Explore Deep Reinforcement Learning to Energy-Efficient Data Synchronism in 5G Self-Powered Sensor Networks

Abstract

Recently, the explosive data traffic growth and large-scale Internet of Things (IoT) equipment connection have brought 5G mobile communication technology many challenges. The communication delay of network services such as augmented reality, smart grid, disaster warning and emergency communication have been relieved by 5G. However, these services are still confronted with the serious challenge in energy conservation. Most of traditional optimization methods usually need complex operations and iterations to get the optimal results, which are not suitable for communication systems with high real-time performance. Based on this argument, this paper has focused on the green communication in Mobile Edge Computing (MEC)-based self-powered sensor network and proposed a novel approach called MIDS (Mobile Intelligent Data Synchronization based on deep reinforcement learning). It exploits Deep Deterministic Policy Gradient (DDPG), continuous interacting with the environment and making tryouts to evaluate the feedback from the environment to optimize future decision-making on the path selection scheme for data transmission as well as achieve the energy efficiency and energy consumption balance of mobile devices with self-powered sensors. We provide experiments to show the capability of our approach in reducing the additional energy consumption of mobile devices and the base station as well as balancing the energy consumption of mobile devices with sensor communication. The superiority of our approach for energy conservation in data synchronization is convincingly demonstrated by comparing with state-of-the-art methods in MEC-based self-powered sensor network.