基于深度强化学习的无人机可信地理位置路由协议

针对无人机(UAV)通信过程中存在的高移动性和节点异常问题，该文提出一种基于深度强化学习的无人机可信地理位置路由协议(DTGR)。引入可信第三方提供节点的信任度，使用理论与真实的时延偏差和丢包率作为信任度的评估因子，将路由选择建模为马尔可夫决策过程(MDP)，基于节点信任度、地理位置和邻居拓扑信息构建状态空间，然后通过深度Q网络(DQN)输出路由决策。在奖励函数中结合信任度调整动作的价值，引导节点选择最优下一跳。仿真结果表明，在包含异常节点的无人机自组网(UANET)中，DTGR与现有方案相比具有更低的平均端到端时延和更高的包递交率。当异常节点数量或者比例变化时，DTGR能感知环境并高效智能地完成路由决策，保障网络性能。

Trusted Geographic Routing Protocol Based on Deep Reinforcement Learning for Unmanned Aerial Vehicle Network

Considering the problems of high mobility and abnormal nodes in Unmanned Aerial Vehicle (UAV) communication, a Deep reinforcement learning based Trusted Geographic Routing protocol (DTGR) is proposed. A trusted third party is introduced to provide the trust of nodes. The difference between theoretical delay and real delay, and packet loss ratio are used as evaluation factors of trust degree. Routing selection is modeled as the Markov Decision Process (MDP). The state are constructed based on the neighbor nodes’ geographic location, the trust degree and the topology information. Then the routing decision can be output through the Deep Q Network(DQN). The action-value is adjusted by combining trust in reward function, to guide nodes to select the optimal next-hop. The simulation results show that DTGR has a lower average end-to-end delay and higher packet delivery ratio compared with existing schemes in UAV Ad hoc NETwork (UANET) with abnormal nodes. Besides, DTGR can effectively implement route selection and ensure network performance when the number or proportion of abnormal nodes changes.