcher will check whether there is an idle CS among the nearest 3 CSs. If
yes, the dispatcher either assigns the B-AGV to the quickest CS among the selected 3 CSs, or the B-AGV continues working. Fig. 7
illustrates the framework of the progressive recharging policy.
The progressive recharging policy takes advantage of the idle statuses of CSs, improving their utilization. Moreover, the policy can
reduce the probability of the B-AGV reaching the minimum SOC, which may decrease the waiting queue of CSs. However, recharging
more frequently will increase the total setup time of recharging the B-AGVs. The simulation experiments in the next section highlight
the differences between the two policies.

Fig. 5. Framework of conservative policy.

7
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Fig. 6. Illustration of the proposed example.

Fig. 7. Framework of progressive policy.

3.3. Analysis process

Considering that the number of all candidate decisions is finite, we enumerate every decision in the decision pools and run the
simulation model. Decisions comprise facility planning and operation strategies. Simulation results are obtained and key objectives
are analyzed to evaluate the decisions. Fig. 8 illustrates the framework of the optimization process.
The mega terminals require transportation vehicles with longer travel distances and larger fleet sizes to achieve higher
throughput. The key indicators port managers are interested in are listed in the following Table 3. A crucial indicator is QCR, QC rate,
which is the backbone indicator of terminal operational efficiency. The CS idle expresses the idle proportion of CSs, which can
implicitly demonstrate the utilization of CSs. The others denote proportions of different statuses of the B-AGV fleet.

4. Simulation results

In this section, we discuss the various experiments conducted to verify the proposed simulation model, designs of CSs, and the
recharging policies. The experiments are based on a perpendicular terminal, which can effectively decrease the occurrence of

8
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Fig. 8. Framework of analysis process.

Table 3
Key indicators in port operations.
Name Description

QCR Average number of containers moved per QC per hour


CS idle Average proportion of CS on idle status among the whole simulation period
B-AGV idle Average proportion of B-AGV on idle status among the whole simulation period
B-AGV working Average proportion of B-AGV on working status among the whole simulation period
B-AGV traveling Average proportion of B-AGV on traveling to CS status among the whole simulation period
B-AGV waiting Average proportion of B-AGV on waiting at CS status among the whole simulation period
B-AGV setup Average proportion of B-AGV on setup at CS status among the whole simulation period
B-AGV charging Average proportion of B-AGV on charging at CS status among the whole simulation period

deadlock. Section 4.1 presents the main parameters of terminal configuration and the B-AGV. A preliminary analysis is described in
Section 4.2. Capacities, layouts, recharging policies, and D-AGV are compared in Sections 4.3–4.6; this comparison addresses the four
concerns in Section 1 individually. Section 4.7 evaluates the influence of battery degradation on the system performance. Finally,
Section 4.8 offers a few management insights to port operators.

4.1. Parameter setting

The main parameters of the terminal configuration and B-AGV are described in Table 4. For the terminal configuration, the quay
length, number of QCs, yard blocks, and YCs indicate the size of the terminal. The expected annual throughput denotes the normal
operation capacity of the port. The working cycle time for loaded and unloaded states are separately set as 3 and 5 h. The recharging
time of the B-AGVs from empty to 100% SOC is set to 1 h. The entire charging/discharging profile presents a nonlinear relationship
with time. However, the manufacturers of B-AGVs typically block the nonlinear parts (extremely high or extremely low voltage) to
protect the batteries. Hence, the users can only access the linear parts of the batteries, for which the charging process is under a fast-
charge condition to decrease the charging time. Hence, we assume the charging/discharging rates are linear with time. The other
parameters can be referred to the work of [14].
We consider multiple scenarios to evaluate the system performances for different designs. The parameter settings in each scenario
are listed in the following Table 5. The first scenario S1 is the baseline scenario, which verifies the influences of CS capacity. Scenarios
S2 to S3 verify the influences of the higher number of B-AGVs and the coefficient of variation of vessel arrivals; scenarios S4 to S6,

Table 4
Parameters of terminal configuration and B-AGV.
Class Parameter Value

Terminal Quay length (meter) 2500


Number of QCs 40
Number of yard blocks 54
Number of YCs 108
Expected annual throughput (milion) 8
B-AGV Maximum B-AGV traveling speed (m/s) 4.5
Recharging setup time (min) 5
Recharging time from empty to 100% SOC (min) 60
Working cycle time on loaded from 100% to empty SOC (h) 3
Working cycle time on unloaded from 100% to empty SOC (h) 5

9
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Table 5
Parameter settings in different scenarios.
Scenario #B-AGV cv Layout of CS Recharging policy

