Study On Machine Learning Based DTN Routing Protocols For Vchicular Ad Hoc Networks

Vehicular Ad hoc Networks(VANETs) are an important component of intelligent transportation, and its research & development has got lots of attentions from Government-enterprises and research institutions. The message exchange and resource sharing of VANETs is achieved by the opportunistic encounter among vehicles, which leads to some features, e.g., the wireless link is easy to be broken, network size is lager, and opportunistic delivery messages and topology structure changes frequently. The transmission stability and extensibility of VANETs can be efficiently improved by building Vehicular Delay Tolerant Network(VDTN). Therefore, VDTN is a main research emphasis in the VANETs.This thesis proposes a multiple copy routing VDTN protocol based on decision tree for vehicular social network, which can resolve the problem of discontinuity connection and short encounter time, and how to reasonably select corresponding social properties to transfer the data. The proposed algorithm improves the method of selecting the social properties, and firstly collects the historical data by combing with machine learning, and then establishes attribute rule based on the learning results of historical data. Furthermore, predicting the optimal next hop node by utilizing the established attribute rule in the spray and forward phase, which enables the message approach to the objective node fast and directly. The simulation scenarios is simulated by the ONE simulation software and real data, and verifies the performance of routing protocol by changing all aspects of target. Finally, the simulation results show that, the proposed routing protocol can lower routing overhead by 20% compared with the single attribute social routing protocol.Aiming at the problem that the vehicles' node density distribution is uneven in different urban areas, and the message transmission is easily affected by the network environment and traffic flow on the road. This thesis proposes an opportunistic routing scheme based on traffic information, which not only can guarantee the higher delivery rate, but also can lower transmission delay, and further verifies that the node degree has scaleless property in VANETs. The proposed scheme is composed of two aspects, and firstly selecting the data forward road among areas based on the traffic flow, which enables the message quickly arrive to non-self regions. Secondly, utilizing the CART learning algorithm to evaluate the attribute value of community in self region, and then establishing the IF-THEN rule to forward message. The ONE simulation results show that, the proposed algorithm not only can guarantee higher delivery rate, but also can lower transmission delay as compared to the existing algorithm.