Abstract: Targeting the high maintenance costs and poor dynamic adaptability of existing routing algorithms for the Internet of vehicles, this study proposes a multi-hop clustering composite routing algorithm based on improved Q-learning. In the cluster maintenance phase, the cluster head selects the node with good communication performance and relatively stable speed with the edge vehicles as the gateway node according to the utility function of the gateway.In the route establishment stage, by considering the link communication quality, packet transmission direction, and node mobility, a node performance evaluation function evaluates the comprehensive performance of the selected next-hop node to avoid the "blind path" problem.In the Q-learning stage, the link duration and distance of adjacent nodes are expressed by a quantitative method and used as the learning and discount rates to improve the learning efficiency of Q-learning. Simulations on mobile datasets on Cologne, Germany and a city in China show that, compared with the RSAR, GPSR, and TCRA routing algorithms, the routing lifetime and throughput of the proposed algorithm improved by 17.71% and 32.56% on average, respectively, and the communication delay and packet loss rate decreased by 14.3% and 66.32% on average, respectively.These results indicate that the proposed algorithm can adapt to complex and Vehicular Ad Hoc Networks(VANETs).

Key words: Vehicular Ad Hoc Networks(VANETs), clustering mechanism, gateway selection, composite routing, Q-learning

摘要： 针对现有车联网路由算法存在路由维护开销大、动态适应性差的问题，提出一种基于改进Q学习的多跳分簇复合路由算法。在簇维护阶段，簇头根据网关效用性函数选择边缘车辆中通信性能较优且速度相对稳定的节点作为网关节点。在路由建立阶段，通过考虑链路通信质量、数据包传输方向和节点移动性三个方面，设计节点性能评估函数，用于评估所选择下一跳节点的综合性能，以避免出现“盲路”问题，在Q学习阶段，通过定量化方法表示相邻节点的链路持续时间和距离，并将其作为学习率和折扣率，以提升Q学习的学习效率。在德国科隆和国内某市移动数据集上的实验仿真结果表明，相比RSAR、GPSR和TCRA路由算法，该算法的路由生存时间、吞吐量平均提高17.71%和32.56%，通信延迟和丢包率平均降低14.3%和66.32%，能适应复杂多变的车辆自组织网络。

关键词: 车辆自组织网络, 分簇机制, 网关选择, 复合路由, Q学习

CLC Number: 

• TP393