S1 100 0.5 Decentralized Conservative


S2 120 0.5 Decentralized Conservative
S3 100 1.5 Decentralized Conservative
S4 100 0.5 Centralized Conservative
S5 120 0.5 Centralized Conservative
S6 100 1.5 Centralized Conservative
S7 100 0.5 Decentralized Progressive
S8 120 0.5 Decentralized Progressive
S9 100 1.5 Decentralized Progressive

different layouts of charging stations; scenarios S7 to S9, different recharging policies. Finally, we also compare the performances of B-
AGVs and D-AGVs, where the scenario with D-AGVs neglects the battery attribute of the vehicles.
Before presenting the experiment results, a warm-up analysis is required to estimate how long it takes for the simulation model to
reach a steady state. In the warm-up analysis, indicator results of scenario S1 are recorded every hour, as shown in Fig. 9. The number
of CSs is set to 40 to avoid the negative effects of CSs. To ensure that the results are consistent, the first five days (120 h) are treated as
a warm-up period and they are not included in the final indicator results. Following the warm-up period, the indicator values become
steady.
For each scenario, the simulation model is run ten times and the average result is reported. Considering that the warm-up period is
5 days, the simulation period is set as 60 days, which can generate steady results for further evaluations. Table 6 lists the descriptive
statistics of the indicator results of scenario S1. The column “SD” denotes the standard deviation of the results among the ten random
instances. The standard deviations of all indicators are extremely small, which indicates that the average results of the simulation
model with ten repetitions are fairly robust.
To further validate the simulation model, more indicators are recorded and listed in Table 7. The average wharf utilization and
berthing on arrival (BOA) rate are 70.98% and 9.13%, respectively. The average container job distance is 0.66 km. The discharging
and loading cycle times are 2.23 and 2.56 min, respectively; these denote the average times required for QCs for finishing a single
discharging activity and loading activity. All indicator have been verified with industrial experts, which are close to the experience
values in actual operations.

4.2. Preliminary analysis

If we simplify the whole terminal operation process, the B-AGVs arriving at the CSs for battery recharging could be seen as an M/M/
c queuing system. We assume that B-AGV arriving at CSs follows a Poisson process with λ per hour. The recharging time of each B-AGV
follows an exponential distribution with a μ service rate per hour. Let k be the number of CSs and n be the total number of B-AGVs.
According to the technology parameters provided in Table 4, on average, the B-AGV can continuously move for four hours (one
trip consisting of loaded and unloaded sub-trips) after full recharging. The CS can recharge one B-AGV per hour, which indicates that
each B-AGV spends one hour to finish recharging the battery. Considering idle and non-moving statuses do not contribute to battery
consumption, let α be the proportion of B-AGV on moving status throughout day. Then, λ is approximately equal to n . When
4/ +1
avoiding the infinite waiting number, the service rate should be greater than the arrival rate, namely, λ < ku. Therefore, the number
of CSs must be more than n
. If the number of B-AGVs is 100 and α is empirically set as 80%, the minimum number of CSs is
µ (4 / + 1)
approximately 17. Though this is not the accurate minimum capacity to serve the B-AGV fleet, it offers insights on the capacity design
of CSs.

Fig. 9. Plan view of the running process of the simulation model.

10
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Table 6
Descriptive statistics of indicators of scenario S1.
Name Min Max Mean SD

QCR 31.87 32.11 32.34 0.12


CS idle 44.73% 45.82% 46.43% 0.00
B-AGV idle 7.55% 9.17% 9.87% 0.01
B-AGV working 68.05% 68.62% 69.81% 0.01
B-AGV traveling 0.23% 0.25% 0.27% 0.00
B-AGV waiting 0.22% 0.30% 0.36% 0.00
B-AGV setup 1.19% 1.21% 1.23% 0.00
B-AGV charging 20.24% 20.46% 20.88% 0.00

Table 7
Several other indicators of scenario S1.
Name Min Max Mean SD

Wharf utilization (%) 68.35 69.21 70.98 0.01


BOA rate (%) 97.57 98.62 99.13 0.01
Avg. container job distance (km) 0.65 0.65 0.66 0.00
Avg. discharging cycle time (min) 2.21 2.22 2.23 0.01
Avg. loading cycle time (min) 2.52 2.53 2.56 0.02

4.3. Comparison between capacities of CSs

4.3.1. Baseline scenario (S1)


We introduce several scenarios to evaluate the performances of the different capacities of CSs. The parameters in the baseline
scenario are as follows: 100 B-AGVs, cv = 0.5, decentralized layout, and conservative recharging policy. Based on initial simulation
experiments under S1, we consider the capacity choices of CSs in the range of 18–27. Fig. 10 shows the results of S1; the horizontal
axis represents the number of CSs, and the left and right vertical axes denote the values of QCR and CS idle, respectively. It can be
seen that when the capacity is below 21, QCR is lower than the best performance, and the CS idle proportion is close to 0; this shows
that the CSs are insufficient and cannot provide adequate battery energy to power the B-AGV fleet. As the CS capacity increases, both
the QCR and CS idle proportion increase. When the number of CSs is greater than 24, QCR stabilizes around 32.2, indicating that the
CSs are not the resource limitation in the terminal.
Table 8 presents the results of the B-AGV statuses of S1. The first column denotes the number of CSs, and the other columns
represent the average proportion of B-AGV under the corresponding statuses. It can be seen that as the number of CSs increases, the B-
AGV idle and working proportions increase, and the B-AGV traveling and waiting proportions decrease; however, the B-AGV setup
and charging proportions remain almost unchanged. When the number of CSs is greater than 24, the B-AGV working and charging
proportions become constant. This demonstrates that given the scenario configuration, the B-AGVs consume similar amounts of
battery energy. The capacity of the CSs primarily influences the B-AGV traveling and waiting proportions. The terminal should
provide enough CSs to achieve an acceptable QCR and maintain a low B-AGV waiting proportion.

