User Association with Multi-Agent Reinforcement Learning for Energy-Efficient UDN

Ultra-dense network (UDN) has been received a great deal of attention recently. However, an excessive number of data links and frequent handovers incur inefficient energy consumption in the process of user association. Specifically, energy-efficiently associating users for a long time remained as a challenge. In this paper, we propose a multi-agent deep reinforcement learning (MADRL)-based user association technique to minimize the energy consumption UDN. By applying actor-critic, the proposed scheme learns to optimally associate users to minimize the energy consumption based on local information. From the simulations, we demonstrate the proposed user association scheme reduces energy consumption of UDN substantially.