Research On Routing Algorithms Of Vehicular Delay Tolerant Networks Based On Random Forest

Vehicle Delay Tolerant Network(VDTN)is a new type of delay-tolerant network structure,whose messages are stored,carried and forwarded by moving vehicle nodes as carriers.Because there is no definite end-to-end connection between the source vehicle node and the destination vehicle node of the message,the message forwarding mechanism is facing enormous challenges.At the same time,with the high speed of vehicles,the dynamic changes of network topology seriously affect the quality of communication.Therefore,how to design an effective high-performance routing algorithm is one of the focuses of VDTN research.In this thesis,the related theoretical knowledge of VDTN and the current research status of routing algorithms for delay tolerant networks have been analyzed.Then,Random Forest(RF)and Deep Forest(DF)are described in detail,which lays a foundation for the design and simulation of routing algorithms for VDTN.Aiming at the problem of the communication link instability caused by the high-speed movement of vehicle nodes in VDTN,a RF-based VDTN routing algorithm is proposed by utilizing the movement pattern of vehicle nodes.Firstly,the attributes associated with the movement of vehicle nodes are introduced,and the vehicle nodes are classified by a dynamic contact reward mechanism to construct the initial RF model.Then,the model is further optimized from two aspects of the classification performance and the diversity of decision trees,and the decision trees with good classification performance and high diversity are selected to construct an improved RF model,where the classification performance and the diversity of decision trees are measured according to the classification error rate of each tree with the corresponding penalty weights and the similarity between decision trees defined by the disagreement measure,respectively.Finally,a new VDTN routing algorithm is proposed based on the improved RF model.Simulation results show that the proposed routing algorithm can significantly improve the message delivery ratio and reduce the message delivery delay,and thus its effectiveness is verified.Moreover,to extend the above work,a DF-based VDTN routing algorithm is proposed,which improves the accuracy of delivery level prediction by constructing DF model.Firstly,the XGBoost and Adaboost model are introduced to change the basic classifier types of cascade forest,which increases the diversity of the model and improves the overall stability of DF.Secondly,two cascading sub–layers are added in each cascade structure,which use layer by layer processing to transform features within the model,and extract attributes associated with the movement pattern of the vehicle nodes.Finally,the Relief F algorithm is utilized to select features with high weights to compose a new DF input feature dataset,in order to further improve the predictive performance of the model.The simulation results show that the optimized DF model can more accurately predict the delivery level of the vehicle nodes,and the proposed DF-based VDTN routing algorithm improves the message delivery ratio and reduce the message delivery delay effectively.The results of this thesis can provide ideas for the research of VDTN routing algorithm,and have abroad application prospects in the practical vehicle networking environment.