4.3.2. Scenario S1 with more B-AGVs (S2)


In this subsection, we analyze the performances of the CSs under the scenario with more B-AGVs. The number of B-AGVs is set to
120, while the other parameters are the same as in the baseline scenario, S1. Although the number of B-AGVs in S2 is greater than in
S1, the initial experiments suggest that the capacity choices of the CSs are also in the range of 18–27. The factor affecting the required

Fig. 10. Results of QCR and CS idle of S1 with varying capacity.

11
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Table 8
Results of B-AGV statuses of S1 when capacity varies.
#CS B-AGV idle B-AGV working B-AGV traveling B-AGV waiting B-AGV setup B-AGV charging

18 1.58% 56.65% 6.61% 17.39% 0.97% 16.79%


19 1.65% 60.32% 2.25% 17.04% 1.03% 17.71%
20 2.07% 62.77% 0.84% 14.62% 1.08% 18.63%
21 4.55% 65.16% 0.61% 9.21% 1.13% 19.34%
22 6.43% 66.19% 0.58% 5.89% 1.15% 19.76%
23 7.64% 67.04% 0.55% 3.45% 1.18% 20.14%
24 8.11% 67.66% 0.53% 2.00% 1.20% 20.50%
25 8.77% 67.89% 0.45% 1.35% 1.19% 20.34%
26 8.78% 68.08% 0.42% 1.02% 1.20% 20.50%
27 9.28% 68.12% 0.38% 0.66% 1.20% 20.37%

capacity of CSs will be discussed in the following section.


Fig. 11 shows a comparison of the QCR and CS idle proportions between S1 and S2. The detailed results can be seen in the
supplementary material. When there are more B-AGVs, the best performance of QCR is approximately 33.2, which is slightly better
than that in S1. The CS idle proportion of S2 is lower than that of S1. The presence of more B-AGVs contributes to the improved system
performance of the port terminal. The B-AGVs also improve the utilization of the CSs. However, the idle proportion of the B-AGVs
increases because the expected annual throughput is fixed.
The following Fig. 12 shows the B-AGV statuses under S1 and S2. When deploying more B-AGVs, the B-AGV idle proportion of S2 is
higher than that of S1. Conversely, the B-AGV working proportion of S2 is lower than that of S1. The presence of more B-AGVs will
increase the B-AGV traveling and waiting proportions. However, the B-AGV setup and charging proportions of S2 are lower than those
of S1. Fig. 12(b and f) imply that the total recharging capacity of batteries is related to the whole working time of the B-AGVs, and not
to the number of B-AGVs.

4.3.3. Scenario S1 under higher coefficient of variation of vessel arrivals (S3)


In S3, the coefficient of variation of the inter- arrival time of vessels, cv, is set to as 1.5, which is higher than that in S1. A higher
coefficient of variation indicates that the inter- arrival time of vessels is more variable. Fig. 13 shows the results of the QCR and CS
idle proportion under S1 and S3.
When cv increases, the best performance of QCR is around 29.5, which is worse than that in S1. The CS idle proportion of S3 is
higher than that of S1, indicating that the working time of CS or the total charging amount decreases. A higher coefficient of variation
of vessel arrivals will reduce the system performance of the port terminal, as well the utilization of B-AGVs and CSs. Thus, it will be
beneficial for the port terminal to smooth the inter- arrival time of vessels.
Fig. 14 presents the gaps of the B-AGV statuses of S3, compared with those of S1. It can be seen that when cv increases, the B-AGV
idle proportion increases sharply; moreover, the B-AGV working proportion decreases slightly, which implies that the total working
time of the B-AGVs decreases. The B-AGV charging proportion of S3 is also lower than that of S1, showing a trend similar to that of the
B-AGV working proportion. The differences between the B-AGV traveling and waiting proportions of S3 and S1 are negligible.

4.4. Comparison of layouts of CSs

4.4.1. Scenario S1 under centralized layout(S4)


In this subsection, we evaluate the performances of two CS layout designs. In S4, the CS layout is centralized in the terminal, and
the other parameters are the same as in S1. Fig. 15 shows the results of the QCR and CS idle proportion under S1 and S4. Compared
with the decentralized layout, the best performance of QCR is around 31.1, which is lower than that of S1. The CS idle proportion of
S3 is slightly higher than that of S1, indicating that the working time of the CSs decreases. The results show the decentralized CS
layout is better than the centralized layout.

Fig. 11. Comparison of QCR and CS idle of S1 and S2.

12
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Fig. 12. Comparison of B-AGV statuses of S1 and S2.

Fig. 13. Results of QCR and CS idle of S1 and S3.

Table 9 presents the gaps of the B-AGV statuses of scenario S4 compared with those of S1. When CSs are centralized, the B-AGV
idle proportion decreases slightly. The decreasing B-AGV working proportion implies that the B-AGV setup and charging proportions
also decrease. However, the B-AGV traveling and waiting proportions increase significantly, which shows that the B-AGVs spend
more time reaching the CSs. This has a further negative impact on the performance of the terminal.

4.4.2. Scenario S2 under centralized layout (S5)


S5 is performed for evaluating the performances of two CS layouts. The key difference between S5 and S2 is that the layout of the
CSs is centralized. Fig. 16 shows the results of the QCR and CS idle proportion under S2 and S5. The best performance of QCR is
approximately 32.5 in S5, which is lower than that in S2. The CS idle proportion of S5 is slightly higher than that of S2, indicating that
the CS working time decreases. These results show trends similar to those of S3 and S1.

