ArticlePDF Available

Artificial Intelligence-Empowered Edge of Vehicles: Architecture, Enabling Technologies, and Applications

Authors:

Abstract and Figures

With the proliferation of mobile devices and a wealth of rich application services, the Internet of vehicles (IoV) has struggled to handle computationally intensive and delay-sensitive computing tasks. To substantially reduce the latency and the energy consumption, application work is offloaded from a mobile device to a remote cloud or a nearby mobile edge cloud for processing. Compared with remote clouds, mobile edge clouds are located at the edge of the network. Therefore, mobile edge computing (MEC) has the advantages of effectively utilizing idle computing and storage resources at the edge of the network and reducing the network transmission delay. In addition, mobile devices are increasingly moving toward intelligence. To satisfy the service experience and service quality requirements of mobile users, the vehicle Internet is transforming into the intelligent vehicle Internet. Artificial intelligence (AI) technology can adapt to rapidly changing dynamic environments to provide multiple task requirements for resource allocation, computational task scheduling, and vehicle trajectory prediction. On this basis, combined with MEC technology and AI technology, computing and storage resources are placed on the edge of the network to provide real-time data processing while providing more efficient and intelligent services. This article introduces IoV from three aspects, namely, MEC, AI and the advantages of combining the two, and analyzes the corresponding architecture and implementation technology. The application of MEC and AI in IoV is analyzed and compared with current approaches. Finally, several promising future directions in the field of IoV are discussed.
Content may be subject to copyright.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.DOI
Artificial Intelligence-Empowered Edge
of Vehicles: Architecture, Enabling
Technologies, and Applications
HONGJING JI12 , OSAMA ALFARRAJ3, AND AMR TOLBA.34
1School of Software, Dalian University of Technology, 116620, Dalian, China
2School of Software, Taiyuan University of Technology, 030024, Taiyuan, China
3Computer Science Department, Community College, King Saud University, Riyadh 11437, Saudi Arabia
4Mathematics and Computer Science Department, Faculty of Science, Menoufia University, Shebin-El-kom 32511, Egypt
Corresponding author: Osama Alfarraj (e-mail: oalfarraj@ksu.edu.sa).
This work was funded by the Researchers Supporting Project No. (RSP-2019/102) King Saud University, Riyadh, Saudi Arabia.
ABSTRACT With the proliferation of mobile devices and a wealth of rich application services, the Internet
of vehicles (IoV) has struggled to handle computationally intensive and delay-sensitive computing tasks.
To substantially reduce the latency and the energy consumption, application work is offloaded from a
mobile device to a remote cloud or a nearby mobile edge cloud for processing. Compared with remote
clouds, mobile edge clouds are located at the edge of the network. Therefore, mobile edge computing
(MEC) has the advantages of effectively utilizing idle computing and storage resources at the edge of the
network and reducing the network transmission delay. In addition, mobile devices are increasingly moving
toward intelligence. To satisfy the service experience and service quality requirements of mobile users, the
vehicle Internet is transforming into the intelligent vehicle Internet. Artificial intelligence (AI) technology
can adapt to rapidly changing dynamic environments to provide multiple task requirements for resource
allocation, computational task scheduling, and vehicle trajectory prediction. On this basis, combined with
MEC technology and AI technology, computing and storage resources are placed on the edge of the network
to provide real-time data processing while providing more efficient and intelligent services. This article
introduces IoV from three aspects, namely, MEC, AI and the advantages of combining the two, and analyzes
the corresponding architecture and implementation technology. The application of MEC and AI in IoV is
analyzed and compared with current approaches. Finally, several promising future directions in the field of
IoV are discussed.
INDEX TERMS
Internet of Vehicles (IoV), Mobile Edge Computing (MEC), Artificial Intelligence (AI)
I. INTRODUCTION
WITH the rapid development of Internet of things (IoT)
technology and the increasing number of vehicle
networks, the traditional vehicle ad hoc network (VANET)
is gradually being integrated into the Internet of vehicles
(IoV). IoV is a new model that combines VANETs and
vehicle remote information processing to connect vehicles,
people and things [1]. In addition, it is a highly important
field in intelligent transportation systems (ITSs), as it covers
intelligent transportation, cloud computing, vehicle informa-
tion services, logistics transportation services [2] [3], modern
wireless technology, Internet access and communication and
other technologies and applications [4]. According to the
forecast report from Cisco, the global monthly mobile data
usage in 2021 will be approximately 49 exabytes, and the
number of mobile devices will be 11.6 billion, increasing
about approximately seven times between 2016 and 2021
[5]. With the explosion of mobile data, mobile phones are
increasingly being used for various computation-intensive
applications, such as augmented reality; natural language
processing; face, hand gestures, and object recognition; and
various forms of user configurations used for recommenda-
tion [6]; hence, mobile users enjoy a rich experience in the
service network. Faced with the surge of mobile data flow,
reducing the delay of data transmission between vehicles and
improving the throughput of data transmission between vehi-
VOLUME 4, 2016 1
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
cles are urgent problems [7]. Therefore, the vehicle network
must adopt advanced communication technology and data
acquisition technology to improve the safety and efficiency
of the traffic system, reduce accidents and reduce traffic con-
gestion [8]. Generally speaking, public communication inter-
faces are divided into wireless networks (such as bluetooth
and wi-fi) and cellular networks (such as 3G, 4G and 5G)
[9]. However, the limited network bandwidth in traditional
cellular networks limits the fast growth of the data trans-
mission rate. In the emerging 5G network, the application of
D2D (device to device) communication technology promises
to substantially improve the spectrum efficiency to support
data transmission between caching vehicles and mobile users
[5]. The federal communications commission (FCC) autho-
rized the 75-mhz band for the provision of vehicle-to-vehicle
wireless communications as dedicated short-range commu-
nications (DSRC). In addition, IEEE standardizes the entire
communication stack according to IEEE 802.11p as a wire-
less access to the vehicle environment (WAVE) to support the
interconnection between vehicles and between vehicles and
roads [10]. In addition, various communication modes coex-
ist in IoV, which include vehicle-to-vehicle (V2V), vehicle-
to-infrastructure (V2I), vehicle-to-sensor (V2S), vehicle-to-
pedestrian (V2P), and vehicle-to-network (V2N) [11] com-
munications, which form a dynamic mobile communication
system. Fig. 1 illustrates the architecture of IoV. This enables
the sharing and collection of information about vehicles,
roads and their surroundings. While the development of
communication technology can alleviate a certain amount of
traffic congestion, the limited ability of the infrastructure to
communicate, compute, and store resources can lead to long
delays and massive data transmission problems. In order to
overcome this challenge, combined with the deployment of
resources on the edge of the wireless network, the proposed
vehicle edge network has attracted wide attention.
Mobile edge computing (MEC) technology can overcome
the challenges of traditional mobile cloud computing (MCC).
For example, (1) centralized cloud servers are located far
away from the terminal devices, thereby resulting in low
efficiency in computation-intensive environments; (2) the
offloading of computing to the cloud consumes energy,
thereby reducing the service life of mobile batteries; and
(3) providing mobile users with complex memory-utilization
applications and higher data storage capacity is difficult [12].
Reference [13] studied the multi-user computing offload-
ing problem of mobile edge cloud computing under multi-
channel wireless interference, and put forward a distributed
computing offloading algorithm by using the game theory
method. In addition, MEC can provide substantial value
to mobile operators, service providers and end users. The
application scenarios of MEC span multiple fields, such as
augmented reality, online games, big data analysis and health
monitoring in the medical Internet of things [14] [15].
With the emergence of IoV and vehicle intelligence, ve-
hicles are transforming from transport tools to intelligent
terminals [16]. In addition, the variety and quantity of on-
FIGURE 1: The architecture of IoV.
board equipment are increasing, and people’s requirements
for automobile service quality are constantly increasing [17].
In the age of IoV, vehicle-mounted intelligent modules can
provide intelligent vehicle control, traffic management, ac-
cident prevention and navigation capabilities, along with
rich multimedia and mobile Internet application services and
many emerging interactive applications [12] that improve the
user experience, reduce operating costs and promote a safe
driving environment. Artificial intelligence (AI) can substan-
tially improve the cognitive performance and intelligence of
vehicle networks, thereby contributing to the optimal alloca-
tion of resources for problems with diverse, time-varying and
complex characteristics [18]. Reinforcement learning (RL)
is an important branch of machine learning. It refers to the
process of realizing objectives via multiple steps and suitable
decisions in a series of scenarios, which can be regarded as a
multi-step sequential decision problem [8]. To overcome the
problem of decentralized management of connected vehicles
in a distributed intelligent transportation system, reference
[19] designed an ant colony optimization algorithm that
is based on swarm intelligence (SI) for dynamic decision-
making of networked vehicles, which enables vehicles to
automatically and adaptively identify the best path to the des-
tination. In [20], an intelligent resource management strat-
egy for joint communication mode selection, resource block
allocation and power control in D2D-V2V communication
vehicle networks is proposed. The model-free participant
critical learning framework is used to effectively improve the
learning efficiency and identify the optimal strategy to ensure
that the vehicle-to-vehicle link satisfies the communication
requirements of ultra-reliability and low latency while maxi-
mizing the overall network capacity.
The main contributions of this article are as follows:
2VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
We introduce the architecture of IoV, the deployment
of V2X in vehicle-mounted communication, and the
application of MEC and AI in IoV. We describe the
advantages and development history of MEC and the
relationship between AI and DRL, and We analyze the
previous research on the application of AI to vehicle
edge networks.
We study the architecture of MEC-based IoV and dis-
cuss the use of MEC in IoV. In addition, the character-
istics of MEC, FC and MCC are analyzed, and the key
technologies for supporting MEC are introduced. In ad-
dition, the previous studies on efficient MEC calculation
for IoV are analyzed.
We consider the theoretical characteristics of AI; ana-
lyze DRL, which is a key method for realizing AI, and
demonstrate the architecture of AI in IoV. In addition,
We introduce the effective key AI algorithms for calcu-
lating the offload and resource allocation in IoV, and We
analyze the previous AI research on IoV.
We combine the application of AI and MEC technology
in IoV and analyze the key technologies that support the
application of AI in vehicle edge networks. In addition,
the previous studies on edge caches and on joint com-
puting resources and caches are introduced. Finally, the
future development directions and research challenges
of IoV are discussed.
This article reviews the architecture, implementation tech-
nology and application of IoV that is based on AI and
MEC. We explore the characteristics of IoV development, the
communication mode, and the impact of combining AI and
MEC technology on the construction of intelligent IoV. This
article is divided into the following parts: Section II describes
the architecture of MEC in IoV and introduces the devel-
opment history of edge computing and the characteristics
of MEC. In addition, the research on computing offloading
of MEC in IoV is analyzed. Section III mainly studies the
application of AI In IoV, expounds on the architecture in
which AI is combined with IoV, and discusses DRL, which is
an important technique for realizing AI. The key algorithms
and applications of AI in IoV are analyzed. In Section IV, the
significance of the combination of AI and MEC technology
in IoV is discussed, the key technologies in AI-based vehicle
edge networks are studied, and relevant studies on IoV are
analyzed. In Section V, the challenges that are faced by IoV
and the future development directions are discussed. Finally,
Section VI summarizes the study.
II. MOBILE EDGE COMPUTING IN INTERNET OF
VEHICLES
With the proliferation of mobile devices in the IoV, there
are stringent computing and processing requirements for
computation-intensive applications and delay-sensitive appli-
cations. The combination of IoV and MEC has emerged as
a promising approach for addressing the growing demand
for computing by shifting heavy computing tasks to cloud
resources on the edges of mobile networks. In this part,
we describe the development history of MEC and the MEC
architecture in IoV, and we explain the advantages of MEC
in IoV. Then, the key MEC technology of IoV is introduced.
Finally, the research status of MEC-based computational
offloading in IoV is discussed.
A. ARCHITECTURE
With the continuous improvement of the number and intel-
ligence of mobile devices, increasingly many mobile appli-
cations require many computing tasks. However, due to the
limited computing power and battery capacity of the user’s
device, it is difficult to handle computationally intensive tasks
locally. The emergence of cloud computing as a potential
solution formally initiated the third Internet revolution. Based
on the concept and advantages of cloud computing, mobile
cloud computing (MCC) was proposed in 2009 and refers
to a centralized cloud computing platform that migrates data
processing, storage and other tasks of intensive applications
from the original mobile device terminals to the cloud.
For applications that are closely involved in data-intensive
and delay-sensitive computing tasks, MCC has difficultly
satisfying the stringent requirements of real-time operations.
