Deep Reinforcement Learning for Energy Efficiency Optimization in Wireless Networks

We consider a joint decision-making problem of the transmission power and channel in wireless networks with unknown state switching patterns for improving energy efficiency (EE). To overcome the unknown dynamics of network, we model the problem as a sequential decision making process, and apply deep reinforcement learning (DRL), which aggregates reinforcement learning (RL) being appropriate for complex sequential decision problem and deep learning (DL) characterized by its advantages of learning feature and approximating the nonlinear function, for figuring out the difficulties of dealing the problem with massive state space and modeling the temporal characteristics between sequential decisions. Simulation results show the proposed algorithm performs efficiently in term of energy efficiency.