【摘要】 为解决因超可靠低时延通信(URLLC)与海量机器类通信(mMTC)等服务场景特性差异而导致的通信与计算资源动态调度问题,提出一种基于深度确定性策略梯度(DDPG)的资源分配(DDPG-RA)算法．首先构建以系统总时延与能耗最小化为目标的优化模型,再利用DDPG框架设计多维状态空间与连续动作决策机制,并分别为时延敏感的URLLC和能耗敏感的mMTC设计差异化奖励函数,以实现资源的动态优化配置．仿真结果表明:在典型城市场景下,与传统深度Q学习(DQN)算法和均匀分配(FAAS)算法相比,DDPG-RA算法使URLLC用户任务处理时延降低了4.3%;在用户设备增至14台的高负载场景时,DDPG-RA算法较FAAS算法的能耗下降了17.6%.更多还原

【Abstract】 In order to address the dynamic scheduling issue of communication and computing resources caused by the characteristic differences of service scenarios such as ultra-reliable low-latency communication(URLLC) and massive machine type communication(mMTC), a resource allocation algorithm based on deep deterministic policy gradient(DDPG-RA) is proposed. Firstly, an optimization model aimed at minimizing the total system delay and energy consumption is constructed. Then, the DDPG framework is used to design the multi-dimensional state space and continuous action decision-making mechanism. Finally, differentiated reward functions are designed respectively for the delay-sensitive URLLC and the energy-sensitive mMTC to achieve the dynamic optimization allocation of resources. The simulation results show that in typical urban scenarios, compared with the traditional deep Q network(DQN) algorithm and the Function as a Service(FAAS) algorithm, the DDPG-RA algorithm reduces the task processing delay of URLLC users by 4.3%, and that in the high-load scenario where the number of user devices are increased to 14, the energy consumption of the DDPG-RA algorithm is 17.6% lower than that of FAAS algorithm.更多还原