Therefore, a new computing paradigm, namely, fog com-
puting (FC), is extended from cloud computing. A fog can
be described as a cloud that is closer to the ground, which
pushes computing resources and application services to the
edge of data generation and processing. In reference [21],
the author considers the mobility of fog nodes. The task
assignment process between fixed and mobile fog nodes is
regarded as a two-objective optimization problem in which
the service latency and quality loss must be balanced. An
event-triggered dynamic task assignment framework that is
based on linear-programming-based optimization (LBO) and
binary particle swarm optimization (BPSO) is proposed for
solving joint optimization problems. In reference [22], a real-
time traffic management unloading scheme in IoV systems
that is based on fog computing is proposed, which can mini-
mize the average response time of vehicle reporting events.
Although fog computing has the advantages of location-
awareness and low latency, ubiquitous connectivity and ultra-
low latency requirements pose challenges to real-time traffic
management in smart cities [23].
To extend the cloud computing capacity to the edge of
the network, to enable the end users to use cloud computing
services more quickly and efficiently, and to improve the
user experience, in 2013, mobile edge computing (MEC) was
proposed for the provision of IT and cloud computing capa-
bilities for wireless access networks by deploying common
servers on the wireless access side. MEC is not a replacement
for MCC but an extension of cloud computing that relaxes the
transmission bandwidth and delay requirements. Compared
with MCC, MEC has the following characteristics [24]: (1)
low delay and low energy consumption: data generation and
processing are conducted close to the data sources and users,
thereby reducing the data transmission delay and energy
consumption; (2) diversity: edge devices with various com-
VOLUME 4, 2016 3
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
FIGURE 2: The architecture of MEC for IoV.
puting capabilities, such as roadside units (RSUs), vehicles
and WiFi hotspots, coexist; and (3) resource limitation: the
computing power of edge nodes is typically lower than that
of cloud servers.
With the progress of MEC standardization, the focus has
gradually shifted from targeting 3GPP mobile networks to
supporting non-3GPP networks (Wi-Fi and wired networks)
and even 5G networks. The name is also modified from
moving edge calculation to multi-access edge calculation.
The multi-access edge computing technology can realize the
interconnection of multiple wireless access technologies to
enable the computing/storage tasks of multiple servers to
be conducted cooperatively. [25] The emergence of MEC
servers enables wireless access networks to flexibly use com-
puting and storage resources while providing time-sensitive,
computation-intensive and highly reliable application ser-
vices. In reference [26], a scenario in which co-driving ve-
hicles and free-driving vehicles that are facilitated by HD
map interconnection via a wireless network co-existence is
designed, and a multi-access edge computing architecture
that is based on SDN and NFV technology is proposed. The
joint optimization of computing/storage resource manage-
ment between MEC servers and bandwidth resource slices
between Base Satations (BSs) effectively improves the uti-
lization of computing/storage resources. MEC, FC and MCC
are compared in TABLE I.
Fig. 2 shows that the architecture of vehicle edge com-
puting can be divided into a mobile user layer, an MEC
layer and a cloud layer. Communication between mobile
users generates a large amount of mobile data. By offloading
to RSUs, BSs, and other relay nodes with computing and
storage capabilities, tasks that require more computational
processing can be offloaded to MEC servers or even to remote
cloud servers to fully utilize the computing resources to
provide efficient computing services.
B. TECHNOLOGIES
1) Network Features Virtualization(NFV) and
Software-Defined Networks(SDN)
NFV enables the abstraction of physical network resources
and the flexible sharing of resources between isolated users
[27]. Virtualization technology is a key technology of MEC
and realizes the separation of the service layer and the phys-
ical resource layer of edge computing, and can assign tasks
to various physical resources, thereby efficiently utilizing re-
sources. By integrating NFV into the MEC server, virtualized
computing and storage resources can support the functions
of various applications and services and can be applied to
the server for functional programming to support a variety
of application services, thereby enhancing the flexibility of
the server and reducing the cost function supply [26]. SDN is
a new network mode that was proposed by the CLean State
research group of Stanford University in the United States.
It is an implementation method of NFV. By separating the
control surface of network devices from the data surface and
opening the programmability, the logic centralized control
of distributed network nodes and mobile devices can be
realized. Reference [28] studies SDN in super-dense network
task offloading problems and designs the edge of a cloud or
offloads tasks on a local process scheme; the main calculation
and control function is separated from the distributed small
unit base station, which is integrated into the centralized SD
UDN in the macrocell base station controller. Based on the
decision of the SD UDN controller, it is decided whether
the mobile device should perform tasks locally or offload
tasks to the edge cloud for processing, and the computing
resources should be optimally allocated to each task to realize
the objectives of minimizing the delay and preserving the
user’s device battery life.
2) Collaborative Mobile Edge Cloud Computing
Collaborative mobile edge cloud computing combines the
advantages of MEC and MCC, which is of substantial
significance for ensuring the full utilization of MEC and
cloud computing resources. While cloud computing may
produce long delays during offloading, it can provide suffi-
cient cloud computing resources. Although an MEC server
outperforms cloud computing in reducing the communication
delay and the energy consumption, with the increasing use
of computation-intensive applications, the limited computing
resources of the MEC server cannot fully satisfy all the
uninstallation requirements. With the increasing number of
computing tasks, the resource bottleneck problem of the
MEC server becomes increasingly prominent. Therefore,
cloud computing and MEC should be highly complementary.
Reference [29] proposed a collaborative offloading scheme
for vehicle-to-vehicle networks that is based on mobile
edge cloud computing and cloud computing, and it devel-
oped a distributed computing offloading and resource allo-
cation algorithm for computational offloading optimization
in vehicle-to-vehicle networks. Reference [30] proposes a
design framework for edge computing in wireless broadband
access networks that supports smart cities by embedding a
green, viable virtual network. A suitable resource partitioning
approach is used for each virtual network embedding, and
backing up edge devices by using heuristic policies to deter-
mine the number and geographic location is recommended.
4VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
TABLE 1: Comparison of MEC,FC and MCC
MEC FC MCC
Deployment Location Network edge Near edge Remote network center
Ownership Mobile operators Enterprise private Cloud provider
Location awareness High Medium Low
Distance from the user Near Far
Transmission Lelay Low High
Server Hardware Small data center, medium computing resources Large data centers with a large number
of high-performance computing servers
Network Architecture Multi-level, Distributed Centralized
Scalability High General
Application scenario Delay-sensitive applications
Widely distributed mobile applications
that require low latency and a small
amount of calculation
Applications with moderate delay
requirements but heavy calculations
In reference [31], Ning et al. designed an iterative heuristic
MEC resource allocation algorithm for making unloading
decisions dynamically. In reference [32], Hu et al. considered
the collaborative calculation of the offloading, combined
power and time distribution. They proposed a capture-unload
protocol that is based on the block-time-division mechanism
for minimizing the transmission power of wireless access
points.
3) Content Distribution
In the context of mass content delivery, a suitable content
distribution scheme can facilitate the avoidance of repeated
content transmission by the network. In addition, the ap-
plication of the content distribution framework in heteroge-
neous IoV systems can improve the message accuracy and
reduce the communication overhead between vehicles and
the infrastructure [37]. Current mobile users have consumed
a substantial amount of the capacity, and the demand for
in-vehicle infotainment services is still growing rapidly. To
improve the network performance and the user quality of
service (QoS), content distribution is often combined with
content caching technology and data prefetching technology
to further reduce the data access latency. In reference [38],
a content propagation box that is based on edge calculation
is proposed. First, a two-stage relay selection algorithm is
designed to facilitate edge computing devices in the selective
transmission of content via V2I communication. Then, the
vehicle that is selected by the edge computing device relays
the content via V2V communication to the vehicle that is
interested in the content during the trip to the destination.
Reference [39] proposed a content distribution framework
that utilizes 5G edge network caching and wireless link time
slot scheduling. The wireless resource allocation and return
link utilization of vehicle-to-roadside-unit communication
at each information station are considered. To maximize
the throughput, wireless links are dynamically allocated to
vehicles using time slots. In reference [40], we studied the
impacts of the storage cost and the retention time of con-
tent storage on cache optimization in mobile scenarios. In
addition, a cache problem in a vehicle network is modeled,
and its complexity is analyzed. For symmetric cases, an
optimal dynamic programming algorithm with polynomial
time complexity is developed. For general cases, a multi-
helper caching algorithm with low complexity and effective
retention perception is proposed, which can obtain the best
caching solution.
C. APPLICATIONS
For satisfying the strict requirements of limited mobile
terminal resources and computation-intensive and delay-
sensitive applications, computational offloading technology
is regarded as a key technology. Within the framework of
MEC, the mobile terminal can offload a task to the nearby
edge computing server for processing and feed back the
calculation results to the mobile terminal, thereby effectively
overcoming the resource limitation and reducing the power
consumption of the terminal during local calculation. Of-
floading decision, computational resource allocation and mo-
bility management of computational offloading are three key
issues in the field of MEC-based computational offloading.
In the following, we analyze the previous research on MEC
in IoV.
1) Offloading Decision
In offloading decision-making, data transfer between depen-
dent tasks is typically considered. Mobile terminal com-
puting offloading methods mainly include local computing,
offloading to the MEC server for execution and offloading
to the cloud server for execution. Many studies have been
conducted on offloading decision-making, such as studies on
whether to offload, the quantity and location of offloading,
service type, user perfection, access technology, network
traffic, device performance, and edge node property [41]. The
offloading method is mainly based on the resource size, the
calculation and return time and the power consumption of
the calculation. The main influencing factors are the delay
and the energy consumption. To minimize the cost in terms
of communication and computing resources, the author of
reference [25] proposed a task diversion mechanism in the
edge computing network of vehicles under the condition of
high mobility of the vehicles. The task offloading scheme
is analyzed in the scenario of an independent mobile edge
computing device server and in the scenario of a collabo-
VOLUME 4, 2016 5
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
TABLE 2: Comparison of Computing Offloading in MEC
Ref. The key technology User
Numbers
Offloading
Methods
Number of
Compute Nodes Compute
nodes Scheme
Offloading
Decision
Resource
Allocation
Mobility
Management
Single
node
Multi
-node
[6] √ √ multi-user Partial
Cloud server
APs
BSs
Adistributed algorithm for computing
polynomial complexity of equal
distribution of wireless and
cloud resources
[12] multi-user Partial
MEC server,
Cloud server,
RUs
Agreedy heuristic algorithm
for an on-board cloud edge system
[13] multi-user Partial BSs Agame theory method and a distributed
computing offloading algorithm
[21] multi-user Partial Fog node
Adynamic task assignment framework
based on programming optimization
and binary particle swarm optimization
[24] multi-user Partial Macro cell
RSUs
Ajoint power control and channel
allocation scheme
[25] √ √ multi-user Partial MEC server
Scheme 1:
in the independent MEC server scenario,
a task offloading scheme based on mobility.
Scheme 2:
in the collaborative MEC server scenario,
a location-based task unloading scheme .
[28] √ √ multi-user Partial BSs with edge
cloud servers
Atask unloading framework for
computing moving edges in software
-defined ultra-intensive networks
[29] √ √ multi-user Partial MEC server,
cloud server
Adistributed collaborative computing
algorithm for offloading and
resource allocation
[33] √ √ multi-user Partial MEC server
Amulti-user multi-task offload
scheduling scheme in a mobile
edge cloud system that can be updated
[34] √ √ multi-user Partial Cloud server
An efficient heuristic algorithm based
on semidefinite relaxation and a new
random mapping method.
[35] √ √ multi-user Partial MEC server
Ajoint optimization algorithm for the
selection and unloading of vehicle edge
computing servers
[36] multi-user Partial MEC server
Amulti-task two-layer computing
offloading framework for
heterogeneous networks
rative mobile edge computing device server. Reference [6]
proposed a polynomial-complexity algorithm for computing
the equal distribution of wireless and cloud resources in
dense wireless networks to minimize the computing costs.
The resource allocation problem of offloading computing to
the mobile cloud by mobile users is considered, where a
single mobile device can stream computations to the mobile
cloud via multiple access points or a base station.
