基于深度双Q网络的多用户蜂窝网络功率分配算法研究

针对现有蜂窝网络功率分配算法存在泛化能力弱、效率低等问题进行了研究,提出基于深度双Q网络(deep double Q network,DDQN)的功率分配算法。采用双神经网络结构,解决强化学习过程中易出现的维度灾难及值函数过估计问题;对状态信息进行设计并输入神经网络,输出智能体的动作行为,并设计奖赏函数反馈给神经网络,使智能体可以有效地自主学习,多次迭代得到最优的功率分配策略。仿真结果表明,所提的模型可获得的平均速率为1.89,平均运行时间为0.0013 s,在不同用户密度及小区数量下均可达到最高的平均速率,验证了算法的有效性,为蜂窝网络资源分配问题提供了新的思路。

Multi-user cellular network power allocation based on deep double-Q network

In view of the weak generalization ability and low efficiency of the existing cellular network power allocation algorithm,this paper proposed the power allocation algorithm based on the deep double Q network.It adopted the dual neural network structure to solve the problem of dimensional disasters and value function overestimation that were easy to occur in the reinforcement learning process.It designed the state information and inputted it to the neural network,outputted the action behavior of the agent,and designed the reward function to feed back to the neural network to make the agent can effectively learn independently and obtain the optimal power allocation strategy through multiple iterations.The simulation results show that the average speed that the proposed model can obtain is 1.89,and the average running time is 0.0013 s.The highest average speed can be reached under different user densities and cell numbers,which verifies the effectiveness of the algorithm.The problem of network resource allocation provides new ideas.