Research On Multi-network Convergence Based On Intelligent Algorithm And Edge Computing

With the continuous increase in the number and types of cognitive network access devices,scenarios where multiple cognitive networks form a communication environment are increasingly common,and access nodes in the network also face more challenges.In the multi-cognitive network scenario,if there is mutual interference between different links,how can nodes maximize the use of resources when the transmission power is limited,and at the same time ensure that the generated interference is controllable for users with higher priority(primary users).Within the scope,detailed discussion and research are required.When the cognitive network scene is integrated with the edge computing model,it is also a big problem how nodes can process tasks quickly and efficiently.This paper studies and improves the above problems in detailThis paper takes nodes with different priorities in the cognitive network as a breakthrough,discusses the link between master and slave users and the link between slave users separately,and uses the idea of multi-agent to design slave user links.In the case of the interference between links and the limited available power of nodes,an improved multi-agent reinforcement learning method is proposed.By further selecting better samples to complete reinforcement learning,the model can complete the convergence as soon as possible.This method is more effective for resource allocation and management of links in multi-cognitive networks.The actual results show that this method improves the resource perception and preemption ability of the secondary user link,and also ensures the communication quality of the primary user link.As for how the access node performs task execution and offloading in the multi-cognitive network,the reinforcement learning method is also used,and the node in the access system is regarded as the receiving node in edge computing.The computing power of the main user in each cognitive network and bandwidth resources are sufficient,and can be used as a central node in edge computing for task offloading and execution.The distributed design structure is used to simulate multiple network states in real-world scenarios,and the improved deep reinforcement learning algorithm is used to train all nodes in the system,that is,to refine the results of the output layer of the neural network.In the case of knowing the main frequency of each node and the computing cost of the task,the improved algorithm can improve the overall resource utilization of the system and ensure the execution efficiency of the task.