2) Computational Resource Allocation
The objective of computing resource allocation is to mini-
mize the cost of task processing so that resources can be
fully and reasonably utilized. It consists of two processes:
task assignment, namely, the assignment of tasks that can
be executed in parallel to specified resources, and resource
allocation. The execution order of tasks is determined accord-
ing to the pre-established resource allocation strategy. In the
MEC scenario, computing resource allocation is also used
to improve the overall system performance and to reduce
the overall execution time and resource consumption. Com-
puting resources are often considered in conjunction with
offloading decisions, which can be divided into single-node
computing resource allocation and multi-node computing
resource allocation according to the numbers of users and
computing nodes, where in single-node allocation, a base
station can only serve one computing task in a time interval,
and in multi-node allocation, a base station in a time gap can
serve more than one computing task. In the multi-node com-
puting resource allocation scenario, the main problems focus
on communication interference between users and resource
competition [41]. In reference [33], the author proposed a
multi-user and multi-task offloading scheduling scheme in
the updatable mobile edge cloud system. Considering the
energy arrival of the mobile edge cloud and the dynamics
of task arrival of various mobile devices, an energy acqui-
sition strategy is proposed by combining energy acquisition
with mobile edge cloud computing. To maximize the system
utility and match the offloading energy consumption of the
mobile edge cloud with the acquired energy, a task offloading
scheduling scheme is proposed for mapping the computing
workload of a mobile device to multiple wireless devices.
In reference [34], a mobile cloud computing system that
consists of multiple users, a computing access point and a
remote cloud server is studied. An efficient heuristic algo-
6VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
rithm is proposed for handling the joint task of loading and
allocating computing and communication resources to mini-
mize the energy consumption, the calculated weighted total
cost, and the maximum latency among all users. In reference
[35], an algorithm is designed for making joint selection
decisions and calculating resources and the offload rate. A
comprehensive task processing delay is used to develop the
system utilities, which considers both the transmission and
computation times. This scheme substantially improves the
performance of load balancing and maximizes the system
availability.
3) Mobility Management of Computational Offloading
Mobility management of computing resource offloading is
of substantial significance to the integrity of the user com-
puting offloading process. Due to the mobility of the user,
it is inevitable that the user will be disconnected from the
base station. Mobile cellular networks ensure the continuity
and quality of service by switching among base stations.
For scenarios with low user mobility, during the process of
offloading the application to the MEC server, the power of
the current base station can be adjusted adaptively to ensure
uninterrupted service. If the user switches to a new service
base station, virtual machine migration of the compute node
is used to solve the problem. In reference [24], a joint task
allocation, subchannel allocation and power allocation prob-
lem is formulated. Aiming at maximizing the total offloading
rate, a hybrid computing shunt management framework for
real-time traffic in 5G networks is proposed. A joint power
control and channel allocation scheme is designed based on
non-orthogonal multiple access and mobile edge computing.
MEC can reduce the computing limitations and extend the
service life of mobile devices; however, it will lead to the
dense distribution of MEC servers. Although MEC servers
are close to the mobile users, they face user-related chal-
lenges, which will affect the computing shunt. Reference
[36] focuses on the joint computing of offloading and multi-
task user correlation, and it studies the scheduling problem
in distributed MEC systems with densely distributed MEC
servers. To reduce the energy consumption or improve the
performance, an efficient algorithm for calculating offloading
is proposed by considering the distribution of the computa-
tional resources and the transmitted power. A comparison of
computing offloading in MEC is presented in TABLE II.
III. ARTIFICIAL INTELLIGENCE IN INTERNET OF
VEHICLES
DRL is an essential technology for realizing AI. DRL utilizes
the advantages of deep neural networks (DNNs) to train the
learning process, thereby improving the learning speed and
performance of the RL algorithm and overcoming the unsuit-
ability of reinforcement learning for large-scale networks. In
this part, we introduce the development of AI, analyze the
relationship between AI and DRL, discuss the theory and
architecture of AI, and analyze the application of AI in IoV
research.
FIGURE 3: The relationship between AI, ML, RL, DL and
DRL.
A. ARCHITECTURE
The objective of artificial intelligence (AI) is to endow ma-
chines with human intelligence. Machine learning (ML) is a
method for implementing AI by using algorithms to parse
data, learn from data, and make decisions and predictions
regarding real-world events. Deep learning (DL) is a tech-
nology for realizing ML, which enables ML to realize many
applications and expands the scope of AI. Reinforcement
learning (RL), which is also known as evaluation learning,
is a technique of ML. Deep reinforcement learning (DRL)
is the combination of DL and RL. It aims at realizing the
optimization objective of RL with the operation mechanism
of DL to advance toward general AI. Fig. 3 illustrates the
relationship among AI, ML, RL, DL and DRL.
AI is a promising approach for making vehicle networks
intelligent. RL is a powerful tool in ML. In contrast to
traditional ML, RL does not have an immediate end result;
only a temporary reward (set primarily according to human
experience) is observed. Therefore, RL can also be regarded
as delayed supervised learning [16]. In the case of small
state space and behavior space, RL technology can be used
to enable network entities to identify the optimal strategy
for decision-making or behavior. However, in a complex
large-scale network, for improving the learning efficiency, a
learning method that combines RL with DL, namely, DRL, is
regarded as a potential solution [42]. The three key elements
of RL are the system status, the system actions, and the
rewards. In RL, the environment is typically represented as
a Markov decision process (MDP). Agents interact with the
unknown environment through repeated observation, action
and reward to construct the optimal strategy [8]. Due to the
limited data that are obtained from outside, DRL systems
often rely on their own experience to learn by themselves.
Via this approach, knowledge is acquired and solutions are
adapted to the environment. For the spatial-temporal cover-
age problem in mobile crowdsensing systems, reference [43]
proposes a vehicle selection scheme that is based on DRL.
Fig. 4 illustrates the architecture of AI in IoV, where
the agent observes its current environmental state, takes
VOLUME 4, 2016 7
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
FIGURE 4: Architecture for AI in IoV.
action, and receives its immediate reward along with the new
state. The observed information, which includes immediate
rewards and the new status, is used to adjust the agent’s
strategy, and the process is repeated until the agent’s strategy
approaches the optimal strategy.
B. TECHNOLOGIES
1) Markov Decision Processes(MDP)
The theoretical basis of MDP is a Markov chain (MC),
which is a stochastic process in a discrete index set and
state space. MDP provides a mathematical framework for
modeling decision problems in which the results are partially
random and controlled by the decision maker or agent. MDP
is used to model RL problems in ML, which facilitates the
study of dynamic programming and optimization problems
that can be solved via RL technology [42]. In reference [44],
the author proposed an architecture that combines a satellite
network with 5G cloud on-board Internet and designed a
joint optimization problem of computation offloading under
time-delay and cost constraints that is based on an incentive
mechanism. The solution uses the Markov chain Monte Carlo
and simulated annealing algorithms to effectively support
seamless coverage and global resource management. To over-
come the inability of mobile IP to cope with high-speed
and frequent vehicle movements, reference [45] adopts a
switching management scheme that is based on machine
learning in a two-tier intelligent transportation network. In
the first layer, the recursive neural network model is used
to predict the received signal strength of the access point to
obtain the switch trigger decision. In the second layer, the
random Markov model is used to predict the next access point
using the vehicle flow.
2) Q-learning and Deep Q-Learning(DQL)
Q-learning is a typical time-difference RL algorithm. The
Q-function is defined for the evaluation of the long-term
return of the strategy, and a neural network is used instead.
For each event, Q-learning makes a decision that is based
on the Q-value, which evaluates the selected operation in
the current scenario [16].When state space and action space
are small, Q-learning algorithm can effectively obtain the
optimal strategy. However, in practical applications, these
Spaces are often large due to the complexity of the system
model. In this case, the Q-learning algorithm may not be able
to find the optimal strategy. Therefore, the introduction of
DQL algorithm can overcome this shortcoming [42].Using
Q-learning or DQL algorithms can intelligently control the
use of network resources in IoV [20].Reference [46] estab-
lished a generic, green, intelligent, and scalable scheduling
strategy for resource distribution, which is used to adapt to
the randomness of the traffic environment, to learn from high-
dimensional input scheduling policies using the depth of the
Q-network, to support the efficient operation of the vehicle
network and balance the IoT gateway of the available energy,
and to minimize the total cost.
3) Long Short-Term Memory(LSTM)
LSTM enables a recurrent neural network (RNN) to evolve
into one of many network topologies. It is a time-cycling
neural network that can remember features in data at any
time interval. LSTM is composed of forward components
and backward components. LSTM solves the vanishing gra-
dient problem of RNN by explicitly introducing a storage
unit. LSTM can be used to create large recursive networks
to facilitate the solution of difficult ordering problems in
machine learning and to obtain the latest results [11]. In
reference [47], the author used a Markov decision process to
model the content caching problem in the Internet of vehicles
and proposed an active caching strategy of Q-learning which
is based on LSTM. In a service scenario in which Non-
Orthogonal Multiple Access (NOMA) users are randomly
deployed by a BS, reference [48] proposes a method that
is based on deep learning, the NOMA technology with
LSTM integration, a framework that can be automatically
and completely learned via the method of offline learning in
an unknown channel environment, end-to-end processing of
a NOMA wireless channel, and optimization that is based on
NOMA user activity and data detection.
C. APPLICATIONS
Compared with the traditional DRL-based centralized ap-
proach, the DRL-based distributed approach can learn in-
formation from the environment more quickly and can sub-
stantially reduce the communication overhead of vehicles.
In reference [49], an intelligent unloading framework of
a vehicle-mounted network that supports 5G is built. To
balance the transmission load, the cellular channel and the
multiplexed sub-channel are used for the task transmission.
According to the bilateral matching algorithm, all users are
divided into V2R and V2I users to allocate the unauthorized
spectrum. Then, a distributed deep reinforcement learning
algorithm is proposed for scheduling the cellular channel,
which can minimize the unloading cost under the premise
of satisfying individual delay constraints. Reference [50]
uses an online learning algorithm based on reinforcement
learning to propose a collaborative online caching strategy to
achieve content caching and updating. In reference [51], the
author proposes an RSU cloud, which is an infrastructure that
8VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
supports computing and communication in the Internet of
vehicles, that utilizes the dynamic programmability of SDN
and the cost analysis method of reconstruction. Modeling
cloud resource management as a multi-objective optimiza-
tion problem, with a heuristic algorithm and reinforcement
learning approach for the selection of the configuration that
minimizes the cost of reconfiguration, may yield high vir-
tual machine mobility immediately, but in the long run, the
opposite may occur. Most studies focus on the optimization
of mobile edge networks in consideration of the network,
communication and computing costs and cache. To satisfy
the requirements of system resource management scheduling
and system performance optimization, this is considered a
promising solution for improving the predictive performance
of channel state information in an edge computing network.
In reference [11], a channel prediction model that is based
on LSTM is proposed, and the associated algorithm, which
is based on deep learning, can predict future channel param-
eters based on past and present channel parameters. Basic
methods of machine learning often incur a large training cost.
Samples are difficult to obtain in practice, and if the network
parameters change, task mismatch easily occurs. Reference
[7], the authors designed a transmission strategy that is based
on deep learning by considering the social characteristics
of the edge of vehicle equipment and physical properties,
and they established a connection framework for assessing
interactions, in which a clustering algorithm that is based on
triangular patterns is used to control the network size and a
discovery algorithm that is based on a convolutional neural
network is used for data sharing with partners.
IV. ARTIFICIAL INTELLIGENCE EMPOWERED EDGE OF
VEHICLES
In the age of intelligent IoV, the application of AI to vehicle
edge networks is a promising approach for the development
of intelligent transportation. In this part, we introduce the
advantages of AI in applications to vehicle edge networks and
the architecture of AI-based vehicle edge networks. Then, the
related key technologies are described. Finally, we introduce
the previous research on the application of AI to vehicle edge
networks.
A. ARCHITECTURE
Traditional data sources are typically transferred remotely to
the cloud center, and services that are based on the mobile
cloud cannot guarantee the satisfaction of low-latency re-
quirements for content transfer [52]. Therefore, mobile edge
computing has the potential to overcome this challenge. Ac-
cording to [53], not only can MEC reduce the communication
latency, but MEC nodes can also use the potential resources
in the network to reduce the workload of the central base
station. In reference [54], a joint communication, caching
and computing (3C) model is proposed for the provision of
infotainment services in smart cars. It minimizes the latency
of access to infotainment services under resource constraints.
The problem of mixed-integer, nonlinear and non-convex
optimization is transformed into a linear programming prob-
lem via the relaxation technique, and its convergence is
demonstrated. In addition, according to [55], by using the
mobile edge cache to store the content on the edge of the
network, the content can be transmitted directly via wire-
less transmission without the need for backhauling or core
network transmission, thereby reducing the end-to-end delay
and the backhaul pressure.