13
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Fig. 14. Comparison between B-AGV statuses of S1 and S3.

Fig. 15. Results of QCR and CS idle proportion of S1 and S4.

Table 10 presents the gaps of B-AGV statuses of S5 compared with those of S2. Compared with S2, the B-AGV idle and working
proportions decrease, while the B-AGV traveling and waiting proportions increase sharply. The B-AGV setup and charging propor­
tions decrease as the working proportion decreases.

4.4.3. Scenario S3 under centralized layout (S6)


S6 is also for comparing the CS layout performance. The key difference between S6 and S3 is again the CS layout. Fig. 17 shows the
results of the QCR and CS idle proportion under scenarios S3 and S6. The best performance of QCR is approximately 28.6 in S6, which
is lower than that in S3. The CS idle proportion of S6 is slightly higher than that of S3, indicating that the working time of the CSs
decreases. These results show trends similar to those of the previous comparisons.

14
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Table 9
Comparison between gaps of B-AGV statuses of S4 and S1.
#CS B-AGV idle B-AGV working B-AGV traveling B-AGV waiting B-AGV setup B-AGV charging
18 −1.33% −1.29% 2.96% 3.32% −0.10% −0.10%
19 −1.53% −0.59% 4.24% 1.90% −0.17% −0.22%
20 −7.45% −0.27% 8.68% 3.16% −1.04% −1.06%
21 −5.32% −0.32% 9.95% 3.87% −1.93% 0.28%
22 −1.36% −0.55% 15.03% 6.17% −1.62% 0.12%
23 −2.18% −0.12% 13.49% 9.43% −3.65% −0.55%
24 0.21% −0.19% 9.26% 19.06% −4.47% −1.30%
25 −4.18% −0.15% 15.59% 21.83% −3.94% 0.75%
26 −0.43% −0.20% 15.39% 22.61% −4.15% −0.34%
27 −0.57% −0.18% 23.08% 29.07% −2.84% −0.33%
Average −2.41% −0.39% 11.77% 12.04% −2.39% −0.28%

Fig. 16. Results of QCR and CS idle of S2 and S5.

Table 10
Comparison between gaps of B-AGV statuses of S5 and S2.
#CS B-AGV idle B-AGV working B-AGV traveling B-AGV waiting B-AGV setup B-AGV charging

18 −29.69% −2.30% 5.32% 3.97% −0.14% −0.14%


19 −28.23% −3.24% 7.74% 5.78% −0.77% 0.13%
20 −29.29% −3.10% 10.39% 5.80% −0.48% −0.28%
21 −27.63% −0.16% 9.22% 4.13% −0.16% −0.10%
22 −9.85% −0.33% 11.57% 4.18% −0.12% −0.13%
23 −6.52% −0.65% 33.08% 8.50% −3.71% −0.05%
24 −3.90% −1.04% 36.03% 15.47% −2.38% −0.05%
25 −2.97% −0.70% 40.68% 18.98% −2.59% −0.76%
26 −2.00% −0.38% 36.49% 22.54% −2.46% −1.09%
27 −0.26% −0.85% 44.11% 37.80% −4.06% −1.66%
Average −14.03% −1.28% 23.46% 12.72% −1.69% −0.41%

Fig. 17. Results of QCR and CS idle of S3 and S6.

15
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Table 11
Comparison between gaps of B-AGV statuses of S6 and S3.
#CS B-AGV idle B-AGV working B-AGV traveling B-AGV waiting B-AGV setup B-AGV charging

18 −8.30% −1.33% 6.47% 3.35% −0.11% −0.11%


19 −8.10% −0.69% 12.34% 2.06% −0.19% −0.20%
20 −4.21% −0.39% 30.40% 3.26% −1.02% −1.03%
21 −2.45% −0.31% 20.58% 3.88% −1.86% −0.27%
22 −0.80% −0.55% 24.02% 6.09% −1.56% −0.11%
23 −2.59% −0.14% 47.02% 9.39% −3.65% −0.51%
24 −0.82% −0.22% 36.24% 18.92% −4.38% −1.25%
25 −1.20% −0.18% 48.48% 21.71% −3.89% −0.73%
26 −1.07% −0.21% 52.82% 22.61% −4.08% −0.32%
27 −0.87% −0.21% 43.25% 29.42% −2.89% −0.32%
Average −3.04% −0.42% 32.16% 12.07% −2.36% −0.48%

Table 11 presents the gaps of B-AGV statuses of S6 compared with those of S3. Compared with S3, the B-AGV traveling and waiting
proportions increase sharply. The other statuses exhibit similar trends.

4.5. Comparison of recharging policies

4.5.1. Scenario S1 under progressive recharging policy(S7)


We also evaluate the performances of the proposed recharging policy. In S7, we apply a progressive recharging policy to recharge
the B-AGVs, and the other parameters are the same as S1. Fig. 18 shows the results of QCR and CS idle proportions under S1 and S7.
When utilizing a progressive recharging policy, the best QCR performance is around 32.4, which is slightly better than that of S1. The
CS idle proportion of S7 is lower than that of S1, which indicates that the CS working time increases. The results show that the
progressive recharging policy is better than the conservative recharging policy.
The following Table 12 lists the gaps of B-AGV statuses of S7 compared with those of S1. When applying the progressive recharging
policy, the B-AGV traveling and waiting proportions decrease sharply, while the B-AGV setup proportion increases heavily. Although
B-AGVs are recharged more frequently, the reduced traveling and waiting times offset the setup time. Moreover, the B-AGV idle
proportion increases. The progressive policy finally benefits the B-AGV working proportions and further contributes to the system
performance.

