Research On Mobility Management Of Vehicle Network With Machine Learning

With the increasing number of vehicles,the demand for mobile management of vehicle networking is becoming more and more urgent.The rapid development of mobile communication makes the vehicle networking system get better network and technical support guarantee,and the realization of intelligent transportation system goes further.The research on mobile management of vehicle network can optimize the construction and deployment of basic network equipment in urban vehicle network,reduce traffic congestion,improve the user’s communication service experience,provide technical support for unmanned platform,and bring more help to the realization of intelligent transportation system.Traditional vehicle network mobility management is based on mobile IP(the Mobility-enable protocol for the Internet)management,the advent of the Big Data era has brought new ideas.Based on the actual vehicle data,this thesis studies the mobility management of vehicle network,aiming to realize the more efficient management scheme of vehicle network mobility through the characteristic analysis of vehicle data and the modeling method of prediction,so as to meet the needs of networked design in intelligent transportation.The main works of this thesis are introduced as follows:(1)Optimal deployment of roadside units based on vehicle data analysis:The network system communication capability in the V2I(Vehicle-to-Infrastructure)in the vehicle networking is closely related to the deployment of infrastructure such as roadside units,and its deployment affects network reliability and information exchange.Based on the actual trajectory data of taxi in Beijing,this research scheme first gets the candidate location point of the roadside unit according to the road topology and vehicle density data,and then uses the branching bounding algorithm to effectively solve the multi-objective optimization problem of delay and deployment cost demand.Simulation results show that this scheme saves about 50%of the deployment cost when the roadside unit radius is 100 meters.(2)Resear-ch on vehicle individual mobility prediction based on vehicle trajectory data:Aiming at the multi-classification problem of vehicle mobility prediction,the classical machine learning algorithms(random forest,Adaboost,GBDT and SVM)are used to solve the accuracy of vehicle mobility prediction,and compared with the traditional Markov process modeling method.Simulation results show that the effect of random forest is the best,and the prediction accuracy of test set can reach 83%.Based on the stochastic selection characteristics of random forests,the thesis designs the method of selecting features and sample sets in random forest algorithm by introducing the importance weighting of data feature knowledge,and the accuracy rate can be improved stably by more than 1.2%.(3)Research on mobility management of vehicle network based on data analysis:Aiming at the problem of high interrupt rate of communication connection caused by congestion during peak road period,vehicle access mode management scheme based on greedy algorithm is proposed.According to the interrupt probability threshold,some vehicle nodes are set up as mobile relay access mode by greedy algorithm to send data,the result is a reduction of about 25%of the interrupt probability.Then,according to the problem of vehicle power supply in the application scene of autonomous vehicle,based on the historical trajectory of the vehicle combined with the time and space characteristics of the road and data,the location of the reasonable power supply point is obtained by using the DBS CAN clustering algorithm,and the experimental results show that 0.54 coverage can be achieved by setting only 65 places.