As the applications of mobile users become richer and
more intelligent, they are faced with the requirements of mas-
sive data processing, delay-sensitivity, and location aware-
ness, among others. In recent years, artificial intelligence
(AI)-based vehicle edge computing has attracted substantial
attention. DRL is a tool of machine learning. The available
DRL technology can be applied to image processing, pattern
recognition, natural language processing and computation-
intensive applications. The integration of DRL technology
and vehicle edge computing is used to construct the intel-
ligent computing shunt system, which faces such problems
as high mobility of vehicles and difficulty finding continuous
image sequences. In reference [53], the authors use the finite-
state Markov chain, DRL and the calculation integration ve-
hicle edge to build an intelligent offloading system, and they
develop a joint optimization of task scheduling and resource
allocation problem in a traffic network, which is decomposed
into two sub-optimization problems: task scheduling among
multiple vehicles and the allocation of resources. The former
is solved via a bilateral matching algorithm, while the latter
is handled by an integrated DRL method. In addition, shared
edge computing services can be provided by mobile edge
servers that are deployed on the edge of the network to
improve the user quality of service. However, due to the
unevenness of space and the dynamics of time, the distri-
bution of vehicles is unbalanced. Therefore, an unbalanced
communication load of the mobile edge server is generated.
Reference [56] proposes an active load balancing method,
namely, an end-to-end load balancer, which uses a deep CNN
to learn spatiotemporal correlations and predict road traffic
conditions. A new framework that is based on CNN is used
to address the optimization problem of NLP, to fine-tune
the network from end to end, and to implement the efficient
collaborative scheduling of cached data between mobile edge
servers. In reference [57], the hybrid computation offloading
and intelligent cache problems in layered IoV with edge
intelligence are studied. However, to satisfy the demand of
real-time analysis of heterogeneous data from an intelligent
vehicle network and its environment, deep reinforcement
learning still faces many challenges [58].
Fig. 5 illustrates the architecture of a vehicle edge network
that is based on AI. The mobile vehicle communicates with a
roadside unit (RSU) that is equipped with an MEC server
via an on-board unit (OBU). RSUs have computing and
storage capabilities, and multiple RSUs can communicate.
Computationally heavy tasks can be offloaded to the base
station (BS), and the collected data can be used to make in-
telligent decisions (such as predicting the direction of vehicle
VOLUME 4, 2016 9
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
FIGURE 5: The architecture of vehicle edge network based
on AI.
movement) with the help of high computing power, and can
be used to support deep learning.
B. TECHNOLOGIES
1) Collaborative edge caching
Collaborative edge caching combines the advantages of mo-
bile edge caching and collaborative caching. By actively
storing files in the base station (BS), the mobile edge cache
can provide content directly without remote file extraction,
which reduces the end-to-end latency and the backhaul stress.
Simultaneously, to effectively utilize the limited cache size,
collaborative caching can be used to improve the diversity
of the cache. Under the new user-centered network archi-
tecture, multiple base stations can serve users. In addition,
collaborative caching can improve the cache hit performance
to overcome the moderate cache hit performance bottleneck
that is caused by the relatively small cache storage on a
single BS. In reference [55], collaborative edge caching in
large-scale user-centered clustered mobile networks is stud-
ied, and a greedy content layout algorithm that is based on
optimal bandwidth allocation is proposed for minimizing the
cache size and the average file transfer rate under bandwidth
constraints. In reference [59], the author proposed an edge
network cache replacement strategy that is based on deep
learning using a deep LSTM network. The joint framework
is used to merge the smart cache replacement algorithm
and the corresponding collaboration mechanism. The cache
strategy is automatically learned in real time from the request
sequence to reduce the transmission latency and the backhaul
data traffic.
2) Multi Armed Bandit
MAB is a reinforcement learning method. It has been exten-
sively studied for addressing the key tradeoff between explo-
ration and development in sequential decision-making under
uncertain conditions. The original k-armed bandit problem
assumed that one option was selected from k options repeat-
edly. The option is the arm. Each time an option is selected, a
reward is obtained as feedback, and the action selection is
repeated to focus the action on the best arm to maximize
the expected total reward in a period of time. To cope
with the unknown service requirements in the changing user
groups, reference [60] proposed a combined context bandit
learning problem. A spatiotemporal edge service placement
algorithm is used to solve the problem. Multiple learners
are considered, and each learner can maintain a distinct
location-specific context space. The context information of
connected users is collected according to users’ preferences,
and location-awareness and context-awareness are realized
for renting computing resources flexibly and economically
in the shared edge computing platform. Reference [61] pro-
posed a distributed adaptive task offloading algorithm that is
based on learning that is based on multi-arm bandit theory.
It enables the vehicle to learn the offloading delay perfor-
mance of an adjacent vehicle while offloading the calculation
task, eliminates the need for frequent state exchanges, and
increases the input and occurrence awareness for adaptation
to the dynamic environment.
3) Nonorthogonal multiple access (NOMA)
As an emerging technology in 5G networks, NOMA has
advantages in terms of its spectrum, connectivity, energy
efficiency and other aspects, thereby enabling multiple users
to reuse frequency resources nonorthogonally. NOMA tech-
nology not only has advantages in increasing the system
throughput and supporting large-scale connections but can
also be used to eliminate multi-user interference in multi-user
detection systems by assigning power levels according to
users’ channel conditions. Reference [65] studied the cache-
assisted non-orthogonal multiplexing access of the onboard
network that supports 5G. Considering the full-file cache
and split-file cache, the optimization problem of the overall
probability of decoding the files successfully in each vehicle
is formulated and solved in the first scenario. In the second
scenario, a joint power distribution optimization problem
is proposed for determining the power distribution between
vehicles and individual files. In reference [66], the author
considers task offloading and user selection between macro
units and edge devices. A moving edge algorithm that is
based on non-orthogonal multiple access is proposed, and a
heuristic algorithm is designed from the aspects of offloading
decision, channel allocation and power control to improve the
transmission rate gain and the discharge offloading efficiency.
C. APPLICATIONS
The emerging 5G mobile network has the advantages of high
bandwidth and low latency. By expanding the antenna scale,
the 5G wireless network can improve frequency reuse and
increase the capacity of the cellular network via network
densification. However, in the face of massive data, the
traditional caching strategy has encountered a bottleneck.
Therefore, the proposal of a moving edge cache is extremely
important in the 5G network, which can provide higher
service quality for many new applications. Caching content
in the base station can significantly reduce the network
10 VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
TABLE 3: Comparison of Edge Caching Based on AI Algorithm
Cache Type Ref. Optimization Objective Program Result
Non-
cooperative
cache
[52] Minimize content transfer
latency
Content caching strategy based on
two-dimensional Markov chain
It is verified that the prediction
accuracy increases and the delay of
content transmission gradually decreases
[62] Minimize the long-term
cost of acquiring IoT data
Framework of Internet of Things
system based on edge cache
This solution can reduce the long-term
cost for users to obtain IoT data
Cooperative
caching
[18] Maximize system
availability
An edge computing and caching
scheme based on AI algorithm
This solution greatly improves the
practicability of the system
[27]
Reduce latency, network
burden and communication
redundancy
An integrated framework that
dynamically coordinates network,
cache, and computing resources
The framework provides network
functions, caching functions and
computing functions
[55]
Minimize the transmission
rate under bandwidth
constraints
Agreedy content layout algorithm
based on optimal bandwidth allocation
Reduced average file transfer latency
to 45%
[59]
Reduce transmission
latency and backhaul
network load
An edge network cache replacement
strategy based on deep learning
14% to 22% reduction in overall
transmission latency and 15% to 23%
savings in backhaul data traffic
[63]
Reduce latency due to
backhaul bandwidth
consumption
Convolutional neural network analysis
method and Multi-Layer Perceptron
method to predict cache content
The accuracy of predicting infotainment
content to be cached can reach 99.28%
[64] Minimize system cost
under latency constraints
Amulti-time-scale framework based
on artificial intelligence
The scheme effectively mitigates the
harmful effects of limited backhaul
capacity and low BS computing resources
latency, whereas caching content on the edge can reduce
the data traffic in the core network and conserve bandwidth
for the Internet [59]. In addition, edge caching can improve
the spectrum efficiency and reduce the energy consumption
due to device heterogeneity and dense deployment [67]. IoT
data are transient; for example, the popularity distribution
of data may vary with the time and location, and static-
based caching strategies have difficulty satisfying the various
requirements of IoT services, such as mobility and geo-
graphically distributed support. In reference [62], for caching
temporary data on the edge of the IoT, the author proposed a
framework of the IoT system that is based on the edge cache.
Considering both the “data freshness” and the “edge cache”,
the cache strategy of the deep reinforcement learning method
can make smart cache decisions without assuming the data
popularity or user request distribution. In reference [63], the
author uses a convolutional neural network to predict and
obtain the user’s age and gender characteristics. By deploying
multiple-access edge computing servers on roadside units,
WiFi access points and acer stations for caching infotain-
ment content in and around self-driving cars, fog computing
extends the infrastructure of traditional cloud computing to
the edge of the network, thereby substantially reducing the
long-distance latency from the terminal to the cloud server.
Since edge servers are distributed in the surrounding area,
fog computing is expected to improve the data transmission
efficiency. Reference [52] proposed a vehicle edge collabo-
rative filtering content transmission scheme that is based on
fog calculation. A collective filtering algorithm and a two-
dimensional Markov chain are used to combine positional
awareness, content caching, and decentralized computing for
content precaching at the edge of the vehicle network. Due
to the highly dynamic network environment and the uncer-
tainty of mobile users, reference [5] proposed the concept of
vehicle caching, which uses vehicle mobility to improve the
service scope and cache capacity. The interaction between
a cached vehicle and the mobile user is modeled as a two-
dimensional Markov process. On this basis, an online vehicle
cache design scheme that is based on network energy effi-
ciency optimization is proposed. It is proved to outperform
the available scheme in terms of the hit ratio, energy effi-
ciency, cache utilization and system gain. Machine learning
is also an emerging tool for solving caching, computing and
communication problems in 5G wireless communication.
Various studies, such as [18] and [27], have investigated the
joint optimization of computing resources and caching. In
reference [18], large amounts of data and popular content
are produced by computation-intensive applications, time-
delay-sensitive applications, and on-board sensors. This pa-
per discusses the resource processing and storage of vehicles
with limited resources in the Internet of vehicles. An AI-
based algorithm is proposed for dynamically orchestrating
the architecture of edge computing and cache resources, and
a novel resource management scheme is developed, which
uses deep reinforcement learning. In contrast to other studies,
this study uses a two-layer cross-layer offloading model that
combines a heterogeneous network and mobile edge com-
puting to realize dynamic resource allocation. In reference
[27], the principle of programmable control of the network
that is defined by software, the principle of caching in infor-
mation and communication technology and the principle of
network virtualization are used to construct the framework
of the dynamic arrangement of integrated network, cache
and computing resources. The main disadvantage is the lack
of consideration of the energy consumption. In references
[64] and [8], the joint optimization of resource allocation
VOLUME 4, 2016 11
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
for caching, computing and communication is considered. In
reference [64], an algorithm that is based on an AI multi-
temporal framework is designed, which facilitates the con-
figuration of cache placement and the calculation of the pa-
rameters of resource allocation. For cost minimization under
the constraints of limited RSU storage capacity, dynamic
fluctuations in computing resources, vehicle mobility, and
strict end-to-end delay limits, reference [8] proposed a deep
reinforcement learning method that is based on a multi-time-
scale framework. Vehicle mobility is leveraged to enhance
the caching and computing strategies. A long-time-scale
model of motion perception reward estimation is proposed
to reduce the complexity that is due to large action spaces.
Resource allocation and computational offloading are inextri-
cably linked. Reference [68] studied the optimal utility task
offloading scheme in a heterogeneous vehicle network with
multiple mobile edge computing servers under constraints on
the reliability and waiting time and proposed an adaptive
redundancy offloading algorithm that is based on deep Q-
learning to ensure the reliability of offloading and to improve
the practicability of the system. Reference [16] proposed an
energy-saving task offloading scheme that is based on DRL
and combined it with fog computing technology. Considering
load balancing and time delay constraints, an optimization
problem was formulated for minimizing the energy consump-
tion of traffic offloading, which was decomposed into two
parts: flow redirection and offloading decision. Algorithms
that are based on Edmonds-Karp and DRL were developed
for solving the problem. In reference [69], a distributed
dynamic computing offloading strategy that is based on DRL
is proposed for dynamic task offloading control of multi-user
MEC systems to minimize the long-term average computing
cost consumption and the task buffer delay in the power. A
comparison of edge caching that is based on AI algorithms is
presented in TABLE III.