4.5.2. Scenario S2 under progressive recharging policy (S8)


In S8, the B-AGVs are recharged following the progressive recharging policy, and the other parameters are the same as in scenario
S2. Fig. 19 shows the results of QCR and CS idle proportion under scenarios S2 and S8. When utilizing the progressive recharging
policy, the best performance of QCR is approximately 33.5, which is slightly better than that of S2. The CS idle proportion of S8 is also
lower than that of S2. The results show that the progressive recharging policy is better than the conservative recharging policy.
The following Table 13 presents the gaps of the B-AGV statuses of S8 compared with those of S2. Although the result gaps are not
as obvious as those of scenarios S7 and S1, they exhibit similar trends. This may be because the progressive recharging policy is
activated less frequently when there are higher numbers of B-AGVs. The B-AGV traveling and waiting proportions also decrease while
the B-AGV setup proportion increases. Moreover, the B-AGV idle and working proportions increase.

4.5.3. Scenario S3 under progressive recharging policy (S9)


In S9, the B-AGVs are also recharged following the progressive recharging policy, and the other parameters are the same as in
scenario S3. Fig. 20 depicts the results of QCR and CS idle proportion under S3 and S9. Similarly, when applying the progressive
recharging policy, the best performance of QCR is around 29.7, which is slightly better than that of S2. The CS idle proportion of S8 is

Fig. 18. Results of QCR and CS idle of S1 and S7.

16
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Table 12
Result gaps of B-AGV statuses of S7 with S1.
#CS B-AGV idle B-AGV working B-AGV traveling B-AGV waiting B-AGV setup B-AGV charging

18 2.0% 0.7% −5.1% −0.6% 0.5% 0.1%


19 −0.1% −0.4% 9.6% 0.0% 0.5% 0.0%
20 −7.0% −0.1% 10.3% 0.7% 0.4% 0.1%
21 6.8% 0.2% −0.3% −5.8% 1.7% 0.4%
22 3.2% 0.9% −5.2% −19.2% 5.1% 1.5%
23 11.3% 0.9% −14.9% −50.1% 8.2% 1.2%
24 12.2% 0.7% −33.5% −67.7% 12.1% 0.0%
25 4.3% 0.7% −37.8% −85.8% 17.8% 1.3%
26 4.7% 0.4% −46.9% −93.7% 21.5% 0.9%
27 1.1% 0.3% −49.0% −96.4% 24.7% 1.0%
Average 3.9% 0.4% −17.3% −41.9% 9.3% 0.7%

Fig. 19. Results of QCR and CS idle of S2 and S8.

Table 13
Result gaps of B-AGV statuses of S8 with S2.
#CS B-AGV idle B-AGV working B-AGV traveling B-AGV waiting B-AGV setup B-AGV charging

18 0.0% 0.5% −0.4% 0.0% 0.0% 0.0%


19 0.1% 0.0% −0.1% 0.0% 0.0% 0.0%
20 0.5% −0.1% −0.2% −0.2% 0.0% 0.0%
21 0.1% −0.4% 0.8% −0.5% 0.0% 0.0%
22 1.5% 0.2% −0.4% −1.3% 0.0% 0.0%
23 0.1% 0.1% 0.2% −0.6% 0.0% 0.2%
24 1.5% 0.8% −0.1% −2.5% 0.1% 0.2%
25 1.7% 0.3% −0.1% −2.2% 0.1% 0.2%
26 1.4% 0.5% −0.1% −2.0% 0.2% 0.1%
27 1.0% 0.2% −0.2% −1.4% 0.2% 0.1%
Average 0.8% 0.2% −0.1% −1.1% 0.1% 0.1%

Fig. 20. Results of QCR and CS idle of S3 and S9.

17
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Table 14
Result gaps of B-AGV statuses of S9 with S3.
#CS B-AGV idle B-AGV working B-AGV traveling B-AGV waiting B-AGV setup B-AGV charging

18 −4.8% −0.2% 15.0% −5.8% 1.8% 1.8%


19 −5.7% 0.1% 16.8% −2.8% 1.3% 0.8%
20 −9.4% −0.3% 22.8% 3.2% 1.8% 0.6%
21 3.8% −0.1% −0.8% −3.1% 2.8% 0.1%
22 8.1% −0.5% −9.0% −10.9% 7.4% 0.0%
23 6.6% 0.5% −13.2% −39.3% 8.0% 0.6%
24 13.9% −0.2% −31.4% −68.5% 10.9% −1.8%
25 5.5% 0.4% −37.7% −82.6% 16.3% −0.3%
26 2.9% 0.4% −46.5% −92.2% 22.0% 0.3%
27 2.0% 0.2% −51.8% −97.3% 26.3% 0.6%
Average 2.3% 0.0% −13.6% −39.9% 9.9% 0.3%

also lower than that of S2. These three results show that the progressive recharging policy is better than the conservative recharging
policy.
The following Table 14 presents the gaps of B-AGV statuses of S9 compared with those of S3. The results exhibit similar trends to
those of scenarios S7 and S1. The B-AGV traveling and waiting proportions decrease sharply, while the B-AGV setup proportion
increases heavily. The B-AGV setup also increases because the B-AGVs are recharged more frequently. However, the traveling and
waiting times decrease, which increases the B-AGV idle proportion. The progressive policy finally benefits the system performance.

