Research On Energy-Efficient Load Balancing Technology Based On Deep Reinforcement Learning In 5G Cloud Radio Access Network

The rapid popularity of intelligent terminals makes the demand for network traffic explosive growth.In order to improve the efficiency of network transmission and management,5G cloud radio access network(c-ran)arises at the historic moment,which improves the network capacity by deploying a large number of small base stations(BSs).However,the intensive deployment of base stations brings great challenges to network energy efficiency and load balancing.This article focuses on 5G C-RAN,from the BSs load and network resources matching and capacity optimization oriented energy-efficient load balancing mechanism,according to the dynamic network traffic demand,adaptively adjust the BS working mode and the use of network resources,so as to improve the energy efficiency of the network.Aiming at the problem of high energy consumption caused by the mismatch between the load and the number of active BSs in 5G C-RAN,this article proposes a load and resource matching strategy based on deep reinforcement learning.Firstly,the optimal matching method based on load balancing is adopted to associate the BS with the user equipment;Secondly,the load of the next moment of network is predicted based on Markov Decision Process;Finally,the intelligent BS sleeping strategy is made based on the load prediction.Under the constraints of network processing delay and user minimum SINR requirement,the proposed strategy achieves the matching of load and network resources,and greatly improves the network energy efficiency.Aiming at the load imbalance problem caused by the dynamic and random changes of network demand in time and space in 5G C-RAN,this article proposes a double-layer load balancing strategy based on deep reinforcement learning.The strategy constructs a double-layer load control system mechanism,which is composed of top-level and bottom-level parts.The top-level centralized load controller divides the BSs in the region into clusters according to the historical load level,and adapts to the dynamic global load fluctuation from a macro perspective.For the bottom-level self-organizing network load controller of each cluster,the mobility load balancing(MLB)method based on deep reinforcement learning is adopted to achieve more precise load adjustment under the condition of ensuring the network outage probability and the minimum signal to interference noise ratio of users.