V. RESEARCH CHALLENGE AND OPEN ISSUES
In the previous sections, we reviewed the architecture and
related technologies of MEC, AI, and AI-based vehicle edge
networks in IoV. In addition, we analyzed the previous re-
search from three aspects. However, the future IoV still faces
challenges. In this section, we will discuss several possible
research challenges and propose several promising research
directions.
A. SECURITY AND PRIVACY
In recent years, security and privacy issues in IoV have
received extensive attention. Mobile vehicles collect infor-
mation via V2V communication between vehicles and via
V2I communication from vehicles to roadside infrastructure.
Due to the high mobility of vehicles, communication is
often interrupted, thereby resulting in frequent failures of
communication links. In addition, hackers’ security attacks
on communication channels and sensor tampering will lead
to severe privacy invasion. In addressing these security and
privacy issues, challenges in the solution of identity privacy,
data privacy and location privacy issues will be encountered.
Potential solutions include communication authentication,
MEC and access control of cloud computing servers [70].
Reference [71] proposes an architecture of edge auxiliary
network connecting vehicles. To solve the problem of loca-
tion privacy, a location-based differential privacy protection
service framework is proposed for ensuring location privacy
within the coverage of the edge nodes. Li et al. [72] proposed
an online double auction scheme for k-anonymous location
privacy protection, which could solve the problems of op-
timal charging scheduling for electric vehicles and location
privacy protection for owners of electric vehicles. Chen et
al. [73] designed a data trading method for the Internet
of vehicles that is based on block chain. An iterative dual
auction mechanism is used to protect the privacy of both
parties in data transaction, to reduce the data transmission
cost and to improve the system stability.
B. GREEN ENERGY SAVING
Green energy saving has a profound impact on the construc-
tion of a green IoV. Automobile exhaust emission is the
main factor that affects the human environment and the air
environment. To alleviate the current environmental pollution
scenario while adapting to the highly dynamic traffic envi-
ronment, it is highly important to use RSUs to communicate
with nearby vehicles to realize efficient task scheduling to
satisfy vehicle communication requirements. Energy saving
in RSU scheduling and RSU energy collection are essential
for solving the problem of energy consumption. The imple-
mentation of a wind or solar RSU in an energy-constrained
vehicle environment can increase the network capacity and
promote energy recovery. In addition, the minimum number
of active RSUs can be set to maintain the network operation
and connectivity [74]. To minimize the total energy con-
sumption of RSUs under the delay constraint, reference [75]
constructed an MEC-based IoV energy-saving scheduling
framework for balancing the computing tasks among RSUs.
A heuristic algorithm is designed that considers the task
scheduling among MEC servers and the energy consumption
of the RSU downlink. In addition, electric vehicles, which are
powered by electric engines instead of internal combustion
engines, which are powered by fossil fuels, can effectively
reduce the carbon footprint and play an important role in
realizing efficient energy management [76].
C. HIGH MOBILITY
Mobility is an important feature of vehicle networks. With
the rapid increase of the road traffic density, high speed
and frequent vehicle movements are the main factors that
render the network topology dynamic. The high mobility of
intelligent vehicles not only adds considerable complexity
in co-optimizing the allocation of computing and cache re-
sources but also hinders the provision of stable and reliable
wireless communication [77]. First, the data transmission
distance is constantly changing due to vehicle movements.
Therefore, the data rate and the effective duration of channel
12 VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
transmission in V2X communication will also be affected.
Second, the changes in the vehicle speed and direction over
time will lead to frequent handovers between edge servers.
Active communication management is transferred from one
RSU or BS to another RSU or BS. The duration for which a
vehicle remains within the coverage area of the RSU or BS
also varies. Due to the widespread use of various Global Po-
sitioning System (GPS) devices and mobile Internet in daily
life, vehicle trajectory data can be easily obtained on a large
scale [78]. Therefore, addressing the high mobility of IoV by
predicting the vehicle movement direction and studying data
routing distribution protocols is a feasible solution [79].
D. INTELLIGENT COMPUTATION
With many edge nodes deployed in 5G networks, edge com-
puting has the advantage of reducing the traffic load and the
backhaul pressure, but edge devices still face the challenge
of real-time processing. Edge cognitive computing has be-
come a new paradigm. By analyzing and interpreting the
available data and information in cyberspace, the intelligence
of machines can be increased for the prediction and genera-
tion of new information, thereby providing more intelligent
cognitive services. Reference [80] proposed an architecture
of edge cognitive computing by combining edge computing
and cognitive computing. Considering the elastic distribution
of cognitive computing services and the mobility of users,
a dynamic cognitive service migration mechanism that is
based on edge cognitive computing is designed. It integrates
the communication, computing, storage and application on
the edge network, improves the user experience and realizes
rational resource allocation and cognitive information circu-
lation.
VI. CONCLUSION
In this study, two key technologies, namely, MEC and AI,
were analyzed by focusing on the development of intelli-
gent IoV and the previous research on combining the two
technologies. First, the communication mode and architec-
ture of the traditional Internet of vehicles were introduced,
along with the advantages of the emerging 5G network. In
addition, MEC, FC and MCC were compared by studying
the development history of MEC. The advantages of MEC
were analyzed, the MEC, FC and MCC were compared by
studying the development history of MEC. The advantages
of MEC were analyzed, the key technologies of MEC were
evaluated, and several key technologies for calculating the
unloading in MEC were studied. Then, the differences and
connection between AI and DRL in realizing intelligent IoV
were discussed, with a focus on the characteristics and ap-
plication status of DRL in dynamic vehicle networks. Then,
combining the advantages of MEC and AI technologies, the
previous research on the application of AI to vehicle edge
networks was analyzed. Finally, the possible future research
directions of IoV were discussed.
REFERENCES
[1] L. Mendiboure, M. A. Chalouf, and F. Krief, “Towards a 5g vehicular
architecture,” in International Workshop on Communication Technologies
for Vehicles. Springer, 2019, pp. 3–15.
[2] S. Guo, C. Chen, J. Wang, Y. Liu, X. Ke, Z. Yu, D. Zhang, and D.-M. Chiu,
“Rod-revenue: Seeking strategies analysis and revenue prediction in ride-
on-demand service using multi-source urban data,” IEEE Transactions on
Mobile Computing, 2019.
[3] C. Chen, D. Zhang, X. Ma, B. Guo, L. Wang, Y. Wang, and E. Sha,
“Crowddeliver: planning city-wide package delivery paths leveraging the
crowd of taxis,” IEEE Transactions on Intelligent Transportation Systems,
vol. 18, no. 6, pp. 1478–1496, 2016.
[4] P. Lin, Q. Song, and A. Jamalipour, “Multidimensional cooperative
caching in comp-integrated ultra-dense cellular networks,” IEEE Trans-
actions on Wireless Communications, 2019.
[5] Y. Zhang, C. Li, T. H. Luan, Y. Fu, W. Shi, and L. Zhu, “A mobility-
aware vehicular caching scheme in content centric networks: Model and
optimization,” IEEE Transactions on Vehicular Technology, vol. 68, no. 4,
pp. 3100–3112, 2019.
[6] S. Jošilo and G. Dán, “Selfish decentralized computation offloading for
mobile cloud computing in dense wireless networks,” IEEE Transactions
on Mobile Computing, vol. 18, no. 1, pp. 207–220, 2018.
[7] Z. Ning, Y. Feng, M. Collotta, X. Kong, X. Wang, L. Guo, X. Hu, and
B. Hu, “Deep learning in edge of vehicles: Exploring trirelationship for
data transmission,” IEEE Transactions on Industrial Informatics, vol. 15,
no. 10, pp. 5737–5746, 2019.
[8] R. Q. Hu et al., “Mobility-aware edge caching and computing in vehicle
networks: A deep reinforcement learning,” IEEE Transactions on Vehicu-
lar Technology, vol. 67, no. 11, pp. 10 190–10203, 2018.
[9] C. Chen, Y. Ding, Z. Wang, J. Zhao, B. Guo, and D. Zhang, “Vtracer:
When online vehicle trajectory compression meets mobile edge comput-
ing,” IEEE Systems Journal, 2019.
[10] Z. Ning, F. Xia, N. Ullah, X. Kong, and X. Hu, “Vehicular social networks:
Enabling smart mobility,” IEEE Communications Magazine, vol. 55, no. 5,
pp. 16–55, 2017.
[11] G. Liu, Y. Xu, Z. He, Y. Rao, J. Xia, and L. Fan, “Deep learning-
based channel prediction for edge computing networks toward intelligent
connected vehicles,” IEEE Access, vol. 7, pp. 114 487–114 495, 2019.
[12] I. Sorkhoh, D. Ebrahimi, R. Atallah, and C. Assi, “Workload scheduling
in vehicular networks with edge cloud capabilities,” IEEE Transactions on
Vehicular Technology, vol. 68, no. 9, pp. 8472–8486, 2019.
[13] X. Chen, L. Jiao, W. Li, and X. Fu, “Efficient multi-user computation
offloading for mobile-edge cloud computing,” IEEE/ACM Transactions on
Networking, vol. 24, no. 5, pp. 2795–2808, 2015.
[14] Z. Ning, P. Dong, X. Wang, X. Hu, L. Guo, B. Hu, Y. Guo, T. Qiu, and R. Y.
Kwok, “Mobile edge computing enabled 5g health monitoring for internet
of medical things: A decentralized game theoretic approach,” IEEE Journal
on Selected Areas in Communications, 2020.
[15] P. Dong, Z. Ning, M. S. Obaidat, P. Jiang, Y. Guo, X. Hu, B. Hu, and
B. Sadoun, “Edge computing based healthcare systems: Enabling decen-
tralized health monitoring in internet of medical things,” IEEE Network,
2020.
[16] Z. Ning, P. Dong, X. Wang, L. Guo, J. J. Rodrigues, X. Kong, J. Huang,
and R. Y. Kwok, “Deep reinforcement learning for intelligent internet
of vehicles: An energy-efficient computational offloading scheme,” IEEE
Transactions on Cognitive Communications and Networking, vol. 5, no. 4,
pp. 1060–1072, 2019.
[17] Z. Ning, X. Hu, Z. Chen, M. Zhou, B. Hu, J. Cheng, and M. S. Obaidat,
“A cooperative quality-aware service access system for social internet of
vehicles,” IEEE Internet of Things Journal, vol. 5, no. 4, pp. 2506–2517,
2017.
[18] Y. Dai, D. Xu, S. Maharjan, G. Qiao, and Y. Zhang, “Artificial intelligence
empowered edge computing and caching for internet of vehicles,” IEEE
Wireless Communications, vol. 26, no. 3, pp. 12–18, 2019.
[19] K.-H. N. Bui and J. J. Jung, “Aco-based dynamic decision making for
connected vehicles in iot system,” IEEE Transactions on Industrial Infor-
matics, vol. 15, no. 10, pp. 5648–5655, 2019.
[20] H. Yang, X. Xie, and M. Kadoch, “Intelligent resource management
based on reinforcement learning for ultra-reliable and low-latency iov
communication networks,” IEEE Transactions on Vehicular Technology,
vol. 68, no. 5, pp. 4157–4169, 2019.
[21] C. Zhu, J. Tao, G. Pastor, Y. Xiao, Y. Ji, Q. Zhou, Y. Li, and A. Ylä-
Jääski, “Folo: Latency and quality optimized task allocation in vehicular
VOLUME 4, 2016 13
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
fog computing,” IEEE Internet of Things Journal, vol. 6, no. 3, pp. 4150–
4161, 2018.
[22] X. Wang, Z. Ning, and L. Wang, “Offloading in internet of vehicles: A
fog-enabled real-time traffic management system,” IEEE Transactions on
Industrial Informatics, vol. 14, no. 10, pp. 4568–4578, 2018.
[23] Z. Ning, J. Huang, and X. Wang, “Vehicular fog computing: Enabling
real-time traffic management for smart cities,” IEEE Wireless Communi-
cations, vol. 26, no. 1, pp. 87–93, 2019.
[24] Z. Ning, X. Wang, J. J. Rodrigues, and F. Xia, “Joint computation offload-
ing, power allocation, and channel assignment for 5g-enabled traffic man-
agement systems,” IEEE Transactions on Industrial Informatics, vol. 15,
no. 5, pp. 3058–3067, 2019.
