The resource optimization of ultra-dense networks (UDNs) is critical to meet the huge demand of users for wireless data traffic. But the mainstream optimization algorithms have many problems, such as the poor optimization effect, and high computing load. This paper puts forward a wireless resource allocation algorithm based on deep reinforcement learning (DRL), which aims to maximize the total throughput of the entire network and transform the resource allocation problem into a deep Q-learning process. To effectively allocate resources in UDNs, the DRL algorithm was introduced to improve the allocation efficiency of wireless resources; the authors adopted the resource allocation strategy of the deep Q-network (DQN), and employed empirical repetition and target network to overcome the instability and divergence of the results caused by the previous network state, and to solve the overestimation of the Q value. Simulation results show that the proposed algorithm can maximize the total throughput of the network, while making the network more energy-efficient and stable. Thus, it is very meaningful to introduce the DRL to the research of UDN resource allocation.

中文翻译：

---

基于深度强化学习的超密集网络资源分配算法

超密集网络（UDN）的资源优化对于满足用户对无线数据流量的巨大需求至关重要。但是主流的优化算法存在很多问题，如优化效果差，计算量大等。提出了一种基于深度强化学习（DRL）的无线资源分配算法，旨在最大化整个网络的总吞吐​​量，并将资源分配问题转化为深度的Q学习过程。为了有效地在UDN中分配资源，引入了DRL算法以提高无线资源的分配效率。作者采用了深度Q网络（DQN）的资源分配策略，并采用经验重复和目标网络来克服由先前网络状态引起的结果的不稳定性和发散性，并解决对Q值的过高估计。仿真结果表明，该算法可以使网络的总吞吐​​量最大化，同时使网络更加节能，稳定。因此，将DRL引入UDN资源分配的研究具有十分重要的意义。