【摘要】 针对正交频分复用的多蜂窝网络系统，提出了一种基于深度强化学习的通信资源分配算法，该算法在满足资源分配高速率、低延时要求的前提下，同时产生信道分配方案和功率控制方案，从而最大化系统的能量效率。首先，在确定好基于正交频分复用的多蜂窝网络系统模型的基础上，将最大化能量效率的约束优化问题同深度Q强化学习算法进行问题映射。其次，将构建的深度Q神经网络（DQN）的多个隐藏层作为状态值函数，用以输出信道分配方案和功率控制方案，并实时与外界环境保持交互，不断迭代更新网络参数用以最大化系统能量效率。通过仿真对比实验可得，所提出的深度强化学习算法在保证低计算时延的同时，可获得接近于或高于其他算法的系统能量效率，且蜂窝网络规模越大，该算法优势越突出。更多还原

【Abstract】 This paper proposes a resource allocation method for orthogonal frequency division multiplexing(OFDM)multi-cell cellular networks based on deep reinforcement, which generates both channel allocation schemes and power control schemes to maximize the energy efficiency of the system while satisfying the high speed and low latency requirements for resource allocation. First, the constrained optimization problem of maximizing the energy efficiency of the system is mapped to a deep Q reinforcement learning algorithm based on a model of an orthogonal frequency division multiplexing multicellular network system. Second, multiple hidden layers of the constructed deep Q neural network(DQN) are used as state value functions to output channel allocation schemes and power control schemes, and interact with the external environment in real time to continuously iteratively update the network parameters to maximize the system energy efficiency. Finally, by comparing the simulation experiments, the proposed deep reinforcement learning algorithm can obtain a system energy efficiency close to or higher than the other algorithms while ensuring the low computing delay, and the larger the cellular network scale, the more prominent the advantages of the algorithm are.更多还原