[25] C. Yang, Y. Liu, X. Chen, W. Zhong, and S. Xie, “Efficient mobility-aware
task offloading for vehicular edge computing networks,” IEEE Access,
vol. 7, pp. 26 652–26 664, 2019.
[26] H. Peng, Q. Ye, and X. S. Shen, “Sdn-based resource management for au-
tonomous vehicular networks: A multi-access edge computing approach,”
IEEE Wireless Communications, vol. 26, no. 4, pp. 156–162, 2019.
[27] Y. He, F. R. Yu, N. Zhao, V. C. Leung, and H. Yin, “Software-defined
networks with mobile edge computing and caching for smart cities: A
big data deep reinforcement learning approach,” IEEE Communications
Magazine, vol. 55, no. 12, pp. 31–37, 2017.
[28] M. Chen and Y. Hao, “Task offloading for mobile edge computing in
software defined ultra-dense network,” IEEE Journal on Selected Areas
in Communications, vol. 36, no. 3, pp. 587–597, 2018.
[29] J. Zhao, Q. Li, Y. Gong, and K. Zhang, “Computation offloading and
resource allocation for cloud assisted mobile edge computing in vehicular
networks,” IEEE Transactions on Vehicular Technology, vol. 68, no. 8, pp.
7944–7956, 2019.
[30] W. Hou, Z. Ning, and L. Guo, “Green survivable collaborative edge
computing in smart cities,” IEEE Transactions on Industrial Informatics,
vol. 14, no. 4, pp. 1594–1605, 2018.
[31] Z. Ning, P. Dong, X. Kong, and F. Xia, “A cooperative partial computation
offloading scheme for mobile edge computing enabled internet of things,”
IEEE Internet of Things Journal, vol. 6, no. 3, pp. 4804–4814, 2018.
[32] X. Hu, K.-K. Wong, and K. Yang, “Wireless powered cooperation-assisted
mobile edge computing,” IEEE Transactions on Wireless Communica-
tions, vol. 17, no. 4, pp. 2375–2388, 2018.
[33] W. Chen, D. Wang, and K. Li, “Multi-user multi-task computation of-
floading in green mobile edge cloud computing,” IEEE Transactions on
Services Computing, vol. 12, no. 5, pp. 726–738, 2018.
[34] M.-H. Chen, M. Dong, and B. Liang, “Resource sharing of a computing ac-
cess point for multi-user mobile cloud offloading with delay constraints,”
IEEE Transactions on Mobile Computing, vol. 17, no. 12, pp. 2868–2881,
2018.
[35] Y. Dai, D. Xu, S. Maharjan, and Y. Zhang, “Joint load balancing and
offloading in vehicular edge computing and networks,” IEEE Internet of
Things Journal, vol. 6, no. 3, pp. 4377–4387, 2018.
[36] Y. Dai, D. Xu, S. Maharjan, and Y. Zhang, “Joint computation offloading
and user association in multi-task mobile edge computing,” IEEE Transac-
tions on Vehicular Technology, vol. 67, no. 12, pp. 12 313–12325, 2018.
[37] X. Wang, Z. Ning, X. Hu, L. Wang, B. Hu, J. Cheng, and V. C. Leung,
“Optimizing content dissemination for real-time traffic management in
large-scale internet of vehicle systems,” IEEE Transactions on Vehicular
Technology, vol. 68, no. 2, pp. 1093–1105, 2018.
[38] Y. Hui, Z. Su, T. H. Luan, and J. Cai, “Content in motion: An edge
computing based relay scheme for content dissemination in urban vehic-
ular networks,” IEEE Transactions on Intelligent Transportation Systems,
vol. 20, no. 8, pp. 3115–3128, 2018.
[39] H. Zhou, N. Cheng, J. Wang, J. Chen, Q. Yu, and X. Shen, “Toward
dynamic link utilization for efficient vehicular edge content distribution,
IEEE Transactions on Vehicular Technology, vol. 68, no. 9, pp. 8301–
8313, 2019.
[40] T. Deng, P. Fan, and D. Yuan, “Optimizing retention-aware caching in
vehicular networks,” IEEE Transactions on Communications, vol. 67,
no. 9, pp. 6139–6152, 2019.
[41] Q. Qi, J. Wang, Z. Ma, H. Sun, Y. Cao, L. Zhang, and J. Liao, “Knowledge-
driven service offloading decision for vehicular edge computing: A deep
reinforcement learning approach,” IEEE Transactions on Vehicular Tech-
nology, vol. 68, no. 5, pp. 4192–4203, 2019.
[42] N. C. Luong, D. T. Hoang, S. Gong, D. Niyato, P. Wang, Y.-C. Liang,
and D. I. Kim, “Applications of deep reinforcement learning in commu-
nications and networking: A survey,” IEEE Communications Surveys &
Tutorials, vol. 21, no. 4, pp. 3133–3174, 2019.
[43] C. Wang, X. Gaimu, C. Li, H. Zou, and W. Wang, “Smart mobile crowd-
sensing with urban vehicles: A deep reinforcement learning perspective,
IEEE Access, vol. 7, pp. 37 334–37 341, 2019.
[44] M. LiWang, S. Dai, Z. Gao, X. Du, M. Guizani, and H. Dai, “A computa-
tion offloading incentive mechanism with delay and cost constraints under
5g satellite-ground iov architecture,” IEEE Wireless Communications,
vol. 26, no. 4, pp. 124–132, 2019.
[45] N. Aljeri and A. Boukerche, “A two-tier machine learning-based han-
dover management scheme for intelligent vehicular networks,” Ad Hoc
Networks, vol. 94, p. 101930, 2019.
[46] R. F. Atallah, C. M. Assi, and M. J. Khabbaz, “Scheduling the operation of
a connected vehicular network using deep reinforcement learning,” IEEE
Transactions on Intelligent Transportation Systems, vol. 20, no. 5, pp.
1669–1682, 2018.
[47] L. Hou, L. Lei, K. Zheng, and X. Wang, “A q-learning-based proactive
caching strategy for non-safety related services in vehicular networks,
IEEE Internet of Things Journal, vol. 6, no. 3, pp. 4512–4520, 2018.
[48] G. Gui, H. Huang, Y. Song, and H. Sari, “Deep learning for an effective
nonorthogonal multiple access scheme,” IEEE Transactions on Vehicular
Technology, vol. 67, no. 9, pp. 8440–8450, 2018.
[49] Z. Ning, P. Dong, X. Wang, M. S. Obaidat, X. Hu, L. Guo, Y. Guo,
J. Huang, B. Hu, and Y. Li, “When deep reinforcement learning meets 5g-
enabled vehicular networks: A distributed offloading framework for traffic
big data,” IEEE Transactions on Industrial Informatics, vol. 16, no. 2, pp.
1352–1361, 2020.
[50] P. Lin, Q. Song, J. Song, A. Jamalipour, and F. R. Yu, “Cooperative caching
and transmission in comp-integrated cellular networks using reinforce-
ment learning,” IEEE Transactions on Vehicular Technology, 2020.
[51] M. A. Salahuddin, A. Al-Fuqaha, and M. Guizani, “Software-defined
networking for rsu clouds in support of the internet of vehicles,” IEEE
Internet of Things Journal, vol. 2, no. 2, pp. 133–144, 2014.
[52] X. Wang, Y. Feng, Z. Ning, X. Hu, X. Kong, B. Hu, and Y. Guo, “A
collective filtering based content transmission scheme in edge of vehicles,
Information Sciences, vol. 506, pp. 161–173, 2020.
[53] Z. Ning, P. Dong, X. Wang, J. J. Rodrigues, and F. Xia, “Deep reinforce-
ment learning for vehicular edge computing: An intelligent offloading sys-
tem,” ACM Transactions on Intelligent Systems and Technology (TIST),
vol. 10, no. 6, p. 60, 2019.
[54] S. A. Kazmi, T. N. Dang, I. Yaqoob, A. Ndikumana, E. Ahmed, R. Hus-
sain, and C. S. Hong, “Infotainment enabled smart cars: A joint com-
munication, caching, and computation approach,” IEEE Transactions on
Vehicular Technology, vol. 68, no. 9, pp. 8408–8420, 2019.
[55] S. Zhang, P. He, K. Suto, P. Yang, L. Zhao, and X. Shen, “Cooperative edge
caching in user-centric clustered mobile networks,” IEEE Transactions on
Mobile Computing, vol. 17, no. 8, pp. 1791–1805, 2017.
[56] J. Li, G. Luo, N. Cheng, Q. Yuan, Z. Wu, S. Gao, and Z. Liu, “An end-
to-end load balancer based on deep learning for vehicular network traffic
control,” IEEE Internet of Things Journal, vol. 6, no. 1, pp. 953–966, 2018.
[57] Z. Ning, K. Zhang, X. Wang, L. Guo, X. Hu, J. Huang, B. Hu, and R. Y.
Kwok, “Edge intelligence for internet of vehicles: A joint hybrid compu-
tation offloading and intelligent caching algorithm,” IEEE Transactions on
Intelligent Transportation Systems, 2020.
[58] A. Ferdowsi, U. Challita, and W. Saad, “Deep learning for reliable mobile
edge analytics in intelligent transportation systems: An overview,” IEEE
Vehicular Technology Magazine, vol. 14, no. 1, pp. 62–70, 2019.
[59] H. Pang, J. Liu, X. Fan, and L. Sun, “Toward smart and cooperative
edge caching for 5g networks: A deep learning based approach,” in
2018 IEEE/ACM 26th International Symposium on Quality of Service
(IWQoS). IEEE, 2018, pp. 1–6.
[60] L. Chen, J. Xu, S. Ren, and P. Zhou, “Spatio–temporal edge service
placement: A bandit learning approach,” IEEE Transactions on Wireless
Communications, vol. 17, no. 12, pp. 8388–8401, 2018.
[61] Y. Sun, X. Guo, J. Song, S. Zhou, Z. Jiang, X. Liu, and Z. Niu, “Adaptive
learning-based task offloading for vehicular edge computing systems,
IEEE Transactions on Vehicular Technology, vol. 68, no. 4, pp. 3061–
3074, 2019.
[62] H. Zhu, Y. Cao, X. Wei, W. Wang, T. Jiang, and S. Jin, “Caching transient
data for internet of things: A deep reinforcement learning approach,” IEEE
Internet of Things Journal, vol. 6, no. 2, pp. 2074–2083, 2018.
[63] A. Ndikumana and C. S. Hong, “Self-driving car meets multi-access
edge computing for deep learning-based caching,” in 2019 International
Conference on Information Networking (ICOIN). IEEE, 2019, pp. 49–
54.
14 VOLUME 4, 2016
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2020.2983609, IEEE Access
[64] R. Q. Hu, L. Hanzo et al., “Twin-timescale artificial intelligence aided
mobility-aware edge caching and computing in vehicular networks,” IEEE
Transactions on Vehicular Technology, vol. 68, no. 4, pp. 3086–3099,
2019.
[65] S. Gurugopinath, P. C. Sofotasios, Y. Al-Hammadi, and S. Muhaidat,
“Cache-aided non-orthogonal multiple access for 5g-enabled vehicular
networks,” IEEE Transactions on Vehicular Technology, vol. 68, no. 9, pp.
8359–8371, 2019.
[66] Z. Ning, X. Wang, and J. Huang, “Mobile edge computing-enabled 5g
vehicular networks: Toward the integration of communication and com-
puting,” IEEE Vehicular Technology Magazine, vol. 14, no. 1, pp. 54–61,
2018.
[67] Z. Ning, X. Kong, F. Xia, W. Hou, and X. Wang, “Green and sustainable
cloud of things: Enabling collaborative edge computing,” IEEE Commu-
nications Magazine, vol. 57, no. 1, pp. 72–78, 2018.
[68] K. Zhang, Y. Zhu, S. Leng, Y. He, S. Maharjan, and Y. Zhang, “Deep
learning empowered task offloading for mobile edge computing in urban
informatics,” IEEE Internet of Things Journal, vol. 6, no. 5, pp. 7635–
7647, 2019.
[69] Z. Chen and X. Wang, “Decentralized computation offloading for multi-
user mobile edge computing: A deep reinforcement learning approach,”
arXiv preprint arXiv:1812.07394, 2018.
[70] J. A. Onieva, R. Rios, R. Roman, and J. Lopez, “Edge-assisted vehicular
networks security,” IEEE Internet of Things Journal, vol. 6, no. 5, pp.
8038–8045, 2019.
