移动车载边缘网络中基于递归深度强化学习的协作缓存接力算法  

Recursive deep reinforcement learning-based collaborative caching relay algorithm in mobile vehicular edge network

作　　者：吴红海 王白冰 马华红 邢玲 WU Honghai;WANG Baibing;MA Huahong;XING Ling(College of Information Engineering,Henan University of Science and Technology,Luoyang 471023,China)

机构地区：[1]河南科技大学信息工程学院,河南洛阳471023

出　　处：《通信学报》2024年第11期277-286,共10页Journal on Communications

基　　金：国家自然科学基金资助项目(No.62272146,No.62071170,No.62171180,No.62072158,No.U23A20272,No.U22A2069)。

摘　　要：考虑无路侧单元覆盖的场景,充分利用车辆之间的协作来构建缓存系统,提出一种基于递归深度强化学习的协作缓存接力算法。考虑缓存决策的动态特性,将问题建模为部分可观察的马尔可夫决策过程,利用图神经网络预测车辆轨迹,并通过计算车辆间的连接稳定性度量,选择可作为缓存节点的车辆。此外,将长短期记忆网络嵌入深度确定性策略梯度算法中,以实现最终的缓存决策。仿真结果表明,所提算法在缓存命中率和时延方面优于传统缓存算法。Considering scenarios without road side unit coverage,a recursive deep reinforcement learning-based collab‐orative caching relay algorithm was proposed to construct a caching system by leveraging the cooperation among ve‐hicles.Recognizing the dynamic nature of caching decisions,the problem was modeled as a partially observable Markov decision process.Vehicle trajectories were predicted using graph neural network,and the connectivity stability between vehicles was measured to select those that could serve as caching nodes.In addition,long short-term memory network was integrated into the deep deterministic policy gradient algorithm to achieve the final caching decision.Simulation re‐sults demonstrate that the proposed algorithm outperforms traditional caching algorithms in terms of cache hit ratio and latency.

关 键 词：车载边缘网络 协作缓存接力 递归深度强化学习 马尔可夫决策 

分 类 号：TN92[电子电信—通信与信息系统]