Base station power control strategy in ultra-dense networks via deep reinforcement learning

Within the context of 5G, Ultra-Dense Networks (UDNs) are regarded as an important network deployment strategy, employing a large number of low-power small cells to achieve extended coverage and enhanced service quality. However, the deployment of numerous small cells results in a linear increase in energy consumption in wireless communication systems. To enhance system efficiency and establish green wireless communication systems, this paper investigates base station sleeping and power allocation strategy based on deep reinforcement learning in UDNs. Firstly, a system energy consumption model for UDNs is established, which is divided into two sub-problems based on the final optimization problem, namely base station sleep and power allocation. Two Deep Q-networks (DQNs) are employed simultaneously for optimization. In addition to considering traditional system energy efficiency (EE), this study also optimizes system spectral efficiency (SE) and user transmission rate as optimization objectives simultaneously. Simulation results show that the proposed method improves EE and SE by about 70% and 81%.