[71] L. Zhou, L. Yu, S. Du, H. Zhu, and C. Chen, “Achieving differentially pri-
vate location privacy in edge-assistant connected vehicles,” IEEE Internet
of Things Journal, vol. 6, no. 3, pp. 4472–4481, 2018.
[72] D. Li, Q. Yang, D. An, W. Yu, X. Yang, and X. Fu, “On location privacy-
preserving online double auction for electric vehicles in microgrids,” IEEE
Internet of Things Journal, vol. 6, no. 4, pp. 5902–5915, 2018.
[73] C. Chen, J. Wu, H. Lin, W. Chen, and Z. Zheng, “A secure and efficient
blockchain-based data trading approach for internet of vehicles,” IEEE
Transactions on Vehicular Technology, vol. 68, no. 9, pp. 9110–9121,
2019.
[74] X. Wang, Z. Ning, X. Hu, L. Wang, L. Guo, B. Hu, and X. Wu, “Future
communications and energy management in the internet of vehicles:
Toward intelligent energy-harvesting,” IEEE Wireless Communications,
vol. 26, no. 6, pp. 87–93, 2019.
[75] Z. Ning, J. Huang, X. Wang, J. J. Rodrigues, and L. Guo, “Mobile edge
computing-enabled internet of vehicles: Toward energy-efficient schedul-
ing,” IEEE Network, vol. 33, no. 5, pp. 198–205, 2019.
[76] J. James, W. Yu, and J. Gu, “Online vehicle routing with neural combinato-
rial optimization and deep reinforcement learning,” IEEE Transactions on
Intelligent Transportation Systems, vol. 20, no. 10, pp. 3806–3817, 2019.
[77] Z. Ning, F. Xia, X. Hu, Z. Chen, and M. S. Obaidat, “Social-oriented
adaptive transmission in opportunistic internet of smartphones,” IEEE
Transactions on Industrial Informatics, vol. 13, no. 2, pp. 810–820, 2016.
[78] C. Chen, Y. Ding, X. Xie, S. Zhang, Z. Wang, and L. Feng, “Trajcom-
pressor: an online map-matching-based trajectory compression framework
leveraging vehicle heading direction and change,” IEEE Transactions on
Intelligent Transportation Systems, 2019.
[79] Z. Ning, L. Liu, F. Xia, B. Jedari, I. Lee, and W. Zhang, “Cais: A
copy adjustable incentive scheme in community-based socially aware
networking,” IEEE Transactions on Vehicular Technology, vol. 66, no. 4,
pp. 3406–3419, 2016.
[80] M. Chen, W. Li, G. Fortino, Y. Hao, L. Hu, and I. Humar, “A dynamic
service migration mechanism in edge cognitive computing,” ACM Trans-
actions on Internet Technology (TOIT), vol. 19, no. 2, pp. 1–15, 2019.
HONGJING JI received the B.Sc. degrees in
Software Engineering in 2020 from the Taiyuan
University of Technology, Taiyuan, China. She
is currently working toward the M.Sc. degree in
School of Software, Dalian University of Technol-
ogy, Dalian, China. Her research interests include
edge computing, internet of vehicle and resource
management.
OSAMA ALFARRAJ received the master’s and
Ph.D. degrees in information and communication
technology (ICT) from Griffith University, in 2008
and 2013, respectively. His doctoral dissertation
investigates the factors influencing the develop-
ment of Government in Saudi Arabia, and it is
a qualitative investigation of the developers’ per-
spectives. He is currently an Associate Professor
with ICT, King Saud University, Riyadh, Saudi
Arabia. His research interests include electronic
commerce, M-government, the Internet of Things, cloud computing, AI, and
big data analytics.
AMR TOLBA received the M.Sc. and Ph.D.degrees
from the Faculty of Science, MenoufiaUniversity,
Egypt, in 2002 and 2006, respec-tively. He is cur-
rently an Associate Professor withthe Faculty of
Science, Menoufia University. Heis on leave from
Menoufia Univesity with theComputer Science
Department, Community Col-lege, King Saud
University, Saudi Arabia. He hasauthored/co-
authored over 30 scientific papers ininternational
journals and conference proceedings.His main
research interests include socially aware network, Internet ofThings, intel-
ligent systems, big data, recommender systems, and cloudcomputing. He
serves as a technical program committee member in severalconferences.
VOLUME 4, 2016 15
... Therefore, to realise the benefits of integrating AI into maternity services, concerted efforts and collaborations are required from key stakeholders, including governments, telecommunication providers, healthcare providers, and AI developers, to ensure the availability of the needed technological infrastructure and reliable internet connectivity, since investments in digital infrastructure and capacity building are necessary steps towards creating an enabling environment for AI technologies. 22 For instance, governments can collaborate with internet service providers to ensure that AI applications integrated into maternity services are zero-rated to warrant consistency of use. Also, developers of AI applications must consider the limited technological skills of clients and clinicians in Africa when developing AI ...
... Healthcare institutions, AI developers and international organisations can facilitate the development of training programmes to build the skills and capacity of healthcare workers and ensure local expertise in AI applications. 22 Considering that errors in data generation could lead to false predictions and misapplied interventions, 23 there is a need to enhance the skills and capacity of maternity care providers in Africa in data handling to avoid aggregated errors in data generation. ...
... For instance, the sensitivity of health data necessitates the need for robust measures to protect client information from authorised access and misuse. 22 Thus, to gain public trust and safeguard clients' privacy in the utilisation of AI in maternity services, it is important to establish stringent data protection regulations, including data access and management protocols, and ensure compliance. 22 ...
... As technology advanced, earlier criteria that define artificial intelligence became outdated. According to Ji (2020), technologies that enable Artificial Intelligence include: i. Computer vision enables computers to detect objects and people in photographs and photos. ...
Article
Full-text available
Artificial intelligence (AI), defined as the simulation of human intelligence by machines, enables automated data analysis, personalized marketing, and enhanced consumer interactions. This study explores the integration and effects of AI in advertising, revealing that while advertisers increasingly leverage AI technologies for efficiency and effectiveness, consumer awareness of AI's role remains limited and fragmented. This review was anchored on Technological Determinism theory. The research, using secondary data, highlights AI's capability to tailor advertisements, predict consumer behavior, and improve engagement through tools like chatbots and virtual assistants. Despite privacy concerns, AI's precise targeting and personalized content are revolutionizing advertising strategies, driving innovation and satisfaction. The study emphasizes the need for a comprehensive understanding of AI's applications and effects, suggesting that AI is a primary driver of change in advertising, influencing consumer behavior and industry practices. The ongoing evolution of AI technology is expected to further advance advertising techniques, reinforcing its role as a critical force in the digital age.
... Traditional approaches are limited in their ability to effectively manage such environments. In contrast, machine learning approaches have shown great potential for long-term cost optimisation by taking into account factors such as time cost, energy consumption and freshness of information [46,47,48]. Machine learning-based solutions enable wireless networks to improve their performance over time, using continuous learning to adapt to changing conditions and optimise network operations [49,50,51,52]. ...
... Traditional IoV Architecture[9] ...
Article
The race towards a better-connected world has led to innovations in network connectivity and the embedding of intelligence into various types of devices. Intelligent devices require an intelligent network infrastructure and design in order to reach their maximum potential. Recently, the concept of Open Radio Access Networks (Open RAN, O-RAN) has gained a lot of attention with the formation of the O-RAN Alliance which seeks to usher in the next phase of cellular network interoperability, intelligence and design. Another key concept that has emerged in recent years is the Internet of Vehicles (IoV) which has been described as a driver for next generation Intelligent Transportation Systems (ITS). This paper proposes an architecture that brings together these two concepts and explores how the design principles of the O-RAN can be incorporated into the IoV by using a bottom-up approach. It also discusses the benefits of an O-RAN-based IoV, some perceived challenges with its implementation and the way forward.
... AI provides a seamless in-vehicle experience and autonomous driving capabilities [144]. These advanced communication systems utilize AI algorithms to analyze and interpret data, enabling vehicles to communicate with each other and their surroundings [145]. This integration allows for improved safety features such as collision avoidance, adaptive cruise control, and lane-keeping assistance. ...
Article
Full-text available
Electric vehicles (EVs) have gained significant attention in recent years due to their potential to reduce greenhouse gas emissions and improve energy efficiency. An EV’s main source of power is its battery, which plays a crucial role in determining the vehicle’s overall performance and sustainability. The purpose of this paper is to examine the advancements in battery technology associated with EVs and the various charging standards applicable to EVs. Additionally, the most common types of automotive batteries are described and compared. Moreover, the application of artificial intelligence (AI) in EVs has been discussed. Finally, the challenges associated with EV battery development, as well as suggestions for improvement, are discussed. According to the study, Lithium-ion batteries are the most common in EVs due to their high energy density, long lifespan, and cost-effectiveness, despite their temperature sensitivity. Other battery types, like lead-acid and nickel-based, vary in efficiency, but are less commonly used in modern EVs. Solid-state batteries are seen as the future for their higher energy density and faster charging, though they face challenges like flammability. Wireless charging technology, still in development, promises superior convenience and sustainability than traditional methods. AI improves EV performance through enhanced battery management, autonomous driving, vehicle-to-grid communication, etc. Overcoming challenges like battery recycling, metal scarcity, and charging infrastructure will be crucial for the widespread adoption of EVs. This will be supported by government policies and battery technology innovations.
... In clustering, the samples within a particular cluster have a similarity in behavior while other clusters have different samples with different behavior. The conventional algorithms used in clustering are hierarchical clustering [50], k-means [51], spectrum clustering [52], and Dirichlet process [53]. Dimension reduction, also known as dimensionality reduction is a technique used in unsupervised learning to perform feature selection or extraction. ...
Article
Full-text available
Integrating Machine Learning (ML) techniques into Vehicular Adhoc Networks (VANETs) provides promising features in autonomous driving and ITS applications. In this paper, DSRC data is used to evaluate the effectiveness of different ML models, including Naive Bayes, Random Forest, KNN, and Gradient Boosting, in normal and adversarial scenarios. Since the dataset is relatively imbalanced, the Synthetic Minority Over-sampling Technique (SMOTE) is employed for sampling, and defensive distillation for improving model resilience to adversarial perturbations. From the results, it is clear that models such as Gradient Boosting and Random Forest show high accuracy in both cases, thus showing the potential of using Machine Learning to improve VANET security and reliability when new threats appear. Through this research, the significance of the application of ML in the protection of vehicular communication in order to enhance both traffic safety and flow has been articulated.
... Key features of DL include the use of multiple layers, hierarchical feature learning, end-to-end learning, scalability to handle large datasets, and automatic feature extraction. DL has accomplished notable achievement in a variety of fields, as well as in computer vision, NLP, recognition of speech, healthcare, and autonomous systems [6]. ...
Article
Cybersecurity questions of cyber-physical systems (CPS) have become ever more vital in recent times. Advances in technology and digitization have posed new weaknesses in CPS and cyber attackers take advantage of it. Viruses, malware, and sophisticated forms of cyberattacks have become a dangerous reality for critical infrastructure CPS. Lately, artificial intelligence (AI) technology has been extensively applied in the struggle against cyber threats in CPS. AI may improve system security by providing tools to quickly detect cyberthreats and automatically resolve them. Digitization of critical infrastructures (energy distribution networks, smart systems, oil and gas industry, water infrastructure, etc.) has increased their efficient management and at the same time, the number of cyber attackers on sensors, actuators, network and control equipment has also increased. Cyber security of critical objects can be ensured through AI. This article explores the theoretical foundations, application fields, and AI conceptual models. It analyzes the benefits and shortcomings of applying AI technologies to the security of the abovementioned systems. Mechanisms for detecting cyber threats in cyber-physical systems with the help of AI and predicting and preventing security threats are proposed.
