Research On Routing Algorithm Based On Reinforcement Learning In SDN

As next network architecture,SDN(Software Defined Networking)has the characteristics of separation between forwarding and control.Centralized control also brings great convenience to network management.The same as traditional network,packet and stream transmission in SDN architecture also requires network routing.How to allocate links so that traffic is not overloaded and each link is fully used in SDN is a hot topic of current research.However,there is so far not a unified and efficient method for routing in SDN network.In this thesis,the reinforcement learning algorithm is applied to the routing planning of SDN networks,reinforcement learning is an area of artificial intelligence that studies algorithms that dynamically optimize their performance based on experience in an environment.The routing planning algorithms based on QLearning and Q(A)are proposed respectively.Based on the link QoS requirements,for example,data lost,available bandwidth and so on,the links are classified hierarchically.Both methods can find a forwarding path that meets the QoS requirements.For the defects of manual design feature in the routing algorithm based on QLearning and Q(A),a routing algorithm based on Deep QLearning is proposed in this thesis,which uses the neural network model to replace the Q value table in QLearning algorithm.Finally,on the basis of previous algorithm implementation,we propose a routing system based on reinforcement learning and give the overall framework design and module design.The system has been tested,and the effectiveness of the system in network routing is verified.