4.6. Comparison of B-AGVs and D-AGVs

In this subsection, we compare the performances of B-AGVs and D-AGVs. We ignore the battery attribute of the vehicles in S1, and
the other parameters are all fixed. Fig. 21 shows the results of QCR under different numbers of B-AGVs and D-AGVs. For scenarios
with B-AGVs, the number of CSs is set to 30 to avoid their negative effects. It can be seen that the best performance of QCR is
approximately 33.5 under scenario with B-AGVs, which is almost equal to that under scenarios with D-AGVs. However, to achieve the
best performance, the minimum number of B-AGVs should be set to 120 while that for D-AGVs should be 90. If the expected number
of D-AGVs is 85 in this terminal, we can infer that 110 B-AGVs should be deployed when substituting all D-AGVs.
The following Table 15 shows the vehicle statues under a scenario with B-AGVs and D-AGVs. As the number of vehicles increases,
the vehicle idle proportion increases and the vehicle working proportion decreases for both scenarios. The D-AGV recharging pro­
portion is 0 because the vehicles do not recharge.
However, the results in Table 15 cannot aid evaluation of the ratio between the number of B-AGVs and the number of D-AGVs
while achieving the required performance. Let “#Working vehicle” be the number of expected working vehicles, which can be
informally calculated as follows:

#Working vehicle = #Vehicle × Prop . working (4)

In the following Table 16, the columns “#Working B-AGV” and “#Working D-AGV” denote the number of expected working B-AGVs
and D-AGVs under the scenarios with B-AGVs and D-AGVs, respectively. The number of expected working vehicles is approximately
71.2 for the scenario with B-AGVs, while it is 72.4 for the scenario with D-AGVs. The reason for the lower value for the scenario with
B-AGVs may be that the average unloaded traveling distance is lower owing to the higher number of B-AGVs. We can infer that for a
given configuration of one terminal port, the number of expected working vehicles is similar for these two scenarios. Thus, the
suggested ratio of the number of B-AGVs to the number of D-AGVs is approximately equal to the ratio of the number of expected
working D-AGVs to the number of expected working B-AGVs.

Fig. 21. Results of QCR under different number of B-AGVs and D-AGVs.

18
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

Table 15
Vehicle statuses of scenarios with B-AGVs and D-AGVs.
#B-AGV B-AGV idle B-AGV working B-AGV recharging #D-AGV D-AGV idle D-AGV working D-AGV recharging

60 1.12% 74.49% 22.19% 60 0.73% 99.27% 0.00%


70 1.22% 74.86% 21.75% 65 1.44% 98.56% 0.00%
80 1.37% 74.73% 21.73% 70 4.49% 95.51% 0.00%
90 3.89% 72.39% 21.56% 75 8.13% 91.87% 0.00%
100 9.18% 67.97% 20.51% 80 12.23% 87.77% 0.00%
110 15.33% 63.32% 19.11% 85 16.09% 83.91% 0.00%
120 21.34% 58.85% 17.76% 90 20.22% 79.78% 0.00%
130 25.34% 54.85% 17.76% 95 23.98% 76.02% 0.00%
140 31.27% 50.79% 15.44% 100 27.71% 72.29% 0.00%
150 35.83% 47.49% 14.44% 110 34.15% 65.85% 0.00%

Table 16
#Working vehicle of scenarios with B-AGVs and D-AGVs.
#B-AGV B-AGV working # Working B-AGV #D-AGV D-AGV working # Working D-AGV

60 74.49% 44.69 60 99.27% 59.56


70 74.86% 52.40 65 98.56% 64.06
80 74.73% 59.79 70 95.51% 66.86
90 72.39% 65.15 75 91.87% 68.91
100 67.97% 67.97 80 87.77% 70.22
110 63.32% 69.65 85 83.91% 71.33
120 58.85% 70.62 90 79.78% 71.81
130 54.85% 71.31 95 76.02% 72.21
140 50.79% 71.11 100 72.29% 72.29
150 47.49% 71.23 110 65.85% 72.43

4.7. Effect of battery degradation on the system performance

In this subsection, we evaluate the influence of battery degradation on the system performance. Numerous factors affect battery
cycle life. The following operations will shorten the lithium battery life: an extremely high or extremely low voltage, overcharge, and
deep discharge. Manufacturers of B-AGVs are attempting many approaches to increase the battery life substantially. For example,
extending the battery capacity results in a lower average depth of discharge, preventing the users to access the maximum charge part
to avoid the maximum voltage, and controlling the battery pack thermally to avoid both maximum voltage and high temperatures.
Moreover, the battery pack of B-AGVs is designed for industrial usage, which consists of numerous battery cells. These cells can be
easily replaced to maintain the life cycle of the entire battery pack. Anseán et al. [1] tested the influence of usage time on battery
capacity degradation, and found that after 1000 times of fully charging and discharging, the battery still maintains 95% capacity.
Therefore, over a relatively long time (such as one year), the battery working range can be seen to be without change.
Nevertheless, battery degradation still occurs to some extent, which mainly affects the working cycle time of the B-AGV. In this
experiment, we change the working cycle time and recharging rate of the B-AGVs to 90% of previous settings. The other parameters
are all fixed as in S1, and the number of CSs is set to 30 to avoid their negative effects. Table 17 shows the results of QCR and vehicle
statuses under scenarios with full and 90% battery capacities. It can be seen that QCR is nearly same under these two scenarios. The