Article
Artificial intelligence (AI) is becoming more important with transformative impacts in the automotive design process. The role of AI in automotive technology is being chiefly investigated from the perspectives of information technology, mechatronics, mechanical engineering, control science, and computer simulation. This paper discusses the application and potential challenges of AI from the viewpoints of interdisciplinary technologies within automotive engineering. We systematically present the trends and applications regarding the combination of AI and automotive engineering, and discuss the possible problems as a result of integrating these interdisciplinary fields. In addition to discussing these questions, a final key issue is the necessity of educating not only engineers and researchers but also all stakeholders in the automotive industry in these fields. The investigation of such interdisciplinary research will have significant impacts on the future of transformative re-engineering of the world. An overall scenario about the future will occur to the reader at the end of this paper. Future of the vehicles are being transformed from mechanical products into informatics systems. Practical deployment of onboard AI technology in new complex automotive environments demands novel functionality adjustments, adding machine learning processes able to cope with interpretability constraints, uncertainty, and energy efficiency. AI in vehicles has gained remarkable scientific attention as it has several applications in automotive in areas including intelligent systems, speech and image recognition, predictive intelligence, and recommendation systems. AI in automotive can transfer information that is initially in a physical form, such as a spoken command or a written text input, to a medium that can understand it logically or rationally, and thus allow the AI program to develop a response.(1)
Article
Rather than diminishing design to a mechanical process governed by utility, efficiency and economy, this paper conceptualises the architecture of the Basilica of La Sagrada Família in Barcelona as posthuman. The Cathedral’s social, cultural and psychological implications will be considered in this paper. La Sagrada Família is an “unfinished cathedral” in Catalonia built by the architect Antoni Gaudí in 1882. It receives over 3 million visitors a year. Considering La Sagrada Família under the posthumanist lens enables it to be intertwined with the environment and technology. Posthumanism involves delegating particular acts of human agency into technological devices. Using Kenneth Frampton’s concept of architecture dealing with the tension between its ‘representational’ and its ‘ontological’ dimensions, this article highlights how posthumanism can enable new design methods to be integrated with the Cathedral’s original ideas, casting a reassessment of existing theories of design. This article also considers other theories of design and architecture. These are Andrew Pickering’s concept of the ‘mangle,’ Jonathan Hale’s rematerialisation theory, Steve Tomasula’s concept of human scale in architecture, and Francesca Ferrando’s theory of philosophical posthumanism. The paper explores the process of bringing the original architectural idea of La Sagrada Família to expression in material reality. It analyses how the construction of increasingly sophisticated technological devices is a kind of collision and interaction between human goals, material resistance and future sustainability. Finally, the article demonstrates how the ‘dance of agency’—an ongoing, open-ended and temporally structured operation involving a dialectic of resistance and accommodation—can be carried out using the posthumanist theory of future sustainability in order to transform La Sagrada Família into a shared, plural, hopeful architecture that is embodied and entangled.
Preprint
Full-text available
This study discussion point of this paper is to make an in-depth analysis of the development impact of the Internet of Things combined with edge computing and artificial intelligence. In the analysis process, the importance and criticality of data processing and decision making of edge computing as well as the challenges faced should be elaborated respectively. With the rapid popularization and development of Internet of Things devices, edge computing has brought more innovative solutions for different application scenarios such as intelligent furniture industrialization, automatic driving and intelligent transportation by reducing the delay of processing data and improving the characteristics of security data, film and television. Resource and energy efficiency have certain limitations, so it is necessary to combine artificial intelligence to enhance edge computing devices, hardware accelerators, and its utility and federated learning technologies, which can effectively improve the performance and scalability of edge computing and promote the development of more self-service network systems for smart devices. The core of this study is how to promote the Internet through AI-driven edge computing to further develop and provide insights for research priorities and suggest related future research directions.
Article
Full-text available
Mobile edge computing (MEC) emerges recently as a promising solution to relieve resource-limited mobile devices from computation-intensive tasks, which enables devices to offload workloads to nearby MEC servers and improve the quality of computation experience. In this paper, an MEC enabled multi-user multi-input multi-output (MIMO) system with stochastic wireless channels and task arrivals is considered. In order to minimize long-term average computation cost in terms of power consumption and buffering delay at each user, a deep reinforcement learning (DRL)-based dynamic computation offloading strategy is investigated to build a scalable system with limited feedback. Specifically, a continuous action space-based DRL approach named deep deterministic policy gradient (DDPG) is adopted to learn decentralized computation offloading policies at all users respectively, where local execution and task offloading powers will be adaptively allocated according to each user’s local observation. Numerical results demonstrate that the proposed DDPG-based strategy can help each user learn an efficient dynamic offloading policy and also verify the superiority of its continuous power allocation capability to policies learned by conventional discrete action space-based reinforcement learning approaches like deep Q-network (DQN) as well as some other greedy strategies with reduced computation cost. Besides, power-delay tradeoff for computation offloading is also analyzed for both the DDPG-based and DQN-based strategies.
Article
Full-text available
The proliferation of IoTs beside the emergence of various cloud services push the horizon of edge computing. By offering cloud capabilities at the network edge closer to mobile devices, edge computing is a promising paradigm to resolve several vital challenges in IoTs, such as bandwidth saturation, energy constraints, low latency transmission, and data security and privacy. To provide a comprehensive understanding of edge computing supported by the integration of IoTs and cloud computing, that is, CoTs, this article first discusses some distinct research directions in CoTs with respect to edge computing. Given the significance of energy efficiency and sustainability of edge deployment in CoTs, we put forward a green and sustainable virtual network embedding framework for cooperative edge computing in wireless-optical broadband access networks. Specifically, we leverage a reliability function to confirm the number of backup edge devices, and embed virtual networks onto the suitable edge devices in CoTs. Finally, several research challenges and open issues are discussed.
Article
Full-text available
Recently, Internet of Vehicles (IoV) has become one of the most active research fields in both academic and industry, which exploits resources of vehicles and Road Side Units (RSUs) to execute various vehicular applications. Due to the increasing number of vehicles and the asymmetrical distribution of traffic flows, it is essential for the network operator to design intelligent offloading strategies to improve network performance and provide high-quality services for users. However, the lack of global information and the time-variety of IoVs make it challenging to perform effective offloading and caching decisions under long-term energy constraints of RSUs. Since Artificial Intelligence (AI) and machine learning can greatly enhance the intelligence and the performance of IoVs, we push AI inspired computing, caching and communication resources to the proximity of smart vehicles, which jointly enable RSU peer offloading, vehicle-to-RSU offloading and content caching in the IoV framework. A Mix Integer Non-Linear Programming (MINLP) problem is formulated to minimize total network delay, consisting of communication delay, computation delay, network congestion delay and content downloading delay of all users. Then, we develop an online multi-decision making scheme (named OMEN) by leveraging Lyapunov optimization method to solve the formulated problem, and prove that OMEN achieves near-optimal performance. Leveraging strong cognition of AI, we put forward an imitation learning enabled branch-and-bound solution in edge intelligent IoVs to speed up the problem solving process with few training samples. Experimental results based on real-world traffic data demonstrate that our proposed method outperforms other methods from various aspects. Index Terms-Internet of Vehicles, peer offloading, content caching, Lyapunov optimization, imitation learning. Z. Ning is with the Chongqing
Article
Full-text available
The prompt evolution of Internet of Medical Things (IoMT) promotes pervasive in-home health monitoring networks. However, excessive requirements of patients result in insufficient spectrum resources and communication overload. Mobile Edge Computing (MEC) enabled 5G health monitoring is conceived as a favorable paradigm to tackle such an obstacle. In this paper, we construct a cost-efficient in-home health monitoring system for IoMT by dividing it into two sub-networks, i.e., intra-Wireless Body Area Networks (WBANs) and beyond-WBANs. Highlighting the characteristics of IoMT, the cost of patients depends on medical criticality, Age of Information (AoI) and energy consumption. For intra-WBANs, a cooperative game is formulated to allocate the wireless channel resources. While for beyond-WBANs, considering the individual rationality and potential selfishness, a decentralized non-cooperative game is proposed to minimize the system-wide cost in IoMT. We prove that the proposed algorithm can reach a Nash equilibrium. In addition, the upper bound of the algorithm time complexity and the number of patients benefiting from MEC is theoretically derived. Performance evaluations demonstrate the effectiveness of our proposed algorithm with respect to the system-wide cost and the number of patients benefiting from MEC.
Article
With the significant advance of connected vehicles, future demand for vehicular infotainment services will be greatly increased. Traditional content distribution approaches based on typical cellular architecture suffer from long latency and unstable connections in high-dynamic vehicular environments and even may cause congestion on the backhaul network due to a large amount of requested data. In this paper, we propose a novel content delivery framework by leveraging the 5G edge networks, in which the content caching and data prefetching techniques are exploited accordingly. We investigate the comprehensive dynamic link utilization problem in 5G edge networks from the perspectives of vehicle users and network operator, respectively. For the vehicle users' perspective, our aim is to maximize the vehicular content distribution throughput through optimal scheduling of vehicular access link slots, and also the utilization problem of backhaul link slots in edge networks is studied to reduce the data access delay of vehicles. For the network operator's perspective, the objective is to maximize the total profit of the network operator, and therefore, the auction model is utilized for the vehicular access link slot scheduling and the backhaul link utilization is analyzed in terms of compensations and costs. Finally, extensive simulations are conducted to demonstrate the efficiency of the proposed solutions for edge content delivery of connected vehicles.
Article
Caching is an effective way to address the challenges due to explosive data traffic growth and massive device connectivity in fifth-generation (5G) networks. Currently, few works on caching pay attention to the impact of the time duration for which content is stored, called retention time, on caching optimization. The research on retention time is motivated by two practical issues, i.e., flash memory damage and storage rental cost in cloud networks, together giving rise to the storage cost. How to optimize caching contents taking the storage cost into consideration is a challenging problem, especially for the scenarios with cache-enabled mobile nodes. In this paper, a retention-aware caching problem (RACP) in vehicular networks is formulated, considering the impact of the storage cost. The problem's complexity analysis is provided. For symmetric cases, an optimal dynamic programming (DP) algorithm with polynomial time complexity is derived. For general cases, a low complexity and effective retention aware multi-helper caching algorithm (RAMA) is proposed. Numerical results are used to verify the effectiveness of the algorithms.
Article
The rapid development of IoMT facilitates pervasive healthcare networks. However, numerous MUs lead to deficient wireless channel and computation resources. Edge computing based healthcare system is appreciated as a satisfactory paradigm to solve this obstacle. Based on the physical bound of WBANs, the IoMT is separated into two sub-networks, that is, intra-WBANs and beyond-WBANs. Considering the characteristics of healthcare systems, the cost of MUs relies on medical urgency, AoI and energy dissipation. In intra-WBANs, the wireless channel resource allocation problem is modeled by a cooperative game. The Nash bargaining solution is utilized to obtain the unique Pareto optimal point. In beyond-WBANs, MUs are considered to be rational and potentially selfish. Thus, a non-cooperative game is formulated to minimize the system-wide cost. To demonstrate the effectiveness of our proposed solution, performance evaluations are conducted with respect to the system-wide cost and the number of MUs benefiting from edge computing. Finally, several research challenges and open issues are discussed for further work.
Article
Mobile network caching (MNC) and Coordinated MultiPoint (CoMP) joint transmission (JT) techniques are expected to play a significant role in future networks. Yet, when considering caching in a CoMP-integrated network where both single transmission (ST) and JT are permitted, some practical challenges need to be tackled. In this paper, we propose a cooperative online caching strategy to realize autonomous transmission-aware content caching and updating for video content delivery. Different from existing works, this paper considers practical time-varying user request patterns, and exploits storage-level and transmission-level cooperation to jointly optimize content caching under dynamic user demands for JT and ST. The former helps reduce caching redundancy and the latter helps BSs create CoMP-JT opportunities to improve cell-edge throughput. We formulate the content caching problem to be a Markov decision process (MDP), in which the system reward is defined as the delay reduction after performing a caching action. The goal is to maximize the system reward, i.e., to minimize content delivery delay. Then, we develop a reinforcement learning (RL)-based online learning algorithm to search the optimal caching policy. To overcome the "curse of dimensionality", the designed RL-based algorithm is extended with linear approximation, namely approximate Q-learning (AQL) algorithm.} The system performance in terms of delay and cache hit rate is evaluated in the simulations, and the results demonstrate the effectiveness of the proposed strategy.
Article
With the rapid development of the internet of things (IoT), connected vehicles are set to become a huge industry over the next few years. In this study, we take an investigation of the distributed intelligent traffic system by pushing intelligence into connected vehicles in terms of dynamic decision making for traversing a certain area (e.g., roundabout and intersection). In particular, we propose a model for the next generation of intelligent transportation system, which focuses on dynamic decision making of connected vehicles based on Ant Colony Optimization, a typical Swarm Intelligence (SI)-based algorithm. Specifically, we first present a communication framework among connected vehicles for sharing information of traffic flow. Then, by applying the concept of SI, connected vehicles are regarded as artificial ants which are able to self-calculate to make an adaptive decision following the dynamics of traffic flow. Furthermore, for evaluating the effectiveness of the proposed approach, we have constructed a framework to model and simulate the traffic system in IoT environment. Simulations with different scenarios of transportation systems indicate promising results comparing with previous works.