Optimisation de la mobilité des véhicules autonomes pour un système de transport en commun flexible et adaptatif = Tối ưu hóa di chuyển của xe tự lái cho hệ thống vận tải công cộng mềm dẻo
Motivation et Contexte
Intelligent transportation systems (ITS) aim to enhance road safety, manage traffic congestion, and achieve positive economic, energy, and environmental outcomes by equipping vehicles with wireless communication technologies Utilizing wireless communication systems like IEEE 802.11p, vehicles can exchange information with nearby vehicles and road infrastructure A robust wireless communication network serves as the foundation for connected transportation systems.
Reliable and transparent data communication between vehicles (V2V) and between vehicles and infrastructure (V2I) is crucial for connected vehicle technology applications Autonomous vehicles represent a significant innovation that will transform both the industry and society As part of Intelligent Transportation Systems (ITS), optimizing the mobility of autonomous vehicles begins with studying Vehicular Ad Hoc Networks (VANETs) and the V2V and V2I communication processes This project focuses on enhancing communication between vehicles and infrastructure, with a key advantage being the ability to assess demand for mobility models, thus providing optimal network resources for effective resource management By optimizing V2I demand, dynamic resource allocation can be achieved, particularly during peak mobility times.
PROJET ET PROBLÉMATIQUE
Généralité sur le projet
Autonomous vehicles (AVs) are set to disrupt urban transportation significantly Numerous studies indicate that replacing traditional manual vehicles with AVs in current transport services can enhance efficiency However, the impact of AVs will extend beyond mere efficiency gains, transforming the entire transportation ecosystem While ridesharing systems are often seen as competitors to public transport, we envision a scenario where AVs are integrated into public transport offerings, making it more flexible and responsive to passenger demand rather than adhering to fixed routes.
Objectif général du projet
The primary objective of the project is to identify the optimal structure for future public transportation systems that integrates both flexible and fixed modes, selecting one or the other based on the time of day and geographical area while rethinking current routes accordingly Resource management optimization must consider traffic saturation, vehicle availability, and the dynamic nature of demand throughout the day By effectively assessing real-time demand, resource status, and availability, it becomes possible to explore regulatory solutions and incentive policies to enhance service for users.
Machine learning techniques should be employed to analyze traveler mobility patterns and resource concentrations in relation to evolving needs These methods can provide estimates for usage trends and propose algorithms to optimize the deployment of autonomous vehicles (AVs) based on time of day and geographical area Additionally, it can trigger specific, calculated offers designed to enhance user engagement and improve service offerings.
— Minimiser le cỏt de déploiement des véhicules autonomes (AVS)
— Optimiser leur temps d’utilisation et leur disponibilité
— Tout en maximisant la satisfaction des voyageurs
Objectifs spécifiques et Problématique
Pour atteindre ces trois objectifs, il est indispensable de faire une étude sur les réseaux véhiculaires, en passant par les communications entre véhicules et infrastructures.
— l’étude des performances au travers des métriques de connectivité probabiliste dans un réseau véhiculaire ;
— la prédiction de la demande V2I par application des techniques de machine learning ;
The project aims to analyze the impacts of road traffic on vehicle-infrastructure communication, seeking to develop a comprehensive understanding of the communication system to optimize network resource allocation.
Conclusion
In this chapter, we introduced intelligent transportation systems, outlining the work context, primary and specific objectives, and the focus of our project along with the associated challenges We emphasized a particular mode of communication within intelligent transportation systems: vehicle-infrastructure communication, which is the basis of our study The literature reveals various mobility models within vehicular networks In the next chapter, we will describe a range of mobility models, vehicular networks, and the technologies employed for dynamic demand adaptation These definitions will lead us to characterize related approaches, refining our selection of the communication method to be studied in our research.
Études sur les modèles de mobilités dans un système de transport intelligent 5
Bref aperỗu sur la mobilitộ
Mobility is a crucial aspect of daily life that has been enhanced through digitalization, leading to the emergence of smart mobility over the past three decades This intelligent mobility is evolving within our cities, significantly improving the daily experiences of users As a result, various models of mobility can be identified based on user perspectives.
— la mobilité partagée (covoiturage et auto-partage)
Each mobility model has specific needs and required resources for optimal functionality Various impacts on infrastructure are reported concerning each mobility user.
According to Alonso-Mora et al (2017), ridesharing systems utilizing autonomous vehicles enable efficient and reliable urban mobility on demand This approach supports a dynamic vehicle routing method that leverages historical data to enhance the performance of an autonomous vehicle network The article discusses how to effectively route vehicles and allocate requests while incorporating future demand predictions, presenting an optimization method that intelligently assigns requests to autonomous vehicles to minimize the expected costs associated with meeting both current and future transportation needs.
The article focuses exclusively on carpooling, disregarding other mobility models This approach may lead to increased distances traveled by each vehicle and can be more expensive when calculating reductions in travel time.
Besoins des usagers et impacts de la mobilité sur l’infrastructure 6
Various needs and impacts are associated with each mobility model; for instance, in the case of autonomous mobility, we consider the autonomous vehicle Key requirements for users of autonomous vehicles include enhanced safety, seamless user experience, and efficient navigation systems.
— Réduire les congestions et donc les émissions de polluants dues au trafic ;
Les impacts de véhicules autonomes sur l’infrastructure sont multiples, tels que :
— Réduction du nombre d’infrastructures routières sur le long terme ;
— Optimiser la demande entre l’infrastructure et les véhicules ;
— Fluidifier le trafic, de réduire le nombre d’accidents graves,
According to Fửldes and Csiszỏr (2018), autonomous vehicles are transforming passenger transport by enabling the emergence of demand-responsive, shared mobility services, particularly in urban areas These services require flexibility in scheduling and capacity to adapt to current demand levels To enhance user acceptance and provide highly personalized services, it is essential to understand the type of information needed for planning various service types that align with user expectations and operational constraints.
The authors have developed various methodologies, including the definition and characterization of shared transport services utilizing small-capacity autonomous vehicles (AVs) They identified planning functions that consider the characteristics of these service types and modeled the planning information system, along with determining data collection methods and user expectation processing However, the article by Fửldes and Csiszỏr (2018) highlights several limitations, such as the failure to account for the dynamic nature of data for input formation, neglecting user demand offers during peak hours, and only considering individual mobility without addressing other mobility models.
