Deep Reinforcement Learning for Dynamic Access Control with Battery Prediction for Mobile-Edge Computing in Green IoT Networks

Abstract

Mobile Edge Computing (MEC) technology has emerged as a promising paradigm to reduce the energy consumption for the resource-constrained and energy-limited Internet of Things (IoT) networks. In this paper, benefiting from energy harvesting technique (EH), we study the dynamic MEC-access control problem for maximizing the long-term average uplink transmission rate whilst minimizing the transmission energy consumption for green IoT networks, in which the IoT device is powered by a rechargeable battery that can harvest energy from the surrounding environments. In particular, this problem is formulated as a Markov decision process with system dynamics unknown. On accounting of the dynamics of the wireless channel state, the energy arrival, and the mobility of the IoT device, a Long Short-Term Memory (LSTM) enhanced Deep Q-Network (DQN) based (LSDQN) access control algorithm is proposed for the IoT network. In the proposed algorithm, the LSTM model is used to predict the battery status for assisting the IoT device to determine the optimal access control decision by DQN with the target of maximizing the average uplink rate whilst minimizing the energy consumption. Finally, extensive simulation results verify the performance of the proposed algorithm.