Table 17
Results of scenarios with full and 90% battery capacities.
#B-AGV Full battery capacity 90% battery capacity

QCR B-AGV idle B-AGV B-AGV B-AGV QCR B-AGV idle B-AGV B-AGV B-AGV
working setup recharging working setup recharging

60 17.81 1.12% 74.49% 1.30% 22.19% 18.06 1.12% 74.52% 1.43% 21.97%
70 20.57 1.22% 74.86% 1.27% 21.75% 20.91 1.45% 74.47% 1.41% 21.70%
80 23.57 1.37% 74.73% 1.27% 21.73% 23.39 1.38% 74.59% 1.41% 21.69%
90 29.93 3.89% 72.39% 1.26% 21.56% 29.40 3.74% 72.22% 1.40% 21.54%
100 32.27 9.18% 67.97% 1.20% 20.51% 32.24 9.26% 68.02% 1.33% 20.37%
110 33.08 15.33% 63.32% 1.12% 19.11% 33.12 15.16% 63.20% 1.25% 19.10%
120 33.38 21.34% 58.85% 1.05% 17.76% 33.37 21.29% 58.89% 1.16% 17.59%
130 33.38 25.34% 54.85% 1.05% 17.76% 33.38 26.57% 54.76% 1.09% 16.53%
140 33.46 31.27% 50.79% 0.91% 15.44% 33.48 31.23% 50.74% 1.00% 15.25%
150 33.48 35.83% 47.49% 0.85% 14.44% 33.48 35.56% 47.55% 0.95% 14.43%
Average 29.09 14.59% 63.97% 1.13% 19.22% 29.08 14.68% 63.90% 1.24% 19.02%

19
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

average proportions of B-AGV on different statuses are also similar under the two scenarios. However, the B-AGV setup under 90%
battery capacity increases 10% compared with that under full battery capacity, from 1.13% to 1.24%.
The results show that if the working cycle time decreases to 90% of a new B-AGV, and the recharging rate also decreases to a same
proportion, there will be no significant impact to the system performance. Because the proportional relationship of working and
recharging for the B-AGV stays the same. Hence, the average proportions of B-AGV on different statuses will also remain unchanged,
such as B-AGV idle, B-AGV working, and B-AGV charging. The proportion of B-AGV setup will slightly increase if the setup time per
recharge does not decease to 90%. However, if the reduction of the recharging time is not linear to the battery degradation, results
will be different. The key factor affecting the proportions of B-AGV on different statuses is the ratio of working cycle time to
recharging time. We can input the new settings into the simulation model and obtain proper results.

4.8. Management insights

The simulation results presented in previous sections indicated that the decentralized CS layout and progressive recharging policy
could improve performances. Although higher numbers of B-AGVs contribute to the system performance, the idle proportion of the B-
AGV fleet increases and offsets the benefit. Moreover, the port should smoothen the inter- arrival time of vessels to improve the
system performance.
It can be seen that the required minimum number of CSs to achieve best system performance is related to the working time of B-
AGVs, other than the number of B-AGVs. Although the B-AGVs and D-AGVs are quite different, the working time of the B-AGV fleet is
also approximately equal to that of the D-AGV fleet. Port operators should coordinate the equipment in the terminal elaborately to
minimize the total working time of B-AGVs, which could further optimize system performance.

5. Conclusions

This study considers how to deploy B-AGVs in automated container terminals; it consists of long-term decisions on the facility
planning of CSs and short-term decisions on recharging operations for B-AGVs. Port operators are concerned with the performance of
B-AGVs when substituting diesel-powered AGVs in container terminals. This study proposes a flexible discrete event simulation
model to describe the container terminal with B-AGV system. It presents two CS layout designs and practicable recharging policies for
B-AGVs. It also offers some management insights to better understand how to achieve the best performance of terminals. This study
helps port managers to deploy the B-AGV system in terminals and evaluate the performances of different designs and operational
strategies. Future work can focus on embedding optimization algorithms for scheduling the B-AGV activities (job fetching, battery
recharging) into the simulation model.

Acknowledgments

This research is supported by the National Natural Science Foundation of China (Grant No. 71573204 and No. 71390333) and
Singapore Maritime Institute (Grant No. SMI-2017-SP-002). The authors are grateful to the anonymous reviewers whose constructive
comments make an improvement on the paper.

Supplementary material

Supplementary material associated with this article can be found, in the online version, at 10.1016/j.simpat.2020.102146.