Les Ressources réseaux pour l’optimisation de besoins
Mobility models are diverse, but our project focuses on a network resource model that optimally addresses these needs We highlight essential network resources required for vehicular networks in general, with a particular emphasis on autonomous vehicles.
— Avoir une bande passante considérable
— Une connexion en temps réel entre les infrastructures et des terminaux embarqués dans l’habitacle des véhicules ;
— Niveau de sécurité des équipements ;
— Ressources en communications véhiculaires (V2X : Vehicle to Everything) opti- males ;
Les réseaux véhiculaires
Généralité
Over the past decade, research has focused on developing an intelligent transportation system (ITS) that enhances user experience, improves traffic efficiency, and promotes a safe driving environment Vehicle networks can leverage a fully operational ITS that supports various applications related to vehicle safety and traffic management As the number of connected vehicles continues to rise, thousands of vehicles will transmit data to report traffic conditions and enhance driving safety This surge necessitates improved network capacity and imposes strict requirements for low latency In this context, 5G technology is proposed as a solution to meet the demanding needs of vehicle networks.
Système de transport intelligent(STI)
Intelligent transportation systems (ITS) facilitate data exchange between vehicles and road infrastructure, enhancing driving safety A key feature of ITS is the ability to share specific information among vehicles, supporting a wide range of vehicular applications To ensure effective communication, ITS components, including vehicles and pedestrians, are equipped with network interfaces, sensors, onboard computing capabilities, and video cameras This enables infrastructure elements to collaborate with vehicles in the communication process Vehicle networks are fundamental to ITS, leveraging information and communication technologies to improve safety, reliability, efficiency, and service quality The goal is to provide ubiquitous network connectivity for moving vehicles while meeting stringent requirements for latency and reliability in various ITS services.
Intelligent Transportation System (ITS) services have specific network requirements According to Toufga et al (2018), these service requirements can be categorized into safety-related needs, which demand very low latency and high reliability, and non-safety-related needs, which may have lower reliability and less stringent latency requirements Each service within the ITS generates traffic, which is defined by the traffic associated with that particular service.
— Un comportement (périodique ou non-périodique) : heures de pointe ;
— un mode de transmission (unicast, broadcast) ;
— une fréquence minimale pour les messages périodiques ;
— un niveau de transmission de la fiabilité
Les réseaux véhiculaires
VANETs (Vehicular Ad hoc Networks) represent an innovative type of mobile ad hoc network (MANET) that facilitates communication between vehicles and roadside infrastructure Unlike traditional ad hoc networks, VANETs are distinguished by the high mobility of their nodes, resulting in a highly dynamic network topology.
To achieve a high level of security, vehicles must communicate with all nearby elements of the Intelligent Transportation System (ITS) This is where Vehicle-to-Everything (V2X) communication comes into play, encompassing various modes such as Vehicle-to-Vehicle (V2V), Vehicle-to-Infrastructure (V2I), Vehicle-to-Network (V2N), and Vehicle-to-Pedestrian (V2P).
Le mode de communication véhicule à infrastructure est l’objet de notre travail.
Figure 2.1 – Mode de Communication dans VANET[Ang et al., 2019]
Vehicle-to-Vehicle (V2V) communication enables nearby vehicles to exchange critical information such as speed, direction, and braking status This technology allows vehicles to communicate directly with one another or through intermediary vehicles, enhancing road safety and traffic efficiency.
Vehicle-to-Infrastructure (V2I) technology enables communication between vehicles and fixed road infrastructure, enhancing traffic efficiency by facilitating the exchange of information regarding lane markings, traffic signs, and signal timings Additionally, V2I plays a crucial role in improving urgent road safety messages by providing real-time updates on roadway accidents and construction zones Our focus is on V2I communication, as it presents various challenges related to resource management and the demand between vehicles and infrastructure.
Vehicle-to-Network (V2N) technology enables vehicles to communicate with a dedicated server that supports V2N applications, known as the Vehicle-to-Everything (V2X) Application Server (AS) This server delivers crucial information regarding traffic distribution, road conditions, and various services, enhancing the overall driving experience.
— Véhicule à piéton (V2P):Le V2Ppermet l’échange de messages entre les véhicules et les piộtons qui envoient et reỗoivent des messages à l’aide de leur tộlộphone ou d’autres appareils portables sans fil.
2.2.3.2 Composants de réseaux des véhicules
Three essential components for enabling V2X communication, as defined in the works of Al-Sultan et al (2014) and Martinez et al (2011), include the Application Unit (AU), the On-Board Unit (OBU), and the Road Side Unit (RSU).
The On Board Unit (OBU) is a communication device installed in vehicles that facilitates the exchange of information with Road Side Units (RSU) or other OBUs It features a resource management processor, an interface for connecting to additional OBUs, and a short-range wireless communication network device.
The Application Unit (AU) is a device installed in vehicles specifically for security applications It can connect to the On-Board Unit (OBU) through either wired or wireless connections and communicates with the network solely via the OBU.
Roadside Unit (RSU) is a fixed point located along the road or in designated areas like intersections RSUs facilitate communication between vehicles and the Internet Their primary function is to execute safety applications, such as accident alerts and work zone notifications, through Vehicle-to-Infrastructure (V2I) communication.
Figure 2.2 – Composants du réseau de véhicules[Al-Sultan et al., 2014]
In 2002, the Federal Communications Commission (FCC) of the United States established the IEEE 802.11p standard, known as Dedicated Short Range Communications (DSRC) Meanwhile, the European Telecommunications Standards Institute (ETSI) defined ITS-G5, which outlines the protocols and requirements for the European version of this technology ITS-G5 is based on DSRC with some architectural modifications DSRC specifies the physical transmission and media access control (MAC), derived from the older IEEE 802.11a standard and tailored for Vehicle-to-Everything (V2X) communication needs The FCC allocated 75 MHz of spectrum in the 5.9 GHz range for DSRC, dividing it into 10 MHz channels and employing Orthogonal Frequency Division Multiplexing (OFDM) for efficient data transmission.
