Research On Cloud Edge Collaboration Architecture For Network Autonomy

The future network is a large-scale ubiquitous network that covers all scenarios.By introducing the concept of minimal and flexible architecture,it realizes demand-driven intelligent control mechanism,thus achieving the goal of network self-growth,self-evolution and self-organization.Network autonomy is an important feature of future networks.Network autonomy can realize deep integration of computing and communication resources,automatically perceive and analyze task requirements,and make decisions and execute tasks.In order to realize network autonomy,cloud-edge collaboration is an important technical means.It can comprehensively utilize the computing and resource management capabilities of the cloud side,as well as the low-latency and flexible deployment capabilities of the edge side,to achieve efficient collaboration of computing and communication,and improve the performance of network autonomy.This paper focuses on two problems in network autonomous scenarios:wireless coverage self-optimization and network service self-optimization.It proposes solutions based on cloud-edge collaborative architecture technology,using machine learning and reinforcement learning algorithms to model,evaluate and optimize autonomous objectives.The wireless coverage self-optimization scheme uses cloud-side data analysis and edgeside prediction feedback capabilities to realize multi-cell joint beamforming decision optimization for 3D MIMO,which improves communication signal coverage quality in sudden crowd gathering mobile scenarios.The network service self-optimization scheme uses cloud-side resource management and edge-side flexible deployment capabilities to perform offloading optimization for communication and computing services in wireless cloud gaming.It uses multi-agent reinforcement learning algorithm to reduce latency and energy consumption in network service processing.In wireless coverage self-optimization,this paper proposes a coverage optimization scheme based on cloud-edge collaboration for 3D MIMO multi-cell joint beamforming problem in sudden crowd gathering scenarios.It improves user signal coverage quality.In sudden crowd gathering scenarios,traditional machine learning algorithms cannot accurately predict user trajectories due to lack of historical information for guiding antenna beam codebook selection.The wireless coverage self-optimization scheme proposed in this paper consists of centralized cloud brain and distributed edge brain.The edge brain uses ConvLSTM neural network to predict user location.The cloud brain is responsible for collecting global data and calculating hotspot crowd feature information.Through cloudedge collaborative feedback,it accurately predicts global user location distribution at future time instants.Simulation results show that compared with schemes using only cloud brain or only edge brain,the proposed scheme improves prediction accuracy by 24.66%and 68.26%respectively,user signal coverage quality by 0.2dB and 0.4dB respectively.In network service self-optimization,this paper proposes a computing offloading optimization scheme based on cloud-edge collaboration for communication and computing service offloading optimization problem in wireless cloud gaming scenario.It reduces communication latency and energy consumption in cloud gaming service processing computation process.The network service self-optimization scheme consists of two layers:the first layer is edge server layer,responsible for local resource scheduling and computation.The second layer is cloud server layer,responsible for business computation offloading and transmission.This paper uses multi-agent reinforcement learning algorithm BiCNet to optimize computing offloading strategy.It simulates communication and computation process of cloud gaming service,and compares different reinforcement learning algorithms.Experimental results show that the cloud-edge collaborative scheme based on BiCNet compared with schemes using only edge brain and only cloud brain,reduce latency by 77%and 59%respectively,average energy consumption by 58%and 37%respectively.