Research On SDN Intelligent Multi-path Planning Technology

Recently,artificial intelligence,a new subject in the field of computer,has developed rapidly,especially in the field of reinforcement learning and deep reinforcement learning.Software Defined Networking(SDN),a new network architecture,has been widely concerned in academic and business field because of its characteristic of centralized control.How to realize load balancing of network traffic through reasonable traffic scheduling algorithm in SDN network,therefore improve the service quality of network business is a hot topic of current research.With the development of artificial intelligence,it provides a new research direction for SDN network to realize intelligent path planning.Based on the research of current SDN traffic scheduling technology,this thesis proposes an intelligent multi-path routing planning method based on reinforcement learning.That combines with the advantages of reinforcement learning on strategy optimization and the characteristics of SDN network resource centralized control.According to the current state information and traffic characteristics of the network,this method performs multi-path routing for network traffic by using reinforcement learning model.Experimental results show that this method can find multiple forwarding paths for different flows and improve the utilization of network bandwidth.Compared with current traffic scheduling algorithms,the proposed traffic scheduling algorithm can reduce network delay jitter and packet loss effectively.In real network environment,many network states exist.There are some drawbacks of reinforcement learning,e.g.: it is difficult to extract artificial features and the state-action space is usually narrow.After combining deep learning with reinforcement learning and making full use of neural network in feature extraction,this thesis proposes an intelligent multi-path routing planning method based on deep reinforcement learning.It collects training samples from a large number of historical experience data and uses the neural network of deep learning to fit the Qvalue table in the traditional reinforcement learning algorithm.Finally,experiments show the feasibility of fitting reinforcement learning Q-value table by this method.Compared with other routing planning methods based on reinforcement learning,it is found that the multi-path routing planning method based on reinforcement learning proposed in this thesis has certain advantages.