Figure 2.3 – Pile de protocoles et normes de base associées pour le DSRC aux États-Unis [Sayadi et al., 2016]
Comme DSRC, ITS-G5 fonctionne dans la bande 5,9 GHz, tandis que l’attribution européenne du spectre est subdivisée en parties A à D Principalement, ITS-G5A avec
The primary frequency band dedicated to safety and traffic efficiency applications is 30 MHz, while ITS-G5B allocates 20 MHz for non-safety-related applications At the PHY layer, ITS-G5 employs OFDM, and at the MAC layer, it utilizes CSMA/CA A notable feature of this protocol, distinct from DSRC, is the use of geographic coordinates for addressing and transmission, ensuring that all vehicles within a specific geographical area can receive broadcasted information.
In September 2016, 3GPP released the first version of Release 14, introducing support for V2X communications, commonly known as LTE-V, LTE-V2X, or cellular V2X Compared to IEEE 802.11p, LTE-V enhances the link budget in the physical layer and incorporates redundant transmission, which significantly boosts reliability.
The LTE-V standard features two radio interfaces: the Uu interface, which supports Vehicle-to-Infrastructure (V2I) communications, and the PC5 interface, designed for Vehicle-to-Vehicle (V2V) communications via a direct LTE sidelink Introduced in version 12, the PC5 interface enhances public safety and operates in two modes: mode 1 and mode 2.
2 Les deux modes sont conỗus avec l’objectif principal d’augmenter la durộe de vie de la batterie des appareils mobiles réalisés au prix d’une augmentation de la latence.
Les véhicules connectés nécessitent des communications V2X à faible latence ; par conséquent, les modes 1 et 2 ne conviennent pas aux applications véhiculaires.
Figure 2.4 – Interfaces radio[Mouawad et al., 2020]
Vehicular networks possess unique characteristics that set them apart from mobile ad hoc networks, and these features must be considered when designing protocols for VANETs This section highlights several properties and constraints associated with this type of network.
The energy capacity and storage in VANET networks differ significantly from MANET networks, where energy constraints pose challenges for researchers In VANETs, vehicles have ample energy resources, which can effectively power various electronic devices within smart cars.
Donc, les nœuds sont censés avoir une grande capacité de traitement et de stockage de données.
Software Defined Networking
Introduction
Traditional network functionalities are primarily based on a model where routers integrate both control and data logic Each network device features a control plane to build a forwarding table and a data plane to transmit traffic according to the rules of that table This architecture leads to significant computational load, as it requires programming a large topology; each network device must be individually configured, making traditional IP networks complex and challenging to manage.
An alternative paradigm is currently capturing the attention of an increasing number of vertical sectors: Software Defined Networking (SDN) SDN is a promising technology that paves the way for a new software-based and programmable network architecture.
Il utilise principalement openflow comme protocole de communication.
SDN is an emerging paradigm that enables innovative network system design Its dual roles include facilitating network programmability and enhancing network control and management.
Traditionally, network devices like switches and routers are designed with the intelligence to manage traffic in relation to adjacent devices, resulting in a distributed intelligence model In contrast, Software-Defined Networking (SDN) offers an architectural principle where network control and management are centralized and decoupled from the data plane, as illustrated in Figure 2.5.
Figure 2.5 – Software Defined Networking vs Réseau traditionnel [Farinacci et al., 2013]
SDN reproduit la conception d’un routeur traditionnel à travers deux éléments diffé- rents :
1 un contrôleur responsable de la mise à jour des tables de routage des équipements réseau,
2 Commutateurs qui transfèrent les paquets d’après les règles spécifiées par le contrô- leur.
It is important to highlight that the proposed decoupling of control and data planes in SDN architecture is not a new concept; it is a fundamental aspect of MPLS (Multiprotocol Label Switching) technology The key distinction between MPLS and SDN lies in SDN's focus on providing programming interfaces within the network topology.
Architecture SDN
According to the Open Networking Foundation (ONF), the SDN architecture consists of three layers: the application layer, the control layer, and the data layer Each layer communicates with its adjacent layer using appropriate NorthBound (NBI) or SouthBound (SBI) interfaces This article examines the ONF-proposed SDN architecture, highlighting its key components as outlined by Farinacci et al (2013).
The data plane consists of network components such as routers and switches that are responsible for processing and transmitting data traffic Its primary function is to forward packets to the next hop based on predefined rules.
The control plane consists of SDN controllers that have exclusive control over the data plane elements Each SDN controller acts as a logically centralized entity, translating the requirements of the SDN application plane into the data plane and offering SDN applications an abstract view of the network.
The application plane consists of one or more applications that communicate their network requirements to the SDN controller through the NBI interface This layer deploys various SDN applications, including network topology discovery, resource provisioning, and path reservation.
The SDN Southbound Interface (SBI) serves as the defined link between an SDN controller and data plane elements, enabling programmatic control over data transfer operations, statistical reporting, and event notifications This interface allows the SDN controller to make dynamic adjustments based on real-time network demands The most commonly used southbound interface is OpenFlow, which will be discussed in detail in the following section.
The Northbound Interfaces (NBI) in Software-Defined Networking (SDN) serve as crucial connections between SDN applications and SDN controllers These interfaces offer abstract representations of the network, enabling direct expression of network behavior and requirements.
The management and administration plan addresses static tasks that are more effectively handled outside of application, control, and data plans This includes managing business relationships between suppliers and clients, allocating resources to customers, and configuring physical equipment.
Figure 2.6 – Composants architecturaux SDN et leurs interactions[Specification, 2013]
Protocole Openflow
The most commonly used protocol for communication between network devices and controllers is OpenFlow, as standardized by the Open Networking Foundation (ONF) and widely adopted by many network vendors While the terms OpenFlow and Software-Defined Networking (SDN) are often used interchangeably, it is crucial to distinguish between them SDN refers to the decoupling of control and data planes, whereas OpenFlow is a protocol that enables SDN controllers to remotely manage and configure OpenFlow switches, allowing for the monitoring of network device states and the collection of traffic statistics Developed by the ONF, the OpenFlow protocol serves as an industry standard that outlines how an SDN controller should interact with the data plane, adapting the network to meet evolving business needs To enhance network responsiveness to traffic demands, the SDN controller can install forwarding rules in data plane elements via the OpenFlow protocol.
