Deep Reinforcement Learning Based Green Resource Allocation Mechanism in Mobile Edge Network for Ubiquitous Power IoT
作者单位：State Key Laboratory of Networking and Switching Technology Beijing University of Posts and Telecommunications Global Energy Interconnection Research Institute Co. Ltd Communication branch State Grid Jiangsu Electric Power Co. Ltd
会议日期：2020年
学科分类：080904[工学-电磁场与微波技术] 13[艺术学] 080802[工学-电力系统及其自动化] 0808[工学-电气工程] 0809[工学-电子科学与技术（可授工学、理学学位）] 08[工学] 1305[艺术学-设计学（可授艺术学、工学学位）] 0810[工学-信息与通信工程] 081104[工学-模式识别与智能系统] 080402[工学-测试计量技术及仪器] 0804[工学-仪器科学与技术] 081001[工学-通信与信息系统] 081101[工学-控制理论与控制工程] 0811[工学-控制科学与工程]
关 键 词：Deep Reinforcement Learning Green Resource Allocation Mobile Edge Network
摘 要：Ubiquitous Power IoT is an important part of smart grid or energy internet. Edge network is a key component of smart grid and future energy Internet. IoT contains various terminals, diverse access networks and equipment. The allocation of energy-efficient resources under QoS guarantee is an effective means to ensure the reliable operation of the network. Due to the heterogeneity of the services carried and in order to reduce business operation risks and impact on the power grid, it is necessary to provide highly reliable resource allocation method to the ubiquitous power IoT business for the energy Internet. Edge network can reduce end-to-end latency and reduce backhaul link traffic. However, mobile edge networks moving the function of storage and computing down to the edge nodes of network, which makes resource management more complex. Moreover, the computing resources in the edge network is limited and the effect to the ecological environment and economic cost should also be considered. So how to allocate resources such as bandwidth and power more effectively while meeting the needs of users becomes an important issue. Even if the Deep Reinforcement Learning（DRL） algorithm has been used to a lot of the work which are correlated to edge networks, there lacks the applications for green resource allocation. In this paper, a mechanism based on Deep Reinforcement Learning（DRL） for the resource allocation problem is proposed oriented to edge networks. The mechanism aims at how to allocate resources more efficiently and energy-saving while satisfying the requirements of each user. The simulation results show that the energy efficiency value could be obtained when the algorithm converges and stabilizes. The Energy Efficiency（EE） of the proposed mechanism and the productiveness in satisfying user requirements and implementing green resource allocation are validated.