Project topics and materials
Get Complete Project Material File(s) Now! »
Mobility Models in Vehicular Ad hoc Network
In an ITS, attention is paid to both individual intelligence and systematical intelligence because of the different research focuses. Individual intelligence refers to the intelligent vehicles, the self-driving car and cruise aid system, while the systematic intelligence means the implementation of vehicular net-works.
Brief introduction to VANET
Although we are so eager to achieve system-level intelligent transportation, it cannot be accomplished by simply replacing the vehicles on road with intelli-gent vehicles. One of the key features in ITS is the communication ability of all the participants which enables information exchange among them. A Vehicu-lar Ad hoc Network (VANET) is a part of the whole ITS framework. It is an ideal network model to represent the communication among vehicles, while vehicles are regarded as the mobile nodes within the network.
As for a real-world vehicle, driver makes his decision according to his judg-ment, which is based on his acknowledgment of his surrounding. There are methods that dedicate to path planing meanwhile concern the modeling of sight range (Geraerts & Schager (2010)). This kind of vehicle can be repre-sented by an intelligent vehicle, and the sight range depends on the perfor-mance of its sensors. However, suppose the driver is aware of the whole sit-uation, it is possible for him to make the global optimal decision, which will greatly increase the efficiency of the whole traffic system. This total awareness is possible for an intelligent vehicle if its sensors accuracy and range perfor-mance can be ensured fully functional. Assuming that, the drivers can share their sight and acknowledgment of the environment, it could help them to predict the future road condition and make even better decisions. This infor-mation share is normally represented by the communication among intelligent vehicles. Thus the individual intelligence can be turned into group intelligence and systematic intelligence.
To achieve these features, there are several kinds of such communication in VANET. The communication between a vehicle and another vehicle or non-vehicle agent is usually noted as V2X, including V2V (Vehicle-to-Vehicle), V2I (Vehicle-to-Infrastructure), V2P (Vehicle-to-Pedestrian) and V2C (Vehicle-to-Cloud) (Park & Min (2015)). With VANET, we can upgrade autonomous sys-tems to cooperative systems.
Major concerns of VANET
While being promising in saving time and saving lives and being conceptually straightforward, design and deployment of VANET is technically and econom-ically challenging (Hartenstein & Laberteaux (2010)). In technical terms, its concerns include the following issues:
• High mobility of the nodes. Due to high relative speed between cars, network’s topology changes very fast, and many paths disconnect before they can be used (Yousefi et al. (2006)). Wang (2004a,b) tried to find the approximation of link’s lifetime. In (Artimy et al. (2004, 2005)), the au-thors attempted to capture some relationship between the model of ve-hicular mobility and connectivity of the networks. In (Blum et al. (2004)), the effective network diameter in a typical VANET is studied.
• Self-organized structure. Lacking of centralized management and coor-dination entity. It is difficult to synchronize and manage the transmission events of different nodes.
• The signal’s physical transmission. The transmission could be blocked by obstacles, e.g. buildings in cities, the propagation models of the signal and the influence on the performance of VANET need to be studied.
• Standardization and flexibility. Standardization is necessary to make the equipment from different manufactures compatible. while the manufac-tures will want to create some product differentiation.
Traffic prediction with Deep Learning approaches
Another important application of the realistic model is in traffic prediction. A good traffic prediction has the potential to improve traffic conditions and re-duce travel delays by facilitating better utilization of available capacity. Since deep learning method with neural network model will be used in the following chapters to produce our realistic convincing mobility models, some represen-tative uses of neural network in traffic prediction methods will be examined in this section.
Song et al. (2016) constructed a deep recurrent neural network to predict the human mobility. Satellite GPS records are collected and used, which means the human mobility in this model may contain mobility with various kinds of transportation, such as by foot, by car or by train. Lv et al. (2015) however, makes the prediction for traffic flows. Stacked autoencoders structure is used to learn the patterns in the traffic data with training performed by a greedy layer-wise fashion. Likewise, Ma et al. (2015) uses a long short-term memory neural network for traffic speed prediction. Although these methods and our method in this thesis show some degrees of similarity, the traffic predictions normally come with traffic flow, as a conse-quence, it will be always acted on macroscopic level, such as in (Lv et al. (2015)). There is no individual trajectories there. In other representative examples, such as (Song et al. (2016)), GPS records from satellite can only produce rough tra-jectories data. As for the result, it is more likely to the Original-Destination structure than a detailed trajectory.
Table of contents :
Acknowledgements
Table of Contents
List of Figures
List of Tables
List of Algorithms
Abbreviations
General Introduction
1 Introduction
1.1 Background
1.1.1 Motivation
1.1.2 Intelligent Transportation Systems
1.2 Mobility Models in Vehicular Ad hoc Network
1.2.1 Brief introduction to VANET
1.2.2 Major concerns of VANET
1.2.3 Self-Driving Cars and ACC systems
1.3 Challenges, Opportunities and Connections
1.3.1 Realistic level: another starting point
1.3.2 Traffic prediction with Deep Learning approaches
1.3.3 Relationship with Self Driving
1.4 Conclusion
2 Neural network based data-driven mobility model
2.1 Introduction
2.1.1 Neural Network in intelligent traffic
2.2 Problem descriptions – Why using NN
2.3 Neural network structure and backpropagation algorithm
2.4 NN Application in traffic simulation problem
2.4.1 Pre-processing the obtained data
2.4.2 Deep Learning Method
2.4.3 Learning Process
2.4.4 Implement simulator for virtual traffic flow
2.4.5 Performances
2.5 Discussion and Conclusion
2.5.1 Discussion
2.5.2 Conclusion
3 Enhanced Mobility Model with HMM
3.1 Introduction
3.2 Traditional Mobility Models and their adaptation
3.2.1 Mobility Models
3.2.1.1 Basic categories on Mobility Models
3.2.1.2 Movement restrictions in Mobility Models
3.2.1.3 Adaptation to Probabilistic Models
3.2.2 Basic adaptation in 1-dimension
3.2.3 Adaptation of Car-following model in 2-dimension
3.2.4 Summary on Adaptation method of Mobility Models
3.3 HMM for Mobility Model improvements
3.3.1 HMM in intelligent traffic
3.3.2 HMM: Basic knowledge and its denotation
3.3.2.1 Definitions and denotations
3.3.2.2 The three problems
3.3.3 HMM combined Mobility Model
3.3.3.1 Scaling Problem
3.3.3.2 Multi-observation sequences
3.3.3.3 Partial knowledge estimation
3.3.4 Performance and result
3.4 Discussion and Conclusion
3.4.1 Discussion
3.4.2 Conclusion
4 Experiment platform and scenario simulation
4.1 Introduction
4.2 UML design of NN-VMM
4.2.1 Commercial usage
4.2.2 Open-source platform
4.3 Scenario simulation
4.3.1 Highway scenario reproduction
4.3.2 Scenario extension: MIXED NN-VMM/TP-AIM
4.3.3 Applications in Transportation technology
4.4 Discussion and Conclusion
Conclusions and Perspectives
Résumé Étendu en Français
References