Il est à noter que Openflow n’est pas le seul protocole disponible utilisé pour l’interface
Southbound SDN protocols include various options such as LISP, Forces, and SoftRouter However, OpenFlow stands out as the most popular and thoroughly tested open-source protocol for the southbound SDN interface.
Travaux connexes
Numerous research studies have addressed the optimization of V2X communication demand, transitioning from static to dynamic demand estimation while considering peak hours and utilizing Software Defined Networking (SDN) This approach aims to adapt dynamically based on network traffic and mobility, ultimately facilitating an intelligent transportation system (ITS) with excellent Quality of Service (QoS), reliability, and minimized latency Aujla et al (2018) outline the anticipated challenges in intelligent transportation systems, including high mobility, minimal latency, real-time services, and high QoS This article presents two key concepts to tackle these challenges in modern ITS: the first is Software Defined Networking, which manages network resources through decoupled data and control planes, and the second is Edge Computing, which optimizes data processing at the network's edge.
The article by Aujla et al (2018) introduces a Software-Defined Networking (SDN) framework for audiovisual vehicles that addresses mobility and Quality of Service (QoS) issues in vehicular networks However, it overlooks the detailed demand during peak hours, where traffic congestion is significant, which is crucial for optimizing mobility Additionally, the article lacks insights on how mobility considerations relate to the types of services users require, aiming to enhance their usage time and availability.
According to Mouawad et al (2019), V2X communication presents unique challenges for 5G systems, with network slicing emerging as a leading solution Network slicing offers several advantages, including the creation of isolated virtual networks, allowing different traffic flows to operate independently The article addresses the mobility management issues within the V2X network slicing environment The authors introduce the SSF-V2X (Slice Selection Function-V2X) approach, along with an algorithm focused on inter-slice handover Their implementation relies on a slice selection function tailored for V2X use cases (SSF-V2X algorithm) However, the article does not define the demand within V2X communication, which is crucial for adapting networks.
The author [Mouawad et al., 2019] highlights that V2X (Vehicle to Everything) communication, which includes V2V (Vehicle to Vehicle) and V2I (Vehicle to Infrastructure), enables real-time interaction between vehicles and their environment, enhancing road safety, traffic management, and entertainment information delivery The article addresses challenges related to mobility management and improving Quality of Service (QoS) in V2X communication It proposes an innovative solution leveraging 5G concepts such as network slicing and Software Defined Networking (SDN), featuring a suitable network selection algorithm along with a routing algorithm and a handover mechanism.
The author proposes a vehicle architecture for mobility management based on Software Defined Networking (SDN), focusing on essential mobility elements and the Quality of Service (QoS) requirements for Vehicle-to-Everything (V2X) applications, particularly addressing latency constraints A Dynamic Optimal SDN Topology (DOST) controller is introduced to minimize communication latency between the control and data planes Additionally, the integration of SDN with network slicing is suggested to enhance network performance and ensure QoS guarantees This framework includes an orchestration layer responsible for resource management across slices and mobility management within a network slicing environment for V2X However, the article does not consider Vehicle-to-Infrastructure (V2I) communication demands or performance metrics [Mouawad et al., 2019].
According to Dutra et al (2017), the use of Software Defined Networking (SDN) technology enhances network management capabilities and enables control plane programmability by abstracting the underlying network infrastructure for applications and services through the OpenFlow protocol The authors address key issues such as managing end-to-end Quality of Service (QoS) based on queue support in OpenFlow, optimally allocating network resources according to user demands, and eliminating the necessity for over-provisioning.
Dutra et al (2017) propose a multi-path routing approach combined with precise bandwidth allocation through SDN quality of service support, ensuring end-to-end service quality while efficiently managing the use of Open vSwitch (OVS) virtual switches This strategy not only guarantees the required service quality but also reduces costs by minimizing the number of OVS needed Open vSwitch aims to provide a virtual switch with functionalities comparable to a physical managed switch, including features like NetFlow, RSPAN, ERSPAN, and a command-line interface similar to IOS, while being scalable across multiple physical servers in a virtualized environment (Quality, 2018).
The study by Dutra et al (2017) demonstrates that it is possible to configure a network that provides QoS guarantees while minimizing the number of active OVS in the network The computational cost for finding a solution increases linearly with the number of endpoint nodes and shows a positive correlation with the number of users However, the article does not address resource allocations considering peak hours, user demands, or the volume of requests between infrastructures for resource distribution.
In [Neggaz and Toufga, 2018], vehicular networks are identified as a cornerstone of Intelligent Transportation Systems (ITS) due to their ability to provide ubiquitous network connectivity to moving vehicles while supporting various ITS services These networks also face stringent requirements regarding latency and reliability The authors address the challenges related to managing the quality of service provided by vehicular networks and the additional services offered by Intelligent Transportation Systems.
Un seul réseau est chargé de prendre en charge chaque service dans un STI.
The article by Neggaz and Toufga (2018) presents a hybrid network architecture based on Software-Defined Networking (SDN), aimed at jointly controlling networks that ensure connectivity for multi-hosted vehicles while exploring the quality of service opportunities offered by such an architecture However, it overlooks the dynamic demands of users, including the specific resource requirements for each service, end-to-end delay management, and lacks details on mobility management.
Subsequently, the authors [Dey et al., 2016] indicate that connected vehicle technology (CVT) relies on wireless data transmission between vehicles (V2V) and between vehicles and infrastructure (V2I) Evaluating the performance of various network options for V2V and V2I communication, which ensures optimal resource utilization, is a prerequisite for designing and developing robust wireless networks for CVT applications.
While Dedicated Short Range Communication (DSRC) has been regarded as the primary communication option for Vehicle-to-Transportation (V2T) safety applications, the use of alternative wireless technologies such as Wi-Fi, LTE, and WiMAX enables longer-range communications and higher throughput requirements that DSRC alone cannot support This article proposes an application layer transfer method to facilitate Het-Net communication for two V2T applications: traffic data collection and collision warning However, it does not address the network resource demands between Vehicle-to-Infrastructure (V2I) and Vehicle-to-Vehicle (V2V) communication.
