异构网络中基于 MADDPG的协作边缘缓存策略研究

由于大量用户和设备共存，移动网络经历了数据量和用户密度的巨大增长。在宏基站(macro base station, MBS)覆盖区域内部署小蜂窝基站(small basic station, SBS),并提前在SBS缓存热门内容，是下一代移动通信网络提供高速、低时延服务的有效手段。针对异构网络环境不稳定以及难以找到精确的数学模型进行优化的问题，提出一种基于传输时延最小的异构网络协作边缘缓存算法。首先以Markov移动预测模型为基础，考虑用户社交关系对于用户移动性的影响，给出了新的用户移动位置预测方法；其次，采用多智能体深度确定性策略梯度(multi-agent deep deterministic policy gradient, MADDPG)算法，通过用户关联、延迟控制和缓存设计来减少内容传输时延并提高缓存命中率。仿真结果表明，同传统DDPG和Greedy算法相比，MADDPG算法缓存命中率分别提高17.89%和42.71%,内容传输时延分别降低9.07%和12.86%,能够有效地解决异构网络中的资源分配和缓存设计问题。

Strategy of MADDPG-based collaborative edge caching in heterogeneous networks

Mobile networks have experienced a tremendous growth in data volume and user density due to the coexistence of a large number of users and devices.Deploying small basic station(SBS)within the coverage area of macro base station(MBS)and caching popular content in SBS in advance is an effective means to provide high-speed and low-latency services in next-generation mobile communication networks.Aiming at the unstable heterogeneous net work environment and the difficulty of finding an accurate mathematical model for optimisation,a collaborative edge caching algorithm for heterogeneous networks based on transmission delay minimization is proposed.Firstly,a new user mobile location prediction method is given based on Markovian mobile prediction model,considering the influence of user social relations on user mobility.Secondly,multi-agent deep deterministic policy gradient(MADDPG)algorithm is used to reduce the content delivery delay and improve the cache hit rate by user association,delay control and cache design.Simulation results show that compared with traditional DDPG and greedy algorithms,the cache hit rate is improved by 17.89%and 42.71%,and the content delivery delay is reduced by 9.07%and 12.86%,respectively,which can effectively solve the resource allocation and cache design problems in heterogeneous networks.