基于双Q学习算法的安全容量的优化  

Secrecy capacity optimization algorithm based on dueling double deep Q-Network

作　　者：杜明[1] 任建国 张清杨 DU Ming;REN Jianguo;ZHANG Qingyang(School of Computer Science and Technology,Jiangsu Normal University,Xuzhou 221116,China)

机构地区：[1]江苏师范大学计算机科学与技术学院,江苏徐州221116

出　　处：《现代电子技术》2024年第11期181-186,共6页Modern Electronics Technique

基　　金：国家自然科学基金项目(62006103);江苏省教育科学规划课题(C/2023/01/24);江苏师范大学实验室研究课题(L2023YB05)。

摘　　要：由于无线传播的广播和开放性,车联网通信系统易遭受到窃听者的窃听,这降低了通信系统的安全容量。为此,文中提出基于深度强化学习的安全容量的优化算法(D3QN-RS)。利用随机过程理论将最优转发节点的选举问题转化为马尔可夫决策过程(MDP),并将截获概率融入奖励函数。利用奖励值抑制截获概率,并将选择转发节点的问题转入强化学习框架。最后,利用双Q学习算法(D3QN)求解,产生最优的转发节点,进而提升系统的安全容量。仿真结果表明,与随机选择转发节点的算法相比,D3QN-RS算法的截获概率下降了约15%,系统的安全容量提升了约12%。Due to the broadcast and openness of wireless communication,the Internet of Vehicle(IoV)communication system is vulnerable to eavesdropping,which reduces the secrecy capacity(SC)of the communication system.Therefore,SC optimization algorithm named D3QN-RS based on deep reinforcement learning(DRL)is proposed.On the basis of the random process theory,the problem of optimal relay selection(RS)is transformed into the Markov decision process(MDP),and the intercept probability is integrated into the reward function.The reward value is used to suppress interception probability,and the problem of RS is transferred into the reinforcement learning framework.Finally,the dueling double deep Q-network(D3QN)algorithm is used to solve the reinforcement learning problem and generate the optimal RS,so as to improve the SC of the system.Simulation results show that the D3QN-RS algorithm performs better than random RS,because it reduces intercept probability by approximately 15%and improves SC by approximately 12%.

关 键 词：车联网 安全容量 截获概率 信道状态信息 转发节点的选择 D3QN 

分 类 号：TN929.5-34[电子电信—通信与信息系统]