The article by Jia and Ngoduy (2016) highlights that vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications are emerging components of intelligent transportation systems (ITS), enabling cooperative driving and significantly enhancing traffic efficiency However, the high mobility of vehicles can lead to unreliable vehicular communications, such as packet loss and transmission delays, which may negatively impact the performance of cooperative driving systems (CDS).
The article employs the IEEE 802.11p standard, the de facto protocol for vehicle networking, alongside road sensors that gather average speed data in the targeted area to serve as a reference for downstream traffic However, it fails to address the resource demands in Vehicle-to-Infrastructure (V2I) and Vehicle-to-Vehicle (V2V) communication.
Conclusion
We presented a study on mobility models within an intelligent transport system, defining the concept of vehicular networks and detailing the types of communications utilized We highlighted the significance of Software-Defined Networking technology for the dynamic adaptation of network resource demand, including its OpenFlow protocol Additionally, we specified the scope of our application by reviewing related literature relevant to our approach Building on the technologies and literature reviewed, the following chapter will outline the selected characteristics for implementing our proposed approach, the architecture of our work, the tools for its execution, and an analysis of machine learning techniques for quantitative communication between vehicles and infrastructure.
Aperỗu sur l’approche proposộe
Choix de la plate-forme et des outils
The study of demand in Vehicle-to-Infrastructure (V2I) wireless communication networks utilizes two types of simulators: traffic simulators and network simulators The traffic simulator generates vehicle mobility on a map, while the network simulator models the interactions between various network entities and the network infrastructure This article will introduce the traffic simulator SUMO and two commonly used network simulators, NS2 and OMNET++, which are employed to simulate Vehicular Ad hoc Networks (VANETs).
3.1.1.1 Le simulateur de trafic routier : SUMO
SUMO (Simulation of Urban Mobility) is an open-source traffic simulation software licensed under the GNU Public License (GPL), designed for managing large road networks It models both urban and highway environments, providing a comprehensive tool for traffic simulation The SUMO software suite includes various applications that assist in preparing and executing traffic scenario simulations In this section, we will present the different applications included in the SUMO suite based on their specific areas of study [Krajzewicz et al., 2012].
The SUMO road network includes traffic light plans and connections between lanes at intersections These road networks can be generated directly using the ônetgenerate application or imported from OpenStreetMap road map files via the ônetconvert application.
De plus, netconvert permet de lire les fichiers XML représentant un graphe réseau routier.
There are five types of XML files that describe nodes, routes, route types, connections between roads, and traffic light plans The applications netgenerate and netconvert utilize the same library to generate or import road networks into the simulator Both applications employ heuristic methods to correct missing data in XML files The process of generating a road network is summarized in Figure 3.2.
Figure 3.2 – Procédure commune de préparation du réseau dans netconvert et netgenerate [Krajzewicz et al., 2012]
The SUMO software performs discrete-time simulations, with a default time step of one second, which can be reduced to as low as one millisecond In the simulator, time is measured in microseconds, and each vehicle's position is determined by its location on the road from the start of its journey The speed of each vehicle, measured in meters per second (m/s), is calculated using an extension of the stochastic model SUMO offers two versions for simulating road traffic.
— Sumo : Une application qui permet d’exécuter les simulations en ligne de com- mande ;
Sumo-gui is an application designed to simulate road traffic through a graphical interface that utilizes OpenGL Users can customize the visualization in various ways, including managing vehicle speeds, controlling wait times, and interacting with traffic signal programs.
The SUMO simulator features an integrated socket known as Traci (Traffic Control Interface), which enables interaction with external applications To facilitate online interaction, SUMO must be launched with an additional option that specifies the network port number This online interaction allows for real-time modifications of simulation objects, including intersections, roads, lanes, traffic lights, and vehicle speeds.
3.1.1.2 Le simulateur réseau NS2(NS3)
NS2 (and NS3) is a network simulation software designed with object-oriented principles, code reuse, and modularity in mind It extends the Tcl (Tool Command Language) programming language to control applications described within NS2 (NS3) Users engage with the simulator through three main steps: programming the network topology and component behavior, running the simulation, and interpreting the results The final step is facilitated by an auxiliary tool called NAM (Network Animator), which visualizes the simulated elements NAM serves two primary purposes: it represents the network topology described in NS2 (NS3) and temporally displays the results from a NS2 (NS3) execution trace.
A Les composants de NS2(NS3)
The NS2 (NS3) simulator is designed for packet-switched networks and supports large-scale simulations It includes essential features for studying unicast and multicast routing algorithms, transport protocols, session protocols, reservation protocols, integrated services, and application services A comprehensive list of the main components available in the NS2 (NS3) simulator is provided in Table 4.2.
Table 3.2 Liste des principaux composants disponible dans NS2, [Lin et al., 2007]
Application web, FTP, Telnet, Générateur de trafic(CBR)
Transport RTP, SRM, UDP, TCP
Routage Statique, dynamique, vecteur de distance, routage MP Gestion des files d’attentes RED, DROP Tail, Token Bucket
Discipline de service CBQ, SFQ, DRR, Fair queueing
Système de synchronisation CSMA/CD, CSMA/CA, lien point to point
OMNET++ is an object-oriented discrete event simulator based on C++, designed for simulating communication networks, multiprocessor systems, and other distributed systems While it offers a robust simulation framework, it falls short in effectively simulating wireless communication networks To address this limitation, the scientific community recommends integrating the MIXIM framework into OMNET++.
MIXIM est un Framework créé pour la modélisation des réseaux filaires et mobiles (ré- seaux de capteurs sans fil, les rộseaux vộhiculaires) dans OMNET++ [Kửpke et al., 2008].
The article provides a detailed overview of radio wave propagation models, interference estimation, energy consumption, and wireless MAC protocols MIXIM, integrated with OMNET++, enables support for the IEEE 802.11p and 1609.4 MAC layers.
OMNET++ facilitates the easy simulation of wireless vehicular network protocols It also supports various probabilistic propagation models, including Log-Normal Shadowing, Nakagami, and Rayleigh, among others.
