Routing Algorithms For Urban VANETs Based On Machine Learning And Fuzzy Logic

As a special Mobile Ad Hoc Network(MANET),the Vehicular Ad Hoc Network(VANET)is composed of vehicle nodes and roadside infrastructures,which combines vehicle technology,sensor technology and wireless communication technology.The VANET is a kind of multi-hop and centreless wireless network and has the characteristics of fast deployment,self-organization,strong scalability and robustness,which can effectively solve the defects of traditional vehicular communication technology and provide important technical support for Intelligent Traffic System(ITS)service.Urban scenario is one of the widely used VANET scenarios.How to design routing algorithm for urban VANET is very important for reliable communication between vehicles,which has become a research hotspot in recent years.Firstly,the network architecture,radio interface technology and routing algorithm of VANETs are introduced in this thesis.And then,some typical routing algorithms in MANETs and their improvements for VANETs are investigated.In addition,the main characteristics of urban VANETs are presented and the routing algorithms for urban VANETs are also reviewed,mainly focusing on some representative traffic aware routing algorithms.Secondly,a Fuzzy Q-learning based Traffic Aware Routing(FQTAR)algorithm is proposed for urban VANETs,that consists of intersections,roads and buildings.In the FQTAR,a real-time traffic aware mechanism is designed to reduce the overhead produced in traffic aware process.In addition,a routing strategy is proposed based on collected traffic results and fuzzy Q-learning algorithm,which consists of different strategies for forwarding data packets within roads and intersections,respectively.Moreover,fuzzy logic algorithm is employed to eliminate the problem of inaccurate link information.Furthermore,an adaptive Q-learning rate is designed to meet the convergence requirements in dynamic VANET environment.FQTAR can reduce the traffic aware overhead and improve the average packet delivery rate without sacrificing too much average packet delivery delay.Thirdly,aiming at the dynamic traffic in urban VANETs,a Scenario Aware Routing algorithm based on Back propagation neural network and Fuzzy logic(SAR-BF)is proposed.By training neural network to fit the vehicle eigenvalues which are nonlinear with the scenarios,the vehicle can accurately determine the state of scenario after the output value of BP neural network is processed by Exponentially Weighted Moving Average(EWMA)technology,and adaptively adjust the metrics(membership functions in fuzzy logic algorithm)of link evaluation.In addition,a dynamic multi-distribution mechanism is proposed,which can control the number of packet copies according to the scenario.SAR-BF can meet the routing requirements of different scenarios,and effectively improve the average packet delivery rate with only a little increase in average overhead and average packet delivery delay.Finally,OPNET and Vanet Mobi Sim are applied to develop the VANET simulation scenarios,and MATLAB is used to train the BP neural nework.Our proposals are evaluated by extensive simulations,and compared with several typical algorithms.The simulation results verify the effectiveness of our proposals.