基于分布式强化学习的蜂窝网络覆盖容量自优化

overage and capacity self-optimization directly affects the user\'s experience of network performance, which is one of the important self-organizing use cases in mobile cellular networks. Although centralized optimization algorithms can achieve good performance, global network information is needed, which can incur huge overhead. Facing above problem, a distributed reinforcement learning based BS radio parameter adjustment algorithm is developed in this paper. Specifically, in order to avoid the occurrence of multiple problem cells and the large adjustment of the problem area which affects the performance of neighboring cells, the optimization objective is defined as the weighting of the coverage and capacity of the entire area, which is then modeled as a Markov Decision Process (MDP). By the decomposition of MDP, each base station can be regarded as an agent that autonomously controls its own radio frequency parameters based on only local information. By interacting with the environment, both the coverage and capacity are optimized. Further, to accelerate the convergence of distributed reinforcement learning, a central controller is introduced to perform cooperative adjustment of multi-BS\'s parameters. Via simulation, the effectiveness of the proposals are verified compared to centralized schemes.