The OMNET++ simulator consists of modules that can be either simple or composed Each simple module is associated with a cc file and a h file, while a composed module contains interconnected simple modules or other composed modules Parameters, submodules, and ports for each module are defined in a ned file Communication between different modules occurs through message exchanges, with messages sent and received via ports that serve as the input and output interfaces for each module.
The design of a network is created in a ned file, with various parameters for each module specified in a configuration file (.ini) At the conclusion of each simulation, OMNET++ generates two new files: a vector file (.vec) and a scalar file (.sca), which are essential for conducting statistical analysis.
3.1.1.4 Comparaison entre les simulateurs NS2(NS3) et OMNET++
Following the description of the two network simulators, NS2 and OMNET++, the table below illustrates a comparison between them The findings of this study are presented in Table 3.3.
Tableau 3.3 Comparaison entre les simulateurs OMNET++ et NS2(NS3)
Les protocoles utilisés dans la simulation
3.1.2.1 Wireless Access in Vehicular Environment( WAVE)
The IEEE has developed a system architecture known as WAVE, designed to provide wireless access in vehicular environments This Wireless Access in Vehicular Environments (WAVE) system is a radio communication framework aimed at delivering seamless and interoperable transportation services These services align with the American Intelligent Transportation Systems (ITS) architecture and encompass various applications envisioned by automotive and transportation infrastructure industries globally, including Vehicle-to-Infrastructure (V2I) and Vehicle-to-Vehicle (V2V) communication.
The IEEE WAVE standards encompass several key specifications, including IEEE Std 1609.0, 1609.2, 1609.3, 1609.4, 1609.11, and 1609.12, along with IEEE Std 802.11, which facilitates communication among stations outside the context of a basic service set.
WAVE systems utilize a highly efficient messaging protocol known as the WAVE Short Message Protocol (WSMP) The messages transmitted via WSMP are referred to as WAVE Short Messages (WSM) The WAVE communication protocol (Wireless Access in Vehicular Environment)/IEEE 802.11p, along with the recently established WSMP, has been proposed as the standard protocol for developing applications for Vehicular Ad Hoc Networks (VANET).
La figure 3.5 ci-dessous illustre un aperỗu du protocole WAVE.
Figure 3.5 – Aperỗu du protocole WAVE[Win et al., 2020]
The IEEE WAVE standards encompass several key specifications, including IEEE Std 1609.0, 1609.2, 1609.3, 1609.4, 1609.11, and 1609.12, along with IEEE Std 802.11, which facilitates communication among stations outside a basic service set context.
Chaque standardisation a un rôle spécifique, décrit ci-après :
IEEE 1609.2 establishes standard security services for management applications and messages within vehicular environments This standard outlines the formats and processing of secure messages for wireless access devices, detailing methods to secure both WAVE management messages and application messages Additionally, it describes the administrative functions required to support essential security operations.
IEEE Std 1609.3 outlines networking services and details the Wave Short Messages Protocol (WSMP), which is designed to enhance communication between vehicles and reduce latency in road safety services.
— IEEE Std 1609.11 : Protocole d’échange de données de paiement électronique en direct pour les systèmes de transport intelligents (ITS) ;
— IEEE Std 802.11 : Fonctionnement en dehors du contexte d’un ensemble de services de base.
— IEEE 1609.4-2010 :La norme IEEE pour l’accès sans fil dans les environnements véhiculaires (WAVE) - Fonctionnement multicanal[IEEE2021].
The protocol for connecting road traffic (SUMO) with network simulators (OMNET++) manages vehicle behavior during simulations Both the network traffic simulator and the road traffic simulator utilize TraCI to gather a set of commands and responses The network flow is regulated based on the number of nodes, with Veins instantiating a network node for each vehicle in motion within SUMO This process is overseen by the TraCIScenario Manager Launchd module, which connects to a TraCI server (either SUMO or sumolaunchd) and subscribes to events such as vehicle creation and movement.
In SUMO, each created vehicle generates an OMNet++ module that includes a TraCIMobility sub-module for mobility simulation This module regularly updates the node's mobility information, such as position, speed, and direction, based on the vehicle's behavior For quick testing, TraCIMobility also features options to stop a vehicle at a predefined time, configured through its Accident Start and Accident Duration parameters The integration of the two simulators (network and road) via the TraCi protocol is illustrated in the accompanying figure.
Figure 3.6 – Simulateur de réseau OMNet ++ couplé au simulateur de trafic SUMO utilisant TraCI[Segata et al., 2014]
Techniques d’apprentissage automatique
Généralités sur les techniques de machine learning
Machine learning (ML) is a branch of artificial intelligence (AI) that enables systems to acquire knowledge without explicit programming The primary goal of ML is to allow computers to learn independently, mimicking human intelligence by adapting to their environment As a pivotal technology in the era of big data, machine learning techniques have been successfully applied across various fields, including pattern recognition, computer vision, aerospace engineering, finance, entertainment, computational biology, and biomedical applications.
Le processus d’apprentissage
The learning process begins with data collection from various sources, followed by data preparation, which involves preprocessing to address data issues and reduce dimensionality by removing irrelevant information Given the vast amount of data used for learning, decision-making becomes challenging; thus, algorithms are designed using logic, probabilities, statistics, and control theory to analyze the data and extract knowledge from past experiences The subsequent step involves testing the model to evaluate its accuracy and performance, leading to system optimization by refining the model with new rules or datasets Machine learning techniques are employed for classification, prediction, and pattern recognition, applicable in diverse fields such as search engines, web page ranking, email filtering, facial recognition, targeted advertising, character recognition, gaming, robotics, disease prediction, and traffic management.
Figure 3.7 – Processus Machine learning[Chiesa et al., 2020]
Types d’algorithmes de machine learning
Le machine learning est principalement divisé en trois catégories, à savoir les approches d’apprentissage supervisé, non supervisé et semi-supervisé.
Supervised algorithms require human input for both the necessary inputs and outputs, as well as feedback on the accuracy of predictions during the training process These algorithms are classified into two main categories: classification methods and regression methods.
— Algorithmes non supervisés : les approches d’apprentissage non supervisé contrastent avec les approches d’apprentissage supervisé ó elles ne nécessitent aucun processus de formation.
