Research On Reinforcement Learning Based Reliable Routing Protocol In VANETs

As an important component of Intelligent Transport System(ITS),Vehicular ad hoc Networks(VANETs)has fascinated researchers,Government agencies and a wide range of companies since the concept of establishing communication network between vehicles has been put forward since the 1980s.This paper studies the reliable and effective routing mechanism in VAENT environment.It is found that the selection criteria of the existing routing protocols for the next-hop forwarding node are single-metric,the selection is not forward-looking and fails to consider the quality of the neighbor nodes of the next hop forwarding node.Based on the analysis,we put forward a position-based routing protocol PbQR(Position-based Q-leaning Routing),the PbQR algorithm uses the reinforcement-learning to evaluate the quality of the neighboring neighbors,and thus guides the selection of the forwarding nodes to ensure the stability and reliability of the routing links.And we simulates the performance of the protocol in the city scene with the network simulation platform NS2.The simulation results show that the proposed PbQR protocol is improved compared with the existing routing protocols in key performance such as packet delivery rate and average end-to-end delay and so on.On the other hand,Considering the VANET nodes in the scene with high moving speed,the network topology changes frequently,thus causing the probability of routing link breakage increases,and degrades the performance of the routing protocol.By using the clustering algorithm to improve the relative stability between nodes and using Dyna framework to provide the planning process to accelerate the learning speed of Q learning,a clustering routing algorithm based on Dyna-Q CbDR(Cluster-based Dyna-Q aided Routing)is proposed.The simulation results show that the proposed CbDR algorithm is scalable in the vehicle scene with different node density.The simulation results show that the proposed CbDR algorithm is scalable in different node density scenario.Moreover,the instability of the vehicle nodes is reduced by the clustering method,which ensures the performance of the routing algorithm in the vehicle scene with faster node movement.