References

[1] D. Anseán, M. González, J. Viera, V. García, C. Blanco, M. Valledor, Fast charging technique for high power lithium iron phosphate batteries: a cycle life analysis,
J. Power Sources 239 (2013) 9–15.
[2] X. Bai, K.-S. Chin, Z. Zhou, A bi-objective model for location planning of electric vehicle charging stations with GPS trajectory data, Comput. Ind. Eng. 128 (2019)
591–604.
[3] I.S. Bayram, A. Tajer, M. Abdallah, K. Qaraqe, Capacity planning frameworks for electric vehicle charging stations with multiclass customers, IEEE Trans. Smart
Grid 6 (4) (2015) 1934–1943.
[4] S.H. Chung, C. Kwon, Multi-period planning for electric car charging station locations: a case of korean expressways, Eur. J. Oper. Res. 242 (2) (2015) 677–687.
[5] J.-F. Côté, M. Iori, The meet-in-the-middle principle for cutting and packing problems, INFORMS J. Comput. 30 (4) (2018) 646–661.
[6] B.R. Farrell, R. McKie, Designing a battery exchange building for automated guided vehicles, Ports 2016: Port Planning and Development, (2016), pp. 71–80.
[7] G. Fernandez, V. Krishnasamy, J.S.M. Ali, Z.M. Ali, S.H.E.A. Aleem, Internet of things based real-time electric vehicle load forecasting and charging station
recommendation, ISA Trans. 97 (2020) 431–447.
[8] F. Hausler, E. Crisostomi, A. Schlote, I. Radusch, R. Shorten, Stochastic park-and-charge balancing for fully electric and plug-in hybrid vehicles, IEEE Trans.
Intell. Transp. Syst. 15 (2) (2014) 895–901.
[9] Q.S. Kabir, Y. Suzuki, Increasing manufacturing flexibility through battery management of automated guided vehicles, Comput. Ind. Eng. 117 (2018) 225–236.
[10] Q.S. Kabir, Y. Suzuki, Comparative analysis of different routing heuristics for the battery management of automated guided vehicles, Int. J. Prod. Res. 57 (2)
(2019) 624–641.
[11] T. Kawakami, S. Takata, Battery life cycle management for automatic guided vehicle systems, Design for Innovative Value Towards a Sustainable Society,
Springer, 2012, pp. 403–408.
[12] A. Klose, A. Drexl, Facility location models for distribution system design, Eur. J. Oper. Res. 162 (1) (2005) 4–29.
[13] A.Y. Lam, Y.-W. Leung, X. Chu, Electric vehicle charging station placement: formulation, complexity, and solutions, IEEE Trans. Smart Grid 5 (6) (2014)
2846–2856.
[14] H. Li, C. Zhou, B.K. Lee, L.H. Lee, E.P. Chew, R.S.M. Goh, Capacity planning for mega container terminals with multi-objective and multi-fidelity simulation

20
N. Ma, et al. Simulation Modelling Practice and Theory 106 (2021) 102146

optimization, IISE Trans. 49 (9) (2017) 849–862.


[15] Z. Liu, Z. Song, Robust planning of dynamic wireless charging infrastructure for battery electric buses, Transp. Res. Part C 83 (2017) 77–103.
[16] E.K. Mohammadi, B. Shirazi, Toward high degree flexible routing in collision-free FMSS through automated guided vehiclesâ;; dynamic strategy: a simulation
metamodel, ISA Trans. 96 (2020) 228–244.
[17] M.D. Ryck, M. Versteyhe, K. Shariatmadar, Resource management in decentralized industrial automated guided vehicle systems, J. Manuf. Syst. 54 (2020)
204–214.
[18] J. Schmidt, C. Meyer-Barlag, M. Eisel, L.M. Kolbe, H.-J. Appelrath, Using battery-electric AGVS in container terminalsâ;;assessing the potential and optimizing
the economic viability, Res. Transp. Bus. Manag. 17 (2015) 99–111.
[19] A. Seitaridis, E.S. Rigas, N. Bassiliades, S.D. Ramchurn, An agent-based negotiation scheme for the distribution of electric vehicles across a set of charging
stations, Simul. Modell. Pract. Theory 100 (2020) 102040.
[20] L. Song, W. Ravesteijn, Responsible port innovation in china: the case of the yangshan port extension project, Int. J. Crit. Infrastruct. 11 (4) (2015) 297–315.
[21] T.M. Sweda, I.S. Dolinskaya, D. Klabjan, Optimal recharging policies for electric vehicles, Transp. Sci. 51 (2) (2017) 457–479.
[22] M.E. Taner, O. Kulak, M.U. Koyuncuoğlu, Layout analysis affecting strategic decisions in artificial container terminals, Comput. Ind. Eng. 75 (2014) 1–12.
[23] H. Zhang, S.J. Moura, Z. Hu, Y. Song, Pev fast-charging station siting and sizing on coupled transportation and power networks, IEEE Trans. Smart Grid 9 (4)
(2018) 2595–2605.
[24] C. Zhou, L.H. Lee, E.P. Chew, H. Li, A modularized simulation for traffic network in container terminals via network of servers with dynamic rates, 2017 Winter
Simulation Conference (WSC), IEEE, 2017, pp. 3150–3161.
[25] C. Zhou, H. Li, B.K. Lee, Z. Qiu, A simulation-based vessel-truck coordination strategy for lighterage terminals, Transp. Res. Part C-Emerg.Technol. 95 (2018)
149–164.
[26] Z. Zhu, Z. Gao, J. Zheng, H. Du, Charging station planning for plug-in electric vehicles, J. Syst. Sci. Syst. Eng. 27 (1) (2018) 24–45.
[27] Z.-H. Zhu, Z.-Y. Gao, J.-F. Zheng, H.-M. Du, Charging station location problem of plug-in electric vehicles, J. Transp. Geogr. 52 (2016) 11–22.

21

You might also like