Semi-supervised learning is a class of machine learning techniques that leverages both labeled and unlabeled datasets Positioned between supervised learning, which relies solely on labeled data, and unsupervised learning, which utilizes only unlabeled data, semi-supervised learning has been shown to significantly enhance learning quality by combining the strengths of both data types.
Chaque catégorie de machine learning implémente ces différents algorithmes selon les approches.
Apprentissage profond
Deep learning, also known as deep learning or hierarchical learning, encompasses a range of machine learning methods that aim to model data at a high level of abstraction through complex architectures involving various nonlinear transformations This approach can learn without human supervision, utilizing both unstructured and unlabeled data.
Deep learning algorithms process data through multiple layers of neural network algorithms, with each layer transmitting a simplified representation of the data to the next At the core of deep learning is the neural network, which consists of interconnected neurons linked by weighted connections.
Several deep learning algorithms have been developed, including Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory networks (LSTM), Stacked Autoencoders, Deep Boltzmann Machines (DBM), and Deep Belief Networks (DBN).
In the following sections, we will provide a detailed overview of the artificial neural networks (ANN) algorithm, as it is the primary focus of our work Other algorithms will be discussed in the appendix.
— Réseaux de neurones artificiels(ANN)
Artificial neural networks (ANNs) mimic the functionality of the human brain through a system of interconnected nodes known as artificial neurons Each neuron can transmit information and is represented by a binary state (0 or 1), with assigned weights that determine its significance within the network The structure of an ANN is organized into layers, starting with the input layer, followed by hidden layers, and culminating in the output layer As data progresses through these layers, it is transformed into meaningful information, ultimately producing the desired output.
Figure 3.8 – Architecture ANN[Jain et al., 2019]
Les fonctions de transfert et d’activation jouent un rôle important dans le fonction- nement des neurones La fonction de transfert résume toutes les entrées pondérées comme suit :
Formule 3.1 : Fonction de transfert[Kaur and Kumari, 2020] ó b est la valeur de biais, qui est généralement de 1.
La fonction d’activation aplatit essentiellement la sortie de la fonction de transfert à une plage spécifique Il peut être linéaire ou non linéaire La fonction d’activation simple est :
Formule 3.2 : Fonction d’activation[Kaur and Kumari, 2020] Étant donné que cette fonction ne fournit aucune limite aux données, la fonction sigmọde est utilisée et peut être exprimée comme suit :
Formule 3.3 : Fonction sigmọde[Kaur and Kumari, 2020]
Conclusion
In this chapter, we outlined our approach and the necessary tools for its implementation, detailing the vehicular protocols used, specifically IEEE 802.11p (WAVE and WSMP), along with the protocol facilitating the interaction between road traffic simulation and network traffic, namely the Traffic Control Interface (TraCI) Road traffic scenarios will be utilized through the SUMO tool, while network traffic will be managed by OMNET++ in conjunction with the VEINS framework We also explored machine learning techniques to develop our predictive model In the next chapter, we will implement our proposed approach to predict demand between vehicle-to-infrastructure communications.
Généralités
The purpose of the experiment is to demonstrate the impact of road traffic on network traffic within a vehicular network using vehicle-to-infrastructure (V2I) communication This approach allows for measuring V2I demand in a defined scenario Additionally, the experiment provides a comprehensive view of the network established at the SDN controller level, enabling smarter and more efficient network behavior management, optimal resource allocation, and demand estimation.
Simulators are essential for analyzing the behavior of entities in wireless vehicular networks and assessing the performance of protocols designed for these networks This article addresses a problem with multiple scenarios, which are detailed in the following section, and highlights performance indicators through various metrics of probabilistic connectivity.
Nous allons décrire dans cette partie, l’environnement de simulation de notre modèle. Ensuite, nous allons présenter les différents scénarios de chaque approche et enfin analyser les résultats des simulations.
This chapter will be structured to first explore metrics and other factors influencing vehicle-to-infrastructure communication as a whole Following this, we will focus on demand prediction by analyzing real-world datasets related to vehicle-to-infrastructure communication, employing machine learning and deep learning techniques.
Étude de performance des métriques de connectivité probabiliste dans un réseau véhiculaire
Étude de cas
In a road network where vehicles enter a specific area defined by the simulation environment, the number of vehicles arriving remains unknown This uncertainty arises from the communication between the vehicles themselves and with the infrastructure, such as Roadside Units (RSUs).
The communication between the SUMO simulator and the network simulator is unidirectional, where SUMO sends commands and requests to OMNET++ to influence network behavior Given the unknown number of vehicles, there is a potential for network saturation within the defined simulation area Assuming a known number of vehicles, our focus is to predict and analyze the saturation rate trend, enabling proactive measures to optimize network resource allocation effectively and robustly based on demand.
Environnement de simulation
To evaluate the performance of our approach, we utilized the SUMO-1.8.0 traffic simulation tool and the OMNET++ 5.6.2 network simulator We chose OMNET++ due to its multiple advantages over the NS2 (NS3) simulator in relation to our study area Assessing wireless vehicular network protocols with OMNET++ requires a framework that enables its operation in parallel with the traffic simulation tool.
The Veins framework, also known as Vehicles in Network Simulation, aims to enable bidirectional coupling between the OMNeT++ and SUMO simulators to achieve simulation results that closely resemble real-world environments For our approach, we utilized Veins version 5.1.
Tableau 4.1 : Configuration de l’environnement matériel
Disque dur 500 GO Processeur i7-4600U CPU 2.10GHz
Tableau 4.2 : Configuration de l’environnement logiciel
Simulateur de trafic routier SUMO-1.8.0
Plateforme de couplage entre OMNET++ et SUMO Veins-5.1 sim-time-limit 500s
Portée du signal des OBUs 150m
Tableau 4.3 : Configuration de la micro-simulation du trafic routier(SUMO)
Paramètres Valeurs maxSpeed 25 m/s(90 km/h) accélération 2.6 m/s2 décélération décélération 4.5 m/s2 minGap(Distance entre véhicule) 2.5m sigma(Driver imperfection ) 0.5
Configuration du simulateur réseau(OMNET++)
Les principaux fichiers d’OMNET++ [Lửbbers and Willkomm, 2007] :
.NED files utilize the NED network description language and can be operated in two modes: graphical and text These modes facilitate the description of module parameters and ports Real-time error notifications are provided through a red dot positioned to the left of the code.
Figure 4.1 – Extrait fichier NED mode graphique
Figure 4.2 – Extrait fichier NED mode code source
— Fichiers(.ini) : Ce fichier est lié directement avec le fichier NED Il permet à l’utilisateur d’initialiser les paramètres des différents modules ainsi que la topologie du réseau.
Figure 4.3 – Extrait du fichier ini
Files with a (.msg) extension are used for communication between modules by exchanging messages that can include additional data fields OMNET++ translates these message definitions into C++ classes The following diagram illustrates the detailed execution process for running a simulation in OMNET++.
Figure 4.4 – Exécution d’une simulation sous OMNET++
Simulation de notre approche
In this section, we outline the step-by-step methodology for our simulation To simulate our approach in an urban environment, we imported a map of Manhattan measuring 1200 m x 1200 m For the highway environment, we utilized a 5 km highway map.
— Lancement du simulateur OMNET++ :mingwenv.cmd est le fichier qui lance le console de OMNET++ pour exécuter les scripts.
Figure 4.5 – Lancement de OMNET++ sur mingwenv.cmd
— Lancement du simulateur SUMO et écoute sur le port 9999 pour la connexion entre le serveur :
Figure 4.7 – Lancement de SUMO sur mingwenv.cm
4.2.4.2 Vu de l’ensemble du réseau coté OMNET++(Qtenv) et SUMO
Figure 4.9 – Présentation de l’environnement Qtenv du réseau
Figure 4.10 – Présentation de l’environnement SUMO du trafic routier
Les métriques de connectivité probabiliste
The performance evaluation of our approach was conducted using the Packet Delivery Ratio (PDR) metric, which measures the ratio of packets received by the destination to the number of packets sent, along with other key indicators.
Nous montrons dans ce qui suit, les résultats de simulations de notre étude de cas en tenant compte de la métrique mentionnée ci-précédente.
Présentation des Scénarios
Dans cette section, nous allons présenter nos différents scénarios effectués lors de la simulation.
4.2.6.1 Aperỗu des Traces de la simulation Étant donné que les véhicules représentent des nœuds, les traces générées par la si- mulation regroupent les différents modes de communications, à savoir la communication véhicule à véhicule et la communication véhicule à infrastructure (RSU) Ainsi, le simu- lateur générant les deux modes V2V et V2I, nous avons filtré les différentes traces pour ne considérer que la communication entre véhicule (nœud) et infrastructure (RSU), telle est notre approche Et c’est ce filtrage qui nous sera servi pour le reste de l’approche.
Figure 4.11 – Extrait de traces de données de la simulation
Considérons 20 nœuds dans la simulation, ces nœuds parcourent une trajectoire de 2.6km avec des vitesses différentes Nous avons évalué quelques nœuds suivant qu’ils ont le même comportement de vitesse Les résultats de la simulation sont représentés sur la figure 4.12 Ainsi, la vitesse varie entre 0 et 90 km/h(25 m/s), chaque véhicule arrive dans simulation à l’instant t i , ce qui veut dire que les véhicules arrivent à différents intervalles de temps, ce qui explique le graphique 4.12.
4.2.6.3 Graphique Nombre de paquets envoyés – Temps
As the number of nodes in a vehicle-infrastructure network increases, the volume of packets transmitted also rises For instance, with 10 nodes, approximately 11 packets are sent, and with 20 nodes, this number increases to around 21 packets Notably, when simulating 10 nodes, the simulation does not reach the expected total time, indicating that vehicles can traverse a defined area without congestion This reduction in network resource demand leads to optimal resource utilization The simulation results are illustrated in Figure 4.13.
Figure 4.13 – Graphique Nombre de paquets envoyés de 10 et 20 noeuds en fonction de Temps
In this scenario, vehicle speeds range from 0 to 90 km/h, with an average of 10 knots Figure 4.14 illustrates the speed trends during the simulation, providing insights into how speed affects the number of packets transmitted Each point on the graph represents the average speed of moving vehicles at that moment The average speed peaks at 75 km/h at the 112th second, while the maximum speed is reached, and from the 277th to the 289th second, the speed drops to zero This variation prompts an analysis of the number of packets sent during these critical moments.
Figure 4.14 – Graphique vitesse moyenne – Temps
4.2.6.5 Graphique Vitesse, Paquets envoyés – Temps
Based on the same assumptions as in Figure 4.14, it is observed that all vehicles begin transmitting at the 28th second, reaching maximum average speed by the 112th second of the simulation, with the packet transmission consistently at its peak However, during the interval from the 277th to the 289th second, speed drops to zero, leading to a significant decrease in packet transmission, indicating that speed significantly influences packet broadcasting behavior The simulation results are illustrated in Figure 4.15.
Figure 4.15 – Graphique vitesse moyenne, paquets – Temps
With a total of 20 nodes, the dissemination of all nodes will occur at the 58th second of the simulation By considering a simulation time t, we can recursively define the moment when all nodes will be disseminated.
L’instant par récurrence ó tous les noms diffuseront est défini de la manière :
Formule 4.1 : Temps de diffusion de tous les véhicules
4.2.6.6 Graphique PDR – Temps pour 10 noeuds
The packet delivery ratio measures the relationship between the number of packets received and the number sent Figure 4.16 demonstrates the packet delivery rate among 10 nodes and RSUs, showing stable performance until the 28th second, when packet loss occurs This loss coincides with the moment when all nodes broadcast simultaneously, indicating that simultaneous transmissions by vehicles can lead to network congestion and packet loss By the 201st second, packet losses decrease, attributed to reduced traffic and fewer vehicles transmitting data.
4.2.6.7 Graphique PDR – Temps pour 20 nœuds
Under the same assumptions as the previous figure, it is observed that with 20 nodes, packet losses begin at the 58th second and cease at the 201st second During this period, all nodes are broadcasting, which can lead to network losses The simulation